National Academies Press: OpenBook

Urban Stormwater Management in the United States (2009)

Chapter: 4 Monitoring and Modeling

« Previous: 3 Hydrologic, Geomorphic, and Biological Effects of Urbanization on Watersheds
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 257
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 258
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 259
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 260
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 261
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 262
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 263
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 264
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 265
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 266
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 267
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 268
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 269
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 270
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 271
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 272
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 273
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 274
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 275
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 276
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 277
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 278
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 279
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 280
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 281
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 282
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 283
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 284
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 285
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 286
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 287
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 288
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 289
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 290
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 291
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 292
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 293
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 294
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 295
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 296
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 297
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 298
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 299
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 300
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 301
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 302
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 303
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 304
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 305
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 306
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 307
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 308
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 309
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 310
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 311
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 312
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 313
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 314
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 315
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 316
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 317
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 318
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 319
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 320
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 321
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 322
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 323
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 324
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 325
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 326
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 327
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 328
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 329
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 330
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 331
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 332
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 333
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 334
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 335
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 336
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 337
Suggested Citation:"4 Monitoring and Modeling." National Research Council. 2009. Urban Stormwater Management in the United States. Washington, DC: The National Academies Press. doi: 10.17226/12465.
×
Page 338

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

4 Monitoring and Modeling As part of its statement of task, the committee was asked to consider several aspects of stormwater monitoring, including how useful the activity is, what should be monitored and when and where, and how benchmarks should be es- tablished. As noted in Chapter 2, the stormwater monitoring requirements under the U.S. Environmental Protection Agency (EPA) stormwater program are vari- able and generally sparse, which has led to considerable skepticism about their usefulness. This chapter first considers the value of the data collected over the years by municipalities and makes suggestions for improvement. It then does the same for industrial stormwater monitoring, which has lagged behind the mu- nicipal separate storm sewer system (MS4) program both in requirements and implementation. It should be noted upfront that this chapter does not discuss the fine details of MS4 and industrial monitoring that pertain to regulatory compliance— questions such as should the average end of pipe concentrations meet water quality standards, how many exceedances should be allowed per year, or should effluent concentrations be compared to acute or chronic criteria. Individual benchmarks and effluent limits for specific chemicals emanating from specific industries are not provided. The current state of MS4 and industrial stormwater monitoring and the paucity of high quality data are such that it is premature and in many cases impossible to make such determinations. Rather, the chapter sug- gests both how to monitor an individual industry and how to determine bench- marks and effluent limits for industrial categories. It suggests how monitoring requirements should be tailored to accommodate the risk level of an individual industrial discharger. Finally, it makes numerous technical suggestions for im- proving the monitoring of MS4s, building on the data already submitted and analyzed as part of the National Stormwater Quality Database. Policy recom- mendations about the monitoring of both industries and MS4s are found in Chapter 6. This chapter’s emphasis on monitoring of stormwater should not be inter- preted as a disinterest in other types of monitoring, such as biomonitoring of receiving waters, precipitation measurements, or determination of land cover. Indeed, these latter activities are extremely important (they are introduced in the preceding chapter) and they underpin the new permitting program proposed in Chapter 6 (especially biological monitoring). Stormwater management would benefit most substantially from a well-balanced monitoring program that en- compasses chemical, biological, and physical parameters from outfalls to receiv- ing waters. Currently, however, decisions about stormwater management are usually made with incomplete information; for example, there are continued recommendations by many that street cleaning will solve a municipality’s prob- lems, even when the municipality does not have any information on the sources of the material being removed. 257

258 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES A second charge to the committee was to define the elements of a “proto- col” to link pollutants in stormwater discharges to ambient water quality criteria. As described in Chapter 3, many processes connect sources of pollution to an effect observed in a downstream receiving water. More and more, these proc- esses can be represented in watershed models, which are the key to linking stormwater sources to effects observed in receiving waters. The latter half of the chapter explores the current capability of models to make such links, including simple models, statistical and conceptual models, and more involved mechanis- tic models. At the present time, associating a single discharger with degraded in-stream conditions is generally not possible because of the state of both model- ing and monitoring of stormwater. MONITORING OF MS4s EPA’s regulations for stormwater monitoring of MS4s is very limited, in that only the application requirements are stated [see 40 CFR § 122.26(d)]. The regulations require the MS4 program to identify five to ten stormwater discharge outfalls and to collect representative stormwater data for conventional and prior- ity toxic pollutants from three representative storm events using both grab and composite sampling methods. Each sampled storm event must have a rainfall of at least 0.1 inch, must be preceded by at least 72 hours of a dry period, and the rain event must be within 50 percent of the average or median of the per storm volume and duration for the region. While the measurement of flow is not spe- cifically required, an MS4 must make estimates of the event mean concentra- tions (EMCs) for pollutants discharged from all outfalls to surface waters, and in order to determine EMCs, flow needs to be measured or calculated. Other than these requirements, the exact type of MS4 monitoring that is to be conducted during the permit term is left to the discretion of the permitting authority. EPA has not issued any guidance on what would be considered an adequate MS4 monitoring program for permitting authorities to evaluate com- pliance. Some guidance for MS4 monitoring based on desired management questions has been developed locally (for example, see the SCCWRP Technical Report No. 419, SMC 2004, Model Monitoring Program for MS4s in Southern California). In the absence of national guidance from EPA, the MS4 monitoring pro- grams for Phase I MS4s vary widely in structure and objectives, and Phase II MS4 programs largely do not perform any monitoring at all. The types of moni- toring typically contained in Phase I MS4 permits include the (1) wet weather outfall screening and monitoring to characterize stormwater flows, (2) dry weather outfall screening and monitoring under illicit discharge detection and elimination programs, (3) biological monitoring to determine storm water im- pacts, (4) ambient water quality monitoring to characterize water quality condi- tions, and (5) stormwater control measure (SCM) effectiveness monitoring.

MONITORING AND MODELING 259 The Nationwide Stormwater Quality Database Stormwater monitoring data collected by a portion of Phase I MS4s has been evaluated for years by the University of Alabama and the Center for Wa- tershed Protection and compiled in a database called the Nationwide Stormwater Quality Database (NSQD). These data were collected in order to describe the characteristics of stormwater on a national level, to provide guidance for future sampling needs, and to enhance local stormwater management activities in areas with limited data. The MS4 monitoring data collected over the past ten years from more than 200 municipalities throughout the country have great potential in characterizing the quality of stormwater runoff and comparing it against his- torical benchmarks. Version 3 of the NSQD is available online at: http://unix.eng.ua.edu/~rpitt/Research/ms4/mainms4.shtml. It contains data from more than 8,500 events and 100 municipalities throughout the country. About 5,800 events are associated with homogeneous land uses, while the re- mainder are for mixed land uses. The general approach to data collection was to contact EPA regional offices to obtain state contacts for the MS4 data, then the individual municipalities with Phase I permits were targeted for data collection. Selected outfall data from the International BMP Database were also included in NSQD version 3, eliminating any source area and any treated stormwater samples. Some of the older National Urban Runoff Program (NURP) (EPA, 1983) data were also included in the NSQD, along with some data from specialized U.S. Geological Survey (USGS) stormwater monitoring activities in order to better represent nationwide condi- tions and additional land uses. Because there were multiple sources of informa- tion, quality assurance and quality control reviews were very important to verify the correctness of data added to the database, and to ensure that no duplicate entries were added. The NSQD includes sampling location information such as city, state, land use, drainage area, and EPA Rain Zone, as well as date, season, and rain depth. The constituents commonly measured for in stormwater include total suspended solids (TSS), 5-day biological oxygen demand (BOD5), chemical oxygen de- mand (COD), total phosphorus (TP), total Kjeldahl nitrogen (TKN), nitrite plus nitrate (NO2+NO3), total copper (Cu), total lead (Pb), and total zinc (Zn). Less information is available for many other constituents (including filterable heavy metals and bacteria). Figure 4-1 is a map showing the EPA Rain Zones in the United States, along with the locations of the communities contributing to the NSQD, version 3. Table 4-1 shows the number of samples for each land use and for each Rain Zone. This table does not show the number of mixed land-use site samples. Rain Zones 8 and 9 have very few samples, and institutional and open- space areas are poorly represented. However, residential, commercial, indus- trial, and freeway data are plentiful, except for the few Rain Zones noted above. Land use has an important impact on the quality of stormwater. For exam- ple, the concentrations of heavy metals are higher for industrial land-use areas

260 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES TABLE 4-1 Number of Samples per Land Use and EPA Rain Zone Single Land Use 1 2 3 4 5 6 7 8 9 Total Commercial 234 484 131 66 42 37 64 0 22 1080 Freeways 0 241 14 0 262 189 28 0 0 734 Industrial 100 327 90 51 83 74 146 0 22 893 Institutional 9 46 0 0 0 0 0 0 0 55 Open Space 68 37 0 18 0 2 0 0 0 125 Residential 294 1470 290 122 105 32 532 7 81 2933 Total 705 2605 525 257 492 334 770 7 125 5820 Note: there are no mixed-use sites in this table. SOURCE: National Stormwater Quality Database. due to manufacturing processes and other activities that generate these materials. Fecal coliform concentrations are relatively high for residential and mixed resi- dential land uses, and nitrate concentrations are higher for the freeway land use. Open-space land-use areas show consistently low concentrations for the con- stituents examined. Seasons could also be a factor in the variation of nutrient concentrations in stormwater due to seasonal uses of fertilizers and leaf drop occurring during the fall season. Most studies also report lower bacteria concen- trations in the winter than in the summer. Lead concentrations in stormwater have also significantly decreased since the elimination of lead in gasoline (see Figure 2-6). Most of the statistical tests used are multivariate statistical evalua- tions that compare different constituent concentrations with land use and geo- graphical location. More detailed discussions of the earlier NSQD results are found in various references, including Maestre et al. (2004, 2005) and Pitt et al. (2003, 2004). How to use the NSQD to Calculate Representative EMC Values EMC values were initially used during the NURP to describe typical con- centrations of pollutants in stormwater for different monitoring locations and land uses. An EMC is intended to represent the average concentration for a sin- gle monitored event, usually based on flow-weighted composite sampling. It can also be calculated from discrete samples taken during an event if flow data are also available. Many individual subsamples should be taken throughout most of the event to calculate the EMC for that event. Being an overall average value, an EMC does not represent possible extremes that may occur during an event. The NSQD includes individual EMC values from about 8,500 separate events. Stormwater managers typically want a representative single value for a land use for their area. As such, they typically evaluate a series of individual

MONITORING AND MODELING 261 FIGURE 4-1 Sampling Locations for Data Contained in the National Stormwater Quality Database, version 3. storm EMC values for conditions similar to those representing their site of con- cern. With the NSQD in a spreadsheet form, it is relatively simple to extract suitable events representing the desired conditions. However, the individual EMC values will likely have a large variability. Maestre and Pitt (2006) re- viewed the NSQD data to better explain the variability according to different site and sampling conditions (land use, geographical location, season, rain depth, amount of impervious area, sampling methods, antecedent dry period, etc.). The most common significant factor was land use, with some geographical and fewer seasonal effects observed. As with the original NURP data, EMCs in the NSQD are usually expressed using medians and coefficients of variation to reflect un- certainty, assuming lognormal distributions of the EMC values. Figure 4-2 shows several lognormal probability plots for a few constituents from the NSQD. Probability plots shown as straight lines indicate that the concentrations can be represented by lognormal distributions (see Box 4-1).

262 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES FIGURE 4-2 Lognormal probability plots of stormwater quality data for selected constitu- ents (pooled data from NSQD version 1.1). Fitting a known distribution is important as it helps indicate the proper sta- tistical tests that may be conducted. Using the median EMC value in load calcu- lations, without considering the data variability, will result in smaller mass loads compared to actual monitored conditions. This is due to the medians underrep- resenting the larger concentrations that are expected to occur. The use of aver- age EMC values will represent the larger values better, although they will still not represent the variability likely to exist. If all of the variability cannot be further explained adequately (such as being affected by rain depth), which would be highly unlikely, then a set of random calculations (such as that ob- tained using Monte Carlo procedures) reflecting the described probability distri- bution of the constituents would be the best method to use when calculating loads. Municipal Monitoring Issues As described in Chapter 2, typical MS4 monitoring requirements involve sampling during several events per year at the most common land uses in the area. Obviously, a few samples will not result in very useful data due to

MONITORING AND MODELING 263 BOX 4-1 Probability Distributions of Stormwater Data The coefficient of variation (COV) values for many constituents in the NSQD range from unusually low values of about 0.1 (for pH) to highs between 1 and 2. One objective of a data analysis procedure is to categorize the data into separate stratifications, each having small variations in the observed concentrations. The only stratification usually applied is for land use. However, further analyses indicated many differences by geographical area and some differences by season. When separated into appropriate stratifications, the COV values are reduced, ranging between about 0.5 to 1.0. With a reasonable confidence of 95 percent ( = 0.05) and power of 80 percent ( = 0.20), and a suitable allowable error goal of 25 percent, the number of samples needed to characterize these conditions would there- fore range from about 25 to 50 (Burton and Pitt, 2002). In a continuing monitoring program (such as the Phase I stormwater National Pollutant Discharge Elimination System [NPDES] permit monitoring effort) characterization data will improve over time as more samples are obtained, even with only a few samples collected each year from each site. Stormwater managers have generally accepted the assumption of lognormality of stormwater constituent concentrations between the 5th and 95th percentiles. Based on this assumption, it is common to use the log-transformed EMC values to evaluate differences between land-use categories and other characteristics. Statistical inference methods, such as estimation and tests of hypothesis, and analysis of variance, require statistical informa- tion about the distribution of the EMC values to evaluate these differences. The use of the log-transformed data usually includes the location and scale parameter, but a lower-bound parameter is usually neglected. Maestre et al. (2005) conducted statistical tests using NSQD data to evaluate the log- normality assumptions of selected common constituents. It was found in almost all cases that the log-transformed data followed a straight line between the 5th and 95th percentile, as illustrated in Figure 4-3 for total dissolved solids (TDS) in residential areas. For many statistical tests focusing on the central tendency (such as for determining the concentrations that are to be used for mass balance calculations), this may be a suit- able fit. As an example, the model WinSLAMM (Pitt, 1986; Pitt and Voorhees, 1995) uses a Monte Carlo component to describe the likely variability of stormwater source flow pollut- ant concentrations using either lognormal or normal probability distributions for each con- stituent. However, if the most extreme values are of importance, such as when dealing with the influence of many non-detectable values on the predicted concentrations, or de- termining the frequency of observations exceeding a numerical standard, a better descrip- tion of the extreme values may be important. The NSQD contains many factors for each sampled event that likely affect the ob- served concentrations. These include such factors as seasons, geographical zones, and rain intensities. These factors may affect the shape of the probability distribution. The only way to evaluate the required number of samples in each category is by using the power of the test, where power is the probability that the test statistic will lead to a rejection of the null hypothesis (Gibbons and Chakraborti, 2003). In the NSQD, most of the data were from residential land uses. The Kolmogorov- Smirnov test was used to indicate if the cumulative empirical probability distribution of the residential stormwater constituents can be adequately represented with a lognormal distri- bution. The number of collected samples was sufficient to detect if the empirical distribu- continues next page

264 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES BOX 4-1 Continued Probability Plot for Different Land Uses Total Dissolved Solids (mg/L) 0.9995 0.999 0.998 0.995 0.99 0.98 0.95 Cumulative Probability 0.9 0.8 0.7 0.6 0.5 0.4 0.3 LAND USE 0.2 Residential 0.1 0.05 0.02 0.01 0.005 1 10 100 1000 10000 Concentration (mg/L) FIGURE 4-3 Probability plot of total dissolved solids in residential land uses (NSQD ver- sion 1.1 data). tion was located inside an interval of width 0.1 above and below the estimated cumulative probability distribution. If the interval was reduced to 0.05, the power varies between 40 and 65 percent. Another factor that must be considered is the importance of relatively small errors in the selected distribution and the problems of false-negative determinations. It may not be practical to collect as many data observations as needed when the distribu- tions are close. Therefore, it is important to understand what types of further statistical and analysis problems may be caused by having fewer samples than optimal. For example, Figure 4-4 (total phosphorus in residential areas) shows that most of the data fall along the straight line (indicating a lognormal fit), with fewer than 10 observations (out of 933) in the tails being outside of the obvious path of the line, or a false-negative rate of about 0.01 (1 percent).

MONITORING AND MODELING 265 FIGURE 4-4 Normality test for total phosphorus in residential land uses using the NSQD. Further analyses to compare the constituent concentration distributions to other com- mon probability distributions (normal, lognormal, gamma, and exponential) were also con- ducted for all land uses by Maestre et al. (2004). Most of the stormwater constituents can be assumed to follow a lognormal distribution with little error. The use of a third parameter in the estimated lognormal distribution may be needed, depending on the number of sam- ples. When the number of samples is large per category (approximately more than 400 samples) the maximum likelihood and the two-parameter lognormal distribution better fit the empirical distribution. For large sample sizes, the L-moments method usually unacceptably truncates the distribution in the lower tail. However, when the sample size is more moder- ate per category (approximately between 100 and 400 samples), the three-parameter log- normal method, estimated by L-moments, better fits the empirical distribution. When the sample size is small (less than 100 samples, as is common for most stormwater programs), the use of the third parameter does not improve the fit with the empirical distribution and the common two-parameter lognormal distribution produces a better fit than the other two methods. The use of the lognormal distribution also has an advantage over the other dis- tribution types because it can be easily transformed to a normal distribution and the data can then be correctly examined using a wide variety of statistical tests.

266 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES the variability of stormwater characteristics. However, during the period of a five-year permit with three samples per year, about 15 events would be sampled for each land use. While still insufficient for many analyses, this number of data points likely allows the confidence limits to be reasonably calculated for the average conditions. When many sites of the same land use are monitored for a region, substantial data may be collected during a permit cycle. This was the premise of the NSQD where MS4 data were collected for many locations throughout the country. These data were evaluated and various findings made. The following comments are partially based on these analyses, along with addi- tional data sources. Sampling Technique and Compositing There are a variety of methods for collecting and compositing stormwater samples that can result in different values for the EMC. The first distinction is the mode of sample collection, either as grab samples or automatic sampling. Obviously, grab sampling is limited by the speed and accuracy of the individuals doing the sampling, and it is personnel intensive. It is for this reason that about 80 percent of the NSQD samples are collected using automatic samplers. Man- ual sampling has been observed to result in slightly lower TSS concentrations compared to automatic sampling procedures. This may occur, for example, if the manual sampling team arrives after the start of runoff and therefore misses an elevated first flush (if it exists for the site), resulting in reduced EMCs. A second important concept is how and whether the samples are combined following collection. With time-based discrete sampling, samplers (people or machines) are programmed to take an aliquot after a set period of time (usually in the range of every 15 minutes) and each aliquot is put into a separate bottle (usually 1 liter). Each bottle is processed separately, so this method can have high laboratory costs. This is the only method, however, that will characterize the changes in pollutant concentrations during the event. Time-based composite sampling refers to samplers being programmed to take an aliquot after a set pe- riod of time (as short as every 3 minutes), but then the aliquots are combined into one container prior to analysis (compositing). All parts of the event receive equal weight with this method, but the large number of aliquots can produce a reasonably accurate composite concentration. Finally, flow-weighted composite sampling refers to samplers being programmed to collect an aliquot (usually 1 liter) for a set volume of discharge. Thus, more samples are collected during the peak of the hydrograph than toward the trailing edge of the hydrograph. All of the aliquots are composited into one container, so the concentration for the event is weighted by flow. Most communities calculate their EMC values using flow-weighted com- posite sample analyses for more accurate mass discharge estimates compared to time-based compositing. This is especially important for areas with a first flush of very short duration, because time-composited samples may overly emphasize

MONITORING AND MODELING 267 these higher flows. An automatic sampler with flow-weighted samples, in con- junction with a bed-load sampler, is likely the most accurate sampling method, but only if the sampler can obtain a representative sample at the location (such as sampling at a cascading location, or using an automated depth-integrated sampler) (Clark et al., 2008). Time- and flow-weighted composite options have been evaluated in resi- dential, commercial, and industrial land uses in EPA Rain Zone 2 and in indus- trial land uses in EPA Rain Zone 3 for the NSQD data. No significant differ- ences were observed for BOD5 concentrations using either of the compositing schemes for any of the four categories. TSS and total lead median concentra- tions in EPA Rain Zone 2 were two to five times higher in concentration when time-based compositing was used instead of flow-based compositing. Nutrients in EPA Rain Zone 2 collected in residential, commercial, and industrial areas showed no significant differences using either compositing method. The only exceptions were for ammonia in residential and commercial land-use areas and total phosphorus in residential areas where time-based composite samples had higher concentrations. Metals were higher when time-based compositing was used in residential and commercial land-use areas. No differences were ob- served in industrial land-use areas, except for lead. Again, in most cases, mass discharges are of the most importance in order to show compliance with TMDL requirements. Flow-weighted sampling is the most accurate method to obtain these values (assuming sufficient numbers of subsamples are obtained). How- ever, if receiving water effects are associated with short-duration high concen- trations, then discrete samples need to be collected and analyzed, with no com- positing of the samples during the event. Of course, this is vastly more costly and fewer events are usually monitored if discrete sampling is conducted. Numbers of Data Observations Needed The biggest issue associated with most monitoring programs is the number of data points needed. In many cases, insufficient data are collected to address the objectives of the monitoring program with a reasonable amount of confi- dence and power. Burton and Pitt (2002) present much guidance in determining the amount of data that should be collected. A basic equation that can be used to estimate the number of samples to characterize a set of conditions is as follows: n = [COV(Z1- + Z1- )/(error)]2 where: n = number of samples needed. = false-positive rate (1– is the degree of confidence; a value of of 0.05 is usually considered statistically significant, corresponding to a 1– degree

268 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES of confidence of 0.95, or 95%). = false-negative rate (1– is the power; if used, a value of of 0.2 is com- mon, but it is frequently and improperly ignored, corresponding to a of 0.5). Z1– = Z score (associated with area under a normal curve) corresponding to 1– ; if is 0.05 (95% degree of confidence), then the corresponding Z1– score is 1.645 (from standard statistical tables). Z1– = Z score corresponding to 1– value; if is 0.2 (power of 80%), then the corresponding Z1– score is 0.85 (from standard statistical tables); how- ever, if power is ignored and is 0.5, then the corresponding Z1– score is 0. error = allowable error, as a fraction of the true value of the mean. COV = coefficient of variation (sometimes noted as CV), the standard de- viation divided by the mean (dataset assumed to be normally distributed). Figures 4-5 and 4-6 can be used to estimate the sampling effort, based on the expected variability of the constituent being monitored, the allowable error in the calculated mean value, and the associated confidence and power. Figure 4-5 can be used for a single sampling point that is being monitored for basic characterization information, while Figure 4-6 is used for paired sampling when two locations are being compared. Confidence and power are needed to control the likelihood of false negatives and false positives. The sample needs increase dramatically as the difference between datasets becomes small when comparing two conditions with a paired analysis, as shown in Figure 4-6 (above and below an outfall, influent vs. effluent, etc.). Typically, being able to detect a difference of at least about 25 percent (requiring about 50 sample pairs with typical sample variabilities) is a reasonable objective for most stormwater projects. This is es- pecially important when monitoring programs attempt to distinguish test and control conditions associated with SCMs. It is easy to confirm significant dif- ferences between influent and effluent conditions at wet detention ponds, as they have relatively high removal rates. Less effective controls are much more diffi- cult to verify, as the sampling program requirements become very expensive. First-Flush Effects First flush refers to an assumed elevated load of pollutants discharged in the beginning of a runoff event. The first-flush effect has been observed more often in small catchments than in large catchments (Thompson et al., 1995, cited by WEF and ASCE, 1998). Indeed, in large catchments (>162 ha, 400 acres), the

MONITORING AND MODELING 269 FIGURE 4-5 Number of samples to characterize median (power of 80% and confidence of 95%). SOURCE: Reprinted, with permission from, Burton and Pitt (2002). Copyright 2002 by CRC Press.

270 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES FIGURE 4-6 Number of paired samples needed to distinguish between two sets of obser- vations (power 80% and confidence of 95%). SOURCE: Reprinted, with permission from, Burton and Pitt (2002). Copyright 2002 by CRC Press.

MONITORING AND MODELING 271 highest concentrations are usually observed at the times of flow peak (Brown et al., 1995; Soeur et al., 1995). Adams and Papa (2000) and Deletic (1998) both concluded that the presence of a first flush depends on numerous site and rain- fall characteristics. Figure 4-7 is a plot of monitoring data from the Villanova first-flush study (Batroney, 2008) showing the flows, rainfall, TSS concentration, TDS concen- tration, and TDS and TSS event mean concentrations for the inflow to an infil- tration trench. Because of the first-flush effect, a grab sample early in the storm would have over-predicted the TSS event mean concentration of the site, and a later sample would have under-predicted this same value, although for TDS the results would have been similar. FIGURE 4-7 Villanova first-flush study showing pollutant concentration as a function of inflow rainfall volume. This study collected runoff leaving the top floor of a parking garage. Samples were taken of the runoff in one-quarter-inch increments, up to an inch of rain, and then every inch thereafter. The plot of TSS concentration versus rainfall increment shows a strong first flush for this storm, while the TDS concentration does not. SOURCE: Re- printed, with permission, Batroney (2008). Copyright 2008 by T. Thomas Batroney.

272 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES Figure 4-8 shows data for a short-duration, high-intensity rain in Tusca- loosa, Alabama, that had rain intensities as great a 6 inches per hour for a 10- minute period. The drainage area was a 0.4-ha paved parking lot with some landscaping along the edges. The turbidity plot shows a strong first flush for this event, and the particle size distributions indicate larger particles at the be- ginning of the event, then becoming smaller as the event progresses, and then larger near the end. Most of the other pollutants analyzed had similar first-flush patterns like the turbidity, with the notable exception of bacteria. Both E. coli and enterococci concentrations started off moderately low, but then increased substantially near the end of the rain. Several rains have been monitored at this site so far, and most show a similar pattern with decreasing turbidity and in- creasing bacteria as the rain continues. Sample collection conducted for some of the NPDES MS4 Phase I permits required both a grab and a composite sample for each event. A grab sample was to be taken during the first 30 minutes of discharge to capture the first flush, and a flow-weighted composite sample was to be taken for the entire time of dis- charge (every 15 to 20 minutes for at least three hours or until the event ended). Maestre et al. (2004) examined about 400 paired sets of 30-minute and 3-hour samples from the NSQD, as shown in Table 4-2. Generally, a statistically sig- nificant first flush is associated with a median concentration ratio of about 1.4 or greater (the exceptions are where the number of samples in a specific category is much smaller). The largest ratios observed were about 2.5, indicating that for these conditions the first 30-minute flush sample concentrations are about 2.5 times greater than the composite sample concentrations. More of the larger ra- tios are found for the commercial and institutional land-use categories, where larger paved areas are likely to be found. The smallest ratios are associated with the residential, industrial, and open-space land uses—locations where there may be larger areas of unpaved surfaces. The data in Table 4-2 were from North Carolina (76.2 percent), Alabama (3.1 percent), Kentucky (13.9 percent), and Kansas (6.7 percent) because most other states’ stormwater permits did not require this sampling strategy. The NSQD investigation of first-flush conditions for these data locations indicated that a first-flush effect was not present for all the land-use categories and cer- tainly not for all constituents. Commercial and residential areas were more likely to show this phenomenon, especially if the peak rainfall occurred near the beginning of the event. It is expected that this effect will more likely occur in a watershed with a high level of imperviousness, but even so, the data indicated first flushes for less than 50 percent of the samples for the most impervious ar- eas. This reduced frequency of observed first flushes in areas most likely to have first flushes is probably associated with the varying rain conditions during the different events, including composite samples that did not represent the complete runoff duration.

MONITORING AND MODELING FIGURE 4-8 Pollutant variations during rain period (0.4-ha drainage area, mostly paved parking with small 273 fringe turf area, Tuscaloosa, Alabama). SOURCE: Robert Pitt, University of Alabama.

274 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES TABLE 4-2 Significant First Flush Ratios (First Flush to Composite Median Concentration) Commercial Industrial Institutional Parameter n sc R ratio n sc R ratio n sc R ratio Turbidity, NTU 11 11 = 1.32 X X COD, mg/L 91 91 2.29 84 84 1.43 18 18 2.73 TSS, mg/L 90 90 1.85 83 83 = 0.97 18 18 2.12 Fecal coliform, 12 12 = 0.87 X X col/100mL TKN, mg/L 93 86 1.71 77 76 1.35 X Phosphorus total, 89 77 1.44 84 71 = 1.42 17 17 = 1.24 mg/L Copper, total, µg/L 92 82 1.62 84 76 1.24 18 7 = 0.94 Lead, total, µg/L 89 83 1.65 84 71 1.41 18 13 2.28 Zinc, total, µg/L 90 90 1.93 83 83 1.54 18 18 2.48 Open Space Residential All Combined Parameter n sc R ratio n sc R ratio n sc R ratio Turbidity, NTU X 12 12 = 1.24 26 26 = 1.26 COD, mg/L 28 28 = 0.67 140 140 1.63 363 363 1.71 TSS, mg/L 32 32 = 0.95 144 144 1.84 372 372 1.60 Fecal coliform, X 10 9 = 0.98 22 21 = 1.21 col/100mL TKN, mg/L 32 14 = 1.28 131 123 1.65 335 301 1.60 Phosphorus, 32 20 = 1.05 140 128 1.46 363 313 1.45 total, mg/L Copper, total, 30 22 = 0.78 144 108 1.33 368 295 1.33 µg/L Lead, total, 31 16 = 0.90 140 93 1.48 364 278 1.50 µg/L Zinc, total, 21 21 = 1.25 136 136 1.58 350 350 1.59 µg/L Note: n, number of total possible events; sc, number of selected events with detected val- ues; R, result; X, not enough data; =, not enough evidence to conclude that median values are different; , median values are different. “Ratio” is the ratio of the first flush to the full- period sample concentrations. SOURCE: NSQD, as reported by Maestre et al. (2004).

MONITORING AND MODELING 275 Groups of constituents showed different behaviors for different land uses. All the heavy metals evaluated showed higher concentrations at the beginning of the event in the commercial land-use category. Similarly, all the nutrients showed higher initial concentrations in residential land-use areas, except for total nitrogen and orthophosphorus. This phenomenon was not found in the bacterial analyses. None of the land uses showed a higher population of bacteria at the beginning of the event. The general conclusion from these data is that, in areas having low and gen- erally even-intensity rains, first-flush observations are more common, especially in small and mostly paved areas. As an area increases in size, multiple routing pathways tend to blend the water, and runoff from the more distant locations reaches the outfall later in the event. SCMs located at outfalls in areas having low levels of impervious cover should be selected and sized to treat the com- plete event, if possible. Preferential treatment of first flushes may only be justi- fied for small impervious areas, but even then, care needs to be taken to prevent undersizing and missing substantial fractions of the event. Seasonal first flushes refer to larger portions of the annual runoff and pol- lutant discharges occurring during a short rain season. Seasonal first flushes may be observed in more arid locations where seasonal rainfalls are predomi- nant. As an example, central and southern California can have dry conditions for extended periods, with the initial rains of the season occurring in the late fall. These rains can be quite large and, since they occur after prolonged dry periods, may carry substantial portions of the annual stormwater pollutant load. This is especially pronounced if later winter rains are more mild in intensity and fre- quent. For these areas, certain types of seasonally applied SCMs may be effec- tive. As an example, extensive street, channel, and inlet cleaning in the late summer and early fall could be used to remove large quantities of debris and leaves from the streets before the first heavy rains occur. Other seasonal main- tenance operations benefiting stormwater quality should also be scheduled be- fore these initial rains. Rain Depth Effects An issue related to first flushes pertains to the effects of rain depth on stormwater quality. The NSQD contains much rainfall data along with runoff data for most areas of the country. Figure 4-9 contains scatter plots showing concentrations plotted against rain depth for some NSQD data. Although many might assume a correlation between concentrations and rain depth, in fact there are no obvious trends of concentration associated with rain depth. Rainfall en- ergy determines erosion and wash-off of particulates, but sufficient runoff vol- ume is needed to carry the particulate pollutants to the outfalls. Different travel times from different locations in the drainage areas results in these materials arriving at different times, plus periods of high rainfall intensity (that increase pollutant wash-off and movement) occur randomly throughout the storm. The

276 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES FIGURE 4-9 Examples of scatter plots by precipitation depth. SOURCE: NSQD. resulting outfall stormwater concentration patterns for a large area having vari- ous surfaces is therefore complex and rain depth is just one of the factors in- volved. Reported Monitoring Problems A number of monitoring problems were described in the local Phase I community MS4 annual monitoring reports that were summarized as part of assembling the NSQD. About 58 percent of the communities described moni- toring problems. Problems were mostly associated with obtaining reliable data for the targeted events. These problems increased costs because equipment fail- ures had to be corrected and sampling excursions had to be rescheduled. One of the basic sampling requirements was to collect three samples every year for each

MONITORING AND MODELING 277 of the land-use stations. These samples were to be collected at least one month apart during storm events having at least 0.1-inch rains, and with at least 72 hours from the previous 0.1-inch storm event. It was also required (when feasi- ble) that the variance in the duration of the event and the total rainfall not exceed the median rainfall for the area. About 47 percent of the communities reported problems meeting these requirements. In many areas of the country, it was dif- ficult to have three storm events per year with these characteristics. Further- more, the complete range of site conditions needs to be represented in the data- collection effort; focusing only on a narrow range of conditions limits the repre- sentativeness of the data. The second most frequent problem, reported by 26 percent of the communi- ties, concerned backwater tidal influences during sampling, or that the outfall became submerged during the event. In other cases, it was observed that there was flow under the pipe (flowing outside of the pipe, in the backfill material, likely groundwater), or sometimes there was no flow at all. These circum- stances all caused contamination of the collected samples, which had to be dis- carded, and prevented accurate flow monitoring. Greater care is obviously needed when locating sampling locations to eliminate these problems. About 12 percent of the communities described errors related to malfunc- tions of the sampling equipment. When reported, the equipment failures were due to incompatibility between the software and the equipment, clogging of the rain gauges, and obstruction in the sampling or bubbler lines. Memory losses in the equipment recording data were also periodically reported. Other reported problems were associated with lighting, false starts of the automatic sampler before the runoff started, and operator error due to misinterpretation of the equipment configuration manual. The reported problems suggest that the following changes should be made. First, the rain gauges need to be placed close to the monitored watersheds. Large watersheds cannot be represented with a single rain gauge at the monitor- ing station. In all cases, a standard rain gauge needs to supplement a tipping bucket rain gauge, and at least three rain gauges should be used in the research watersheds. Second, flow-monitoring instrumentation also needs to be used at all water quality monitoring stations. The lack of flow data greatly hinders the value of the chemical data. Third, monitoring needs to cover the complete storm duration. Automatic samplers need to be properly programmed and maintained to handle very short to very long events. It is unlikely that manual samplers were able to initiate sampling near the beginning of the events, unless they were deployed in anticipation of an event later in the day. A more cost-effective and reliable option would be to have semi-permanent monitoring stations at the vari- ous locations with sampling equipment installed in anticipation of a monitored event. Most monitoring agencies operated three to five land-use stations at one time. This number of samplers, and flow equipment, could have been deployed in anticipation of an acceptable event and would not need to be continuously installed in the field at all sampling locations.

278 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES Non-Detected Analyses Left-censored data involve observations that are reported as below the lim- its of detection, whereas right-censored data involve above-range observations. Unfortunately, many important stormwater measurements (such as for filtered heavy metals) have large fractions of undetected values. These incomplete data greatly hinder many statistical tests. To estimate the problems associated with censored values, it is important to identify the probability distributions of the data in the dataset and the level of censoring. As discussed previously, most of the constituents in the NSQD follow a lognormal distribution. When the fre- quencies of the censored observations were lower than 5 percent, the means, standard deviations, and COVs were almost identical to the values obtained when the censored observations were replaced by half of the detection limit. As the percentage of nondetected values increases, replacing the censored observa- tion by half of the detection limit instead of estimating them using Cohen’s maximum likelihood method produced lower means and larger standard devia- tions. Replacing the censored observations by half of the detection limit is not recommended for levels of censoring larger than 15 percent. Because the Cohen method uses the detected observations to estimate the nondetected values, it is not very accurate, and therefore not recommended, when the percentage of cen- sored observations is larger than 40 percent (Burton and Pitt, 2002). In this case, summaries should only be presented for the detected observations, with clear notations stating the level of nondetected observations. The best method to eliminate problems associated with left-censored data is to use an appropriate analytical method. By keeping the nondetectable level below 5 percent, there are many fewer statistical analysis problems and the value of the datasets can be fully realized. Table 4-3 summarizes the recom- mended minimum detection limits for various stormwater constituents to obtain manageable nondetection frequencies (< 5 percent), based on the NSQD data observations. Some of the open-space stormwater measurements (lead, and oil and grease, for example) would likely have greater than 5 percent nondetections, even with the detection limits shown. The detection limits for filtered heavy metals should also be substantially less than shown on this table. Seasonal Effects Another factor that some believe may affect stormwater quality is the sea- son when the sample was obtained. If the few samples collected for a single site were all collected in the same season, the results may not be representative of the whole year. The NPDES sampling protocols were designed to minimize this effect by requiring the three samples per year to be separated by at least one month. The few samples still could be collected within a single season, but not within the same week. Seasonal variations for residential fecal coliform data are shown in Figure 4-10 for NSQD data for all residential areas. These data were

MONITORING AND MODELING 279 TABLE 4-3 Suggested Analytical Detection Limits for Stormwater Monitoring Programs to Obtain Less Than 5 Percent Nondetections Residential, Commercial, Industrial, Parameter Open Space Freeway Conductivity 20 S/cm 20 S/cm Hardness 10 mg/L 10 mg/L Oil and grease 0.5 mg/L 0.5 mg/L TDS 10 mg/L 10 mg/L TSS 5 mg/L 1 mg/L BOD5 2 mg/L 1 mg/L COD 10 mg/L 5 mg/L Ammonia 0.05 mg/L 0.01 mg/L NO2 + NO3 0.1 mg/L 0.05 mg/L TKN 0.2 mg/L 0.2 mg/L Dissolved P 0.02 mg/L 0.01 mg/L Total P 0.05 mg/L 0.02 mg/L Total Cu 2 g/L 2 g/L Total Pb 3 g/L (residential g/L) 1 g/L Total Ni 2 g/L 1 g/L Total Zn 20 g/L (residential 10 g/L) 5 g/L SOURCE: Maestre and Pitt (2005). FIGURE 4-10 Fecal coliform concentrations in stormwater by season. SOURCE: NSQD.

280 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES the only significant differences in concentration by season for any constituent measured. The bacteria levels are lowest during the winter season and highest during the summer and fall (a similar conclusion was obtained during the NURP data evaluations). Recommendations for MS4 Monitoring Activities The NSQD is an important tool for the analysis of stormwater discharges at outfalls. About a fourth of the total existing information from the NPDES Phase I program is included in the database. Most of the statistical analyses in this research were performed for residential, commercial, and industrial land uses in EPA Rain Zone 2 (the area of emphasis according to the terms of the EPA- funded research). Many more data are available from other stormwater permit holders that are not included in this database. Acquiring these additional data for inclusion in the NSQD is a recommended and cost-effective activity and should be accomplished as additional data are also being obtained from ongoing monitoring projects. The use of automatic samplers, coupled with bed-load samplers, is preferred over manual sampling procedures. In addition, flow monitoring and on-site rainfall monitoring need to be included as part of all stormwater characterization monitoring. The additional information associated with flow and rainfall data will greatly enhance the usefulness of the much more expensive water quality monitoring. Flow monitoring must also be correctly conducted, with adequate verification and correct base-flow subtraction methods applied. A related issue frequently mentioned by the monitoring agencies is the lack of on-site precipita- tion information for many of the sites. Using regional rainfall data from loca- tions distant from the monitoring location is likely to be a major source of error when rainfall factors are being investigated. Many of the stormwater permits only required monitoring during the first three hours of the rain event. This may have influenced the EMCs if the rain event continued much beyond this time. Flow-weighted composite monitoring should continue for the complete rain duration. Monitoring only three events per year from each monitoring location requires many years before statistically adequate numbers of observations are obtained. In addition, it is much more difficult to ensure that such a small fraction of the total number of annual events is representative. Also, there is minimal value in obtaining continued data from an area after sufficient information is obtained. It is recommended that a more concentrated monitoring program be conducted for a two- or three-year period, with a total of about 30 events monitored for each site, covering a wide range of rain conditions. Periodic checks can be made in future years, such as repeating concentrated monitoring every 10 years or so (and for only 15 events during the follow-up surveys). Finally, better watershed area descriptions, especially accurate drainage- area delineations, are needed for all monitored sites. While the data contained in

MONITORING AND MODELING 281 the NSQD are extremely useful, future monitoring information obtained as part of the stormwater permit program would be greatly enhanced with these addi- tional considerations. MONITORING OF INDUSTRIES INCLUDING CONSTRUCTION The various industrial stormwater monitoring requirements of the EPA Stormwater Program have come under considerable scrutiny since the program’s inception. Input to the committee at its first meeting conveyed the strong sense that monitoring as it is being done is nearly useless, is burdensome, and pro- duces data that are not being utilized. The requirements consist of the follow- ing. All industrial sectors covered under the Multi-Sector General Permit (MSGP) must conduct visual monitoring four times a year. This visual monitor- ing is performed by collecting a grab sample within the first hour of stormwater discharge and observing its characteristics qualitatively (except for construction activities—see below). A subset of MSGP industries are required to perform analytical monitoring for benchmark pollutant parameters (see Table 2-5) four times in year 2 of permit coverage and again in year 4 if benchmarks are ex- ceeded in year 2. A benchmark sample is collected as a grab sample within the first hour of stormwater discharge after a rainfall event of 0.1 inch or greater and with an interceding dry period of at least 72 hours. An even smaller subset of MSGP industries that are subject to numerical effluent guidelines under 40 C.F.R. must, in addition, collect grab samples of their stormwater discharge after every discharge event and analyze it for specific pollutant parameters as speci- fied in the effluent guidelines (see Table 2-6). There is no monitoring require- ment for stormwater discharges from construction activity in the Construction General Permit. There is only an elective requirement that the construction site be visually inspected within 24 hours after the end of a storm event that is 0.5 inch or greater, if inspections are not performed weekly. EPA selected the benchmark analytical parameters for industry subsectors to monitor using data submitted by industrial groups in 1993 as part of their group applications. The industrial groups were required to sample a minimum of 10 percent of facilities within an industry group for pH, TSS, BOD5, oil and grease, COD, TKN, nitrate plus nitrite nitrogen, and total phosphorous. Each sampling facility within a group collected a minimum of one grab sample within the first 30 minutes of discharge and one flow-weighted composite sample. Other nonconventional pollutants such as fecal coliform bacteria, iron, and co- balt were analyzed only if the industry group expected it to be present. Simi- larly, toxic pollutants such as lead, copper, and zinc were not sampled but rather self-identified only if expected to be present in the stormwater discharge. As a result of the self-directed nature of these exercises, the data submitted with the group applications were often incomplete, inconsistent, and not representative of the potential risk posed by the stormwater discharge to human health and aquatic

282 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES life. EPA has not conducted or funded independent investigations and has relied solely on the data submitted by industry groups to determine which pollutant parameters are appropriate for the analytical monitoring of an industry subsec- tor. Thus, there are glaring deficiencies; for example, the only benchmark pa- rameter for asphalt paving and roofing materials is TSS, even though current science shows that the most harmful pollutants in stormwater discharges from the asphalt manufacturing industry are polycyclic aromatic hydrocarbons (com- pare Table 2-5 with Mahler et al., 2005). Aside from the suitability of benchmark parameters is the fact the too few samples are collected to sufficiently characterize the variability of pollutant con- centrations associated with industrial facilities within a sector. This is discussed in detail in Box 4-2, which describes one of the few efforts to collect and ana- lyze data from the benchmark monitoring of industries done in Southern Cali- fornia. EPA has not requested a nationwide effort to compile these data, as was done for the MS4 program, although this could potentially lead to average efflu- ent concentrations by industrial sector that could be used for a variety of pur- poses, including more considerate regulations. Finally, the compliance monitor- ing that is presently being conducted under the MSGP is of limited usefulness because it is being done to comply with effluent guidelines that have not been updated to reflect the best available technology relevant to pollutants of most concern. All of these factors have led to an industrial stormwater monitoring program that is not very useful for the purposes of reducing stormwater pollu- tion from industries or informing operators on which harmful pollutants to ex- pect from their sites. Industrial-Area Monitoring Issues Monitoring at industrial sites has some unique issues that must be over- come. The most important aspect for any monitoring program is understanding and specifying the objectives of the monitoring program and developing and following a detained experimental design to allow these objectives to be met. The following discussion is organized around the reasons why monitoring at industrial sites may be conducted. Regional Monitoring of Many Facilities An important monitoring objective would be regional monitoring to cali- brate and verify stormwater quality models, to randomly verify compliance at facilities not normally requiring monitoring, and to establish benchmarks for compliance. As shown in Box 4-2, haphazard monitoring throughout an area would require a very large effort, and would still likely result in large errors in the expected data. It is recommended that a regional stormwater authority coor- dinate regional monitoring as part of the MS4 monitoring requirements, possibly

MONITORING AND MODELING 283 BOX 4-2 The Plight of Industrial Stormwater Data Unlike the data collected by municipalities and stored in the NSQD, the benchmark monitoring data collected by permitted industries are not compiled or analyzed on a na- tional basis. However, there has been at least one attempt to compile these data on a more local basis. California required that industrial facilities submit their benchmark moni- toring data over a nine-year period, and it was subsequently analyzed by Michael Sten- strom and colleagues at UCLA (Stenstrom and Lee, 2005; Lee et al., 2007). The collected data were for such parameters as pH, turbidity, specific conductance, oil and grease (or total organic carbon), and several metals. There are more than 6,000 industries covered under the California general permit, each of which was to have collected two grab samples per year for a limited number of parameters. Whether these data were collected each year and for each industry was highly variable. The analysis of the data from Los Angeles and Ventura counties revealed that storm- water monitoring data are not similar to the types of data that the environmental engineer- ing field is used to collecting, in particular wastewater data. Indeed, as shown in Figure 4- 11, stormwater data are many orders of magnitude more variable than drinking water and wastewater data. The coefficients of variation for municipal and industrial stormwater were almost two orders of magnitude higher than for drinking water and wastewater, with the industrial stormwater data being particularly variable. This variability comes from various sources, including intrinsic variability given the episodic nature of storm events, analytical methods that are more variable when applied to stormwater, and sampling technique prob- lems and error. FIGURE 4-11 A comparison of data from four sources: wastewater influent, drinking water plant effluent, municipal stormwater, and industrial stormwater. SOURCE: Reprinted, with permission, from Stenstrom (2007). Copyright 2007 by Michael K. Stenstrom. continues next page

284 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES BOX 4-2 Continued This enormous variability means that it is extremely difficult to make meaningful state- ments. For example, it was impossible, using different analyses, to correlate certain chemical pollutants with certain industries. Furthermore, although the data revealed that there are exceedances of benchmark values for certain parameters (Al, Cu, Fe, Pb, and Zn in particular), the data are not of sufficient quantity or quality to identify problem polluters. Finally, there were also large numbers of outliers (that is, samples whose concentrations were well above the 75th percentile range). Because of these large coefficients of variation, greater numbers of samples are needed to be able to say there is a significant difference between samples. As shown in Figure 4-12 using COD and a 50 percent difference in means as an example, one would need six data points to tell the difference between two wastewater influents, 80 data points if one had municipal stormwater data, and around 1,000 data points for industrial stormwa- ter. These numbers obviously eclipse what is required under all states’ MSGPs. For drinking water treatment, monitoring is done to ensure the quality of the product, while for wastewater, there is a permit that requires the plant to meet a specific quality of water. Unlike these other areas of water resources, there are few incentives that might compel an industry to increase its frequency of stormwater monitoring. As a result, indus- tries are less invested in the process and rarely have the expertise needed to carry out self- monitoring. Permitted industries are not required to sample flow. However, Stenstrom and col- leagues used Los Angeles rainfall data (see Figure 4-13) as a surrogate for flow and dem- onstrated that there is a seasonal first-flush phenomenon occurring in early fall. That is, samples taken after a prolonged dry spell will have higher pollutant concentrations. There are always high concentrations of contaminants during the first rainfall because contami- nants have had time to accumulate since the previous rainfall. This is important because EPA asks the industrial permittees to collect data from the first rainfall, such that they may end up overestimating the mass emissions for the year. Furthermore, it shows that nu- meric limits for grab samples would be risky because the measured data are highly affected by the timing of the storm. The controversy about numeric limits for industrial stormwater dischargers has existed for more than ten years in California. A recent expert panel concluded that in some cases, numeric limits are appropriate (for construction, but not for municipalities). Stenstrom’s recommendations are that industrial monitoring should be either ended or upgraded (for competent industries). If upgraded, it should include more types of monitored parameters, a sampling method with a lower coefficient of variation, real-time monitoring as opposed to grab samples, more quality assurance/quality control, and web-based reporting. A fee- based program with a subset of randomly selected industries may be better than requiring every industry to sample. Stenstrom and Lee (2005) suggest who might do this monitoring if the industry does not have the necessary trained personnel. There is concern that the California water boards are too understaffed to administer such programs and respond to high emitters.

MONITORING AND MODELING 285 ip mp ww FIGURE 4-12 Number of cases needed to detect a certain percentage difference in the means, using COD as an example. SOURCE: Reprinted, with permission, from Stenstrom (2007). Copyright 2007 by Michael K. Stenstrom. FIGURE 4-13 Annual precipitation in Los Angeles (left) and seasonal first flushes of vari- ous contaminants (right). SOURCE: Reprinted, with permission, from Stenstrom (2007). Copyright 2007 by Michael K. Stenstrom. SOURCES: Stenstrom and Lee (2005), Lee et al. (2007), Stenstrom (2007).

286 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES even at the state level covering several Phase I municipalities. A coordinated effort would be most cost-effective with the results compiled for a specific ob- jective. The general steps in this effort would include the following. (1) Compiling available regional stormwater quality data and comparing the available data to the needs (such as calibration of a regional model; verifying compliance of facilities not requiring monitoring; and establishing regional benchmarks). This may include expanding the NSQD for the region to include all of the collected data, plus examination of data collected as part of other spe- cialized monitoring activities. These objectives will result in different data needs, so it is critical that the uses of the data are identified before sampling plans are established. (2) Identifying monitoring opportunities as part of other on-going activities that can be expanded to also meet data gaps for these specific objectives. It is important to understand the time frame for the monitoring and ensure that it will meet the needs. As an example, current NPDES stormwater monitoring only requires a few events to be sampled per year at a facility. It may take many years before sufficient data are obtained unless the monitoring effort is acceler- ated. (3) Preparing an experimental design that identifies the magnitude of the needed data, considering the allowable errors in the results, and carrying out the sampling program. Different types of data may have varying data quality objec- tives, depending on their use. It may be possible to truncate some of the moni- toring when a sufficient understanding is obtained. A regionally calibrated and verified model can be used to review develop- ment plans and proposed SCMs for new facilities. When suitably integrated with receiving-water modeling tools, a stormwater model can also be used to develop discharge objectives and numeric discharge limits that are expected to meet regulatory requirements. Eventually, it may be possible to couple water- shed stormwater models with regional receiving water assessments and benefi- cial use studies. Haphazard monitoring of a few events each year will be very difficult to correlate with regional receiving water objectives, while a calibrated and verified watershed model, along with receiving water assessments, will re- sult in a much more useful tool and understanding of the local problems. Regional monitoring can also be targeted to categories of industries that were previously determined to be of low priority. This monitoring activity would randomly target a specific number of these facilities for monitoring to verify the assumption that they are of low priority and are still carrying out the minimum management practices. This activity would also quantify the dis- charges from these facilities and the performance of the minimum controls. If the discharges are excessive when compared to the initial assumptions, or the management practices being used are not adequate, then corrective actions would be instigated. A single category of specific industries could be selected for any one year, and a team from the regional stormwater management author-

MONITORING AND MODELING 287 ity could randomly select and monitor a subset of these facilities. An efficient experimental design would need to be developed based on expected conditions, but it is expected that from 10 to 15 such facilities would be monitored for at least a year in a large metropolitan area that has a Phase I stormwater permit, or even state-wide. Regional monitoring is also necessary to more accurately establish bench- marks for numeric permits. Geographical location, along with land use, is nor- mally an important factor affecting stormwater quality. Receiving water im- pacts and desired beneficial uses also vary greatly for different locations. It is therefore obvious that compliance benchmarks also be established that consider these regional differences. This could be a single statewide effort if the state agency has the permit authority and if the state has minimal receiving water and stormwater variations. However, in most cases, significant variations occur throughout the state and separate monitoring activities would be needed for each region. In the simplest case, probability distributions of stormwater discharge quality can be developed for different discharge categories and the benchmarks would be associated with a specific probability value. In some cases, an overall distribution may be appropriate, and only the sites having concentrations greater than the benchmark value would need to have additional treatment. In all cases, a basic level of stormwater management should be expected for all sites, but the benchmark values would identify sites where additional controls are necessary. The random monitoring of sites not requiring extensive monitoring could be used to identify and adjust the basic levels of control needed for all categories of stormwater dischargers. Identification of Critical Source Areas Associated with Specific Industrial Operations The objective of this monitoring activity would be to identify and character- ize critical source areas for specific industries of concern. If critical source areas can be identified, targeted control or treatment can be much more effective than relying only on outfall monitoring. Many of the treatment strategies for indus- trial sites involve pollution prevention, ranging from covering material or prod- uct storage areas to coating galvanized metal. Other treatment strategies involve the use of highly effective treatment devices targeting a small area, such as fil- ters used to treat zinc in roof runoff or lamella plate separators for pretreatment of storage yard runoff before wet pond treatment. Knowledge of the characteris- tics of the runoff from the different areas at a facility is needed in order to select and design the appropriate treatment methods. Box 4-3 is a case study of one such group monitoring effort—for a segment of the telecommunications industry targeting a specific maintenance practice. Instead of having each telecommunication company throughout the country conduct a detailed monitoring program for individual stormwater permits asso- ciated with maintenance efforts, many of the companies joined together under an

288 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES BOX 4-3 Monitoring to Support a General Stormwater Group Permit Application for the Telecommunications Industry This monitoring program was conducted to support a group permit application for the telecommunications industry, specifically to cover maintenance operations associated with pumping water out of communications manholes that is then discharged into the storm drainage system. Under federal and state environmental statues, the generator (owner or operator) is responsible for determining if the discharged water needs treatment. The work performed under this project covered characterization, prevention, and treatment methods of water found in manholes. The objective of this project was to develop a test method to quickly evaluate water in manholes and then to recommend on-site treatment and preventative methods. To meet the telecommunication industry needs, the evaluating tests of water found in manholes need to be simple, quick, inexpensive, field applicable, and accurate indicators of contami- nated conditions. The on-site treatment methods must be cost-effective and quickly reduce the concentrations of the contaminant of concern to acceptable levels before the water from manholes is discharged, to result in a safe environment for workers. A sampling effort was conducted by Pitt et al. (1998) to characterize the quality of the water and sediment found in manholes. More than 700 water samples and 300 sediment samples were analyzed over a three-year period, representing major land-use, age, sea- son, and geographical factors from throughout the United States. The samples were ana- lyzed for a wide range of common and toxic constituents. The statistical procedures identi- fied specific relationships between these main factor categories and other manhole charac- teristics. Part of the project was to evaluate many field analytical methods. Finally, re- search was also conducted to examine possible water treatment methods for water being pumped from telecommunication manholes. Summary of Sampling Effort and Strategy The objective of the monitoring program was to characterize telecommunication man- hole water and sediment. Important variables affecting the quality of these materials were also determined. A stratified random sampling design was followed, with the data organ- ized in a full 24 factorial design, with repeated sampling of the same manholes for each season. The goal for the minimum number of samples per strata was ten. This sampling effort enabled the determination of errors associated with the results, which was expected to be less than 25 percent. In addition, this level of effort enabled comparison tests to be made outside of the factorial design. Table 4-4 lists the constituents that were evaluated for each of the sample types. The immense amount of data collected during this project and the adherence to the original experimental design enabled a comprehensive statistical evaluation of the data. Several steps in data analysis were performed, including: exploratory data analyses (mainly probability plots and grouped box plots), simple correlation analyses (mainly Pearson correlation matrices and associated scatter plots), complex correlation analyses (mainly cluster and principal component analyses, plus Kurskal-Wallis comparison tests), and

MONITORING AND MODELING 289 4 model building (based on complete 2 factorial analyses of the most important factors). The toxicity screening tests (using the Azur Microtox method) conducted on both un- filtered and filtered water samples from telecommunication manholes indicated a wide range of toxicity, with no obvious trends for season, land use, or age. About 60 percent of the samples were not considered toxic (less than an I25 light reduction of 20 percent, the light reduction associated with phosphorescent bacteria after a 25-minute exposure to undi- luted samples), about 20 percent were considered moderately toxic, while about 10 percent were considered toxic (light reductions of greater than 40 percent), and 10 percent were considered highly toxic (light reductions of greater than 60 percent). Surprisingly, samples from residential areas generally had greater toxicities than samples from commercial and industrial areas. Samples from newer areas were also more toxic than those from older areas. Further statistical tests of the data indicated that the high toxicity levels were likely associated with periodic high concentrations of salt (in areas using de-icing salt), heavy metals (especially filterable zinc, with high values found in most areas), and pesticides (associated with newer residential areas). TABLE 4-4 Constituents Examined in Water and Sediment from Telecommunication Manholes Unfiltered Filtered Constituent Sediment Water Water Solids, volatile solids, COD, Cu, Pb, and Zn X X X Turbidity, color, and toxicity (Microtox X X screening method) pH, conductivity, hardness, phosphate, nitrate, X ammonia, boron, fluoride, potassium, and detergents Odor, color, and texture X E. coli, enterococci, particle size, and Selected chromium Metal scan (ICP) Selected PAHs, phenols (GC/MSD), and pesticides X Selected Selected SOURCE: Modified from Pitt et al. (1998). Concentrations of copper, lead, and zinc were evaluated in almost all of the water samples, and some filtered samples were also analyzed for chromium. From 470 to 548 samples (75 to 100 percent of all unfiltered samples analyzed) had detectable concentra- tions of these metals. Filterable lead concentrations in the water were as high as 160 g/L, while total lead concentrations were as high as 810 g/L. Zinc values in filtered and unfil- tered samples were as high as about 3,500 g/L. Some of the copper concentrations were also high in both filtered and unfiltered samples (as high as 1,400 g/L). Chromium con- centrations as high as 45 g/L were also detected. continues next page

290 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES BOX 4-3 Continued About 300 sediment samples were analyzed and reviewed for heavy metals. An ICP/MS was used to obtain a broad range of metals with good detection limits. The follow- ing list shows the median observed concentrations for some of the constituents found in the sediments (expressed as milligrams of the constituent per kilogram of dry sediment): Aluminum 14,000 mg/kg COD 85,000 mg/kg Chromium <10 mg/kg Copper 100 mg/kg Lead 200 mg/kg Strontium 35 mg/kg Zinc 1,330 mg/kg Geographical area had the largest effect on the data observations, while land use, season, and age influenced many fewer parameters. The most obvious relationship was found for high dissolved solids and conductivity associated with winter samples from snowmelt areas. The high winter concentrations slowly decreased with time, with the low- est concentrations noted in the fall. Another important observation was the common asso- ciation between zinc and toxicity. Residential-area samples generally had larger zinc con- centrations than the samples from commercial and industrial areas. Samples from the newest areas also had higher zinc concentrations compared to samples from older areas. No overall patterns were observed for zinc concentrations in sediment samples obtained from manholes. Other constituents (especially nutrients and pesticides) were also found to have higher concentrations in water collected from manholes in newer residential areas. Very few organic toxicants were found in the water samples, but sediment sample organic toxicant concentrations appeared to be well correlated to sediment texture and color. About 10 to 25 percent of the sediment samples had relatively large concentrations of or- ganics. Bacteria analyses indicated some relatively high bacteria counts in a small per- centage of the samples. Bacteria were found in lower amounts during sampling periods that were extremely hot or extremely cold. Pacific Northwest samples also had the lowest bacteria counts. The data were used to develop and test predictive equations based on site conditions. These models were shown to be valid for most of the data, but the highest concentrations were not well predicted. Therefore, special comparisons of many site conditions were made for the manholes having water with the highest concentrations of critical constituents for comparison to the other locations. It was interesting to note that about half of the prob- lem manholes were repeated samples from the same sites (after complete pumping), but at different seasons, indicating continuous problems and not discrete incidents. In addition, the problem manholes were found for all areas of the country and for most rain conditions. Water clarity and color, along with sediment texture, were found to be significant factors associated with the high concentrations of other constituents, while land use was also noted as a significant factor. These factors can be used to help identify problem manholes, but the rates of false positives and false negatives were found to be high. Therefore, these screening criteria can be used to identify more likely problematic manholes, but other methods (such as confirmation chemical analyses) are also needed to identify those that could not be identified using these simpler methods. continues next page

MONITORING AND MODELING 291 BOX 4-3 Continued The field analytical test methods worked reasonably well, but had much higher detec- tion limits than advertised, limiting their usefulness. Due to the complexity and time needs for many of these on-site analyses, it is usually more effective to analyze samples at a central facility. For scheduled maintenance operations, a crew could arrive at the site be- fore the maintenance time to collect samples and have them analyzed before the mainte- nance crew arrives. For emergency repairs, it is possible to pump the collected water into a tank truck for later analyses, treatment, and disposal. The treatment scenario developed and tested is relatively rapid and cheap and can be used for all operations, irrespective of screening analyses. Chemical addition (using ferric chloride) to the standing water in the manhole was found to reduce problematic levels of almost all constituents to low levels. Slow pumping from the water surface over about a 15- to 30-minute period, with the discharged water then treated in 20-µm cartridge filters, allows the manhole to be entered and the repairs made relatively rapidly, with the water safely discharged. The remaining several inches of water in the bottom of the manhole, along with the sediment, can be removed at a later time for proper disposal. SOURCE: Pitt et al. (1998). industrial trade group to coordinate the monitoring and to apply for a group permit. This was a significant effort that was conducted over several years and involved the participation of many regional facilities throughout the nation. This coordinated effort spread the cost over these different participants, and also allowed significant amounts of data to be collected, control practices to be evaluated, and the development of screening methods that allow emergency maintenance operations of the telecommunication system to proceed in a timely manner. The experimental design of this monitoring program allowed an effi- cient examination of factors affecting stormwater discharges from these opera- tions. This enabled the efficient implementation of effective control programs that targeted specific site and operational characteristics. Although the total cost for this monitoring program was high, it was much less costly than if each indi- vidual company had conducted their own monitoring. In addition, this group effort resulted in much more useful information for the industry as a whole. Outfall Monitoring at a Single Industrial Facility for Permit Compliance and to Demonstrate Effectiveness of Control Practices Sampling at an individual facility results in outfall data that can be com- pared to pre-control conditions and numeric standards. There are many guid- ance documents and reports available describing how to monitor stormwater at an outfall. Two comprehensive sources that describe stormwater monitoring procedures include the handbook written by Burton and Pitt (2002) and a recent guidance report prepared by Shaver et al. (2007). There are a number of basic

292 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES components that need to be included for an outfall characterization monitoring effort, many which have been described in this report. These include the follow- ing: rainfall monitoring in the drainage area (rate and depth, at least at two locations). flow monitoring at the outfall (calibrated with known flow or using dye dilution methods). flow-weighted composite sampler, with sampler modified to accom- modate a wide range of rain events. recommended use of water quality sonde to obtain high-resolution and continuous measurements of such parameters as turbidity, conductivity, pH, oxidation reduction potential, dissolved oxygen (DO), and temperature. preparation of adequate experimental design that quantifies the needed sampling effort to meet the data quality objectives (adequate numbers of sam- ples in all rain categories and seasons). selection of constituents that meet monitoring objectives. In addition, the analytical methods must be appropriately selected to minimize “nonde- tected” values. monitoring station maintenance must also be conducted appropriately to ensure reliable sample collection. Sampling plan must also consider sample retrieval, sample preparation and processing, and delivery to the analytical labo- ratory to meet quality control requirements. Burton and Pitt (2002) describe these monitoring components in detail, along with many other monitoring elements of potential interest (e.g., receiving water biological, physical, and chemical monitoring, including sediment and habitat studies), and include many case studies addressing these components, along with basic statistical analyses and interpretation of the collected data. Box 4-4 pro- vides a detailed example of industrial stormwater monitoring at individual sites in Wisconsin. In general, monitoring of industries should be tailored to their stormwater pollution potential, considering receiving water uses and problems. There are a number of site survey methods that have been developed to rank industry by risk that mostly rely on visual inspections and information readily available from regional agencies. The Center for Watershed Protection developed a hot-spot investigation procedure that is included in the Urban Subwatershed Restoration Manual No. 11 (Wright et al., 2005). This site survey reconnaissance method ranks each site according to its likely stormwater pollutant discharge potential. A detailed field sheet is used when surveying each site to assist with the visual inspections. Cross and Duke (2008) developed a methodology, described in greater detail in Chapter 6, to visually assess industrial facilities based on the level of activities exposed to stormwater. They devised four categories— Category A, no activities exposed to stormwater; Category B, low intensity; Category C, medium intensity; and Category D, high intensity—and tested this

MONITORING AND MODELING 293 BOX 4-4 Wisconsin’s Monitoring of Industrial Stormwater The State of Wisconsin also uses a site assessment method to rank industrial opera- tions into three tiers, mostly based on their standard industrial codes. This system groups facilities by industry and how likely they are to contaminate stormwater. The general per- mits differ in monitoring requirements, inspection frequency, plan development require- ments, and the annual permit fee. The Tier 1 general permit covers the facilities that are considered “heavy” industries, such as paper manufacturing, chemical manufacturing, pe- troleum refining, ship building/repair, and bulk storage of coal, minerals, and ores. The monitoring required of these facilities is presented in this box. The Tier 2 general permit covers facilities that are considered “light” industries and includes such sites as furniture manufacturing, printing, warehousing, and textiles. Facilities with no discharge of contami- nated stormwater are in the Tier 3 category and include sites that have no outdoor storage of materials or waste products. In accordance with the Wisconsin MSGP, Tier 1 industries are required to perform an annual chemical stormwater sampling at each outfall for those residual pollutants listed in the industry’s stormwater pollution prevention plan. The one runoff event selected for sam- pling must occur between March and November and the rainfall depth must be at least 0.1 inch. At least 72 hours must separate the sampled event and the previous rainfall of 0.1 inch. The concentration of the pollutant must represent a composite of at least three grab samples collected in the first 30 minutes of the runoff event. There is concern about the value of collecting so few samples from just one storm each year. To evaluate how well this sampling protocol characterizes pollutant concentrations in industrial runoff, the Wisconsin Department of Natural Resources partnered with the USGS to collect stormwater samples from three Tier 2 industrial sites (Roa-Espinosa and Ban- nerman, 1994). Seven runoff events were monitored at each site, and the samples were collected using five different sampling methods, including (1) flow-weighted composites, (2) time-based discrete samples, (3) time-based composites, (4) a composite of discrete sam- ples from first 30 minutes, and (5) time-based composite sheet flow samples. The first three methods have been described previously. For the composite of discrete samples from the first 30 minutes, the sampler is programmed to take an aliquot after a set period of time (usually every 5 minutes) and the aliquots are combined into one container. The sam- pler stops collecting samples after 30 minutes. For many sites the samples are collected manually, so there is a high probability the sample does not represent the first 30 minutes of the event. For the time-based composite sheet flow samples, a sheet flow sampler is programmed to take an aliquot of sheet flow after a set period of time (usually about every 5 to 15 minutes). All the aliquots are deposited in one bottle beneath the surface of the ground. All of the parts of the hydrograph receive equal weight in the final concentration, but the larger number of aliquots makes for a reasonably accurate composite concentra- tion. This method is unique in that it can be placed near the source of concern. Automatic samplers were used for the first four methods, while sheet flow samplers designed by the USGS were used for the fifth method (Bannerman et al., 1993). Samples were collected during the entire event. All the automatic samplers had to be installed at a location with concentrated flow, such as an outfall pipe, while the sheet flow samplers could be installed in the pavement near a potential source, such as a material storage area. continues next page

294 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES BOX 4-4 Continued The time-based discrete, time-based composite, first-30-minute composite, and sheet flow samples were analyzed for COD, total recoverable copper, total recoverable lead, total recoverable zinc, TSS, total solids, and hardness. In addition to these constituents, the flow-weighted composite samples were analyzed for antimony, arsenic, beryllium, chro- mium, ammonia-N, nitrate plus nitrite, TKN, and TP. All the analysis was done at the State Laboratory of Hygiene in Madison, Wisconsin, and the data are stored in the USGS’s QWDATA database. The number of samples collected during a runoff event varied greatly among the five types of sampling. By design, the median number of samples collected for the first 30 min- utes was three. Limits on the funds available for laboratory cost limited the time-based discrete sampling to about six per storm. Since they are not restricted by laboratory cost, the composites can be based on more sub-samples during a storm. Thus, the median numbers of sub-samples collected for the flow-weighted composite and time-based com- posite were 13 and 24, respectively. The time-based composite sheet flow sample could not document the number of samples it collected, but it was set to collect a sample every few minutes. To judge the accuracy of the sampling methods, one method had to be selected as the most representative of the concentration and load affecting the receiving water. Be- cause a relatively large number of samples are collected and the timing of the sampling is weighted by volume, the flow-weighted composite concentrations were used as the best representation of the quality of the industrial runoff. Concentrations in water samples col- lected by the time-based composite method compared very well to those collected by the flow-weighted composite method, especially if the time-based composite resulted in 20 sub-samples or more. This was not true for the discrete sampling method, because many fewer sub-samples were used to represent changes across the hydrograph. The time- based composite sheet flow sampler produced concentrations slightly higher than the time- based composite samplers collecting water in the concentrated flow. Concentrations from the sheet flow sampler are probably not diluted by other source areas such as the roof. Concentrations of total recoverable zinc and TSS collected in the first 30 minutes of the event were usually two to three times higher than the flow-weighted composite sam- ples. For many of the events, the highest concentration of these constituents occurred in the first 10 minutes of the event. Although the concentrations might be higher in the first part of the event, the earlier parts of the event might only represent one third or less of the total runoff volume. Thus, using the concentrations from the first 30 minutes of the event could greatly overestimate the constituent load from the site. Along with accuracy, the selection of an appropriate sampling method must consider cost and the criteria for installing the sampling equipment. To measure flow, the site must have a location where the flow is concentrated, such as a pipe or well-defined channel, and the runoff is just coming from the site. Out of 474 sites evaluated for this project, only 14 met the criteria for an accurate flow measurement. A few more sites might be suitable for using an automatic sampler without flow measurements, but the number of sites would still be limited. Sheet flow samplers can be used on most sites, since they are simply installed in the pavement near the source of concern. For each sampling method, approximate costs were determined including equipment, installation of equipment, and the analysis of one sample (Table 4-5). Collecting the sam- ples and processing the data should also be included, but they were not because this cost is highly variable. Flow-weighted composite and time-based discrete sampling had the highest cost. Flow measurements made the composite sampling more expensive, while the laboratory cost of analyzing six discrete samples increased the cost of the time-based

MONITORING AND MODELING 295 TABLE 4-5 Cost of Using Different Sampling Methods in 1993 Dollars Estimated Cost for Equipment, Method Installation, and Analysis of One Sample Flow-weighted composite $16,052 Time-based discrete $22,682 Time-based composite $5,920 First-30-minutes (automatic sampler) $6,000 1 First-30-minutes (grab sample) $1,800 Time-based composite sheet flow sampler $2,889 1 Cost of laboratory analysis only. SOURCE: Reprinted, with permission, from Roa- Espinosa and Bannerman (1994). Copyright 1994 by the American Society of Civil Engi- neers. discrete method. It should be noted that hand grab samples could be used to collect the discrete samples in the first 30 minutes at lower cost, although this depends strongly on the skill of the person collecting the sample. The sheet flow sampler could be the most cost effective approach to sampling an industrial site. A determination must be made of how many runoff events should be sampled in order to accurately characterize a site’s water quality. As shown in Table 4-6, representing a site with the results from one storm can be very misleading. Concentrations in Table 4-6 were collected by the flow-weighted composite method. The geometric means of EMCs from five or more events were very different than the lowest or highest concentration observed for the set of storms. The chances of observing an extreme value by sampling just one event is increased by selecting a sampling method designed to collect a limited number of sub-samples, such as the first-30-minutes method. Too few storms were monitored in this project to properly evaluate the variability in the EMCs, but sufficient changes occur be- tween the zinc and TSS geometric means in Table 4-6 to suggest that a compliance moni- toring schedule should include a minimum of five events be sampled each year. To overcome the high COV observed for municipal stormwater data collected in Wis- consin, EMCs should be determined for about 40 events (Selbig and Bannerman, 2007; Horwatich et al., 2008). The 40 event mean concentrations would probably represent the long-range distribution of rainfall depths, and there would be sufficient data available to perform some trend analysis, such as evaluating the benefits of an SCM implemented at an industrial site. Monitoring 40 events each year, however, would be too costly for an annual compliance monitoring schedule for each industrial site. Results from this project indicate that the stormwater monitoring required at industrial sites cannot adequately characterize the quality of runoff from an industrial site. Only col- lecting samples from the first 30 minutes of a storm is probably an overestimate of the con- centration, and a load calculated from this concentration would exaggerate the impact of the site on the receiving waters. Time- and flow-based composite sampling would be much better methods for monitoring a site if there are locations to operate an automatic sampler. For sites without such a location, the time-based composite sheet flow sampler offers the best results at the least cost. Given all the variability in concentrations between runoff events, the annual monitoring schedule for any site should include sampling multiple storms. continues next page

296 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES BOX 4-4 Continued TABLE 4-6 Effects of Including a Different Number of Events in the Geometric Mean Cal- a culation for Zinc and TSS Number of Events Total Recoverable Zinc Total Suspended Solids AC Rochester 1 (Lowest Concentration) 57 8 1 (Highest Concentration) 150 84 3 76 24 5 91 36 PPG Industries 1 (Lowest Concentration) 140 32 1 (Highest Concentration) 330 49 3 153 57 6 186 53 Warman International 1 (Lowest Concentration) 68 17 1 (Highest Concentration) 140 56 3 67 15 5 81 26 7 74 19 a Samples were collected using the flow-weighted composite method. SOURCE: Reprinted, with permission, from Roa-Espinosa and Bannerman (1994). Copyright 1994 by the American Society of Civil Engineers. scheme by examining many southern Florida industrial facilities. About 25 per- cent of the facilities surveyed that were officially included in the stormwater permit program had no stormwater exposure (Category A), but very few had submitted the necessary application to qualify for an exception under the “no exposure” rule. Slightly more than half of the of the surveyed facilities were included in the “no exposure” and “low exposure” categories, obviously deserv- ing less attention compared to the higher impact categories. Recommendations for Industrial Stormwater Monitoring Suitable industrial monitoring programs can be implemented for different categories of industrial activities. The following is one such suggestion, based on the likely risks associated with stormwater discharges from each type of fa- cility. No Exposure to Industrial Activities and Other Low-Risk Industrial Operations For sites having limited stormwater exposure to industrial operations, such as no outdoor storage of materials or waste products, basic monitoring would

MONITORING AND MODELING 297 not normally be conducted. However, roof runoff (especially if galvanized met- als are used) and large parking areas need to be addressed under basic stormwa- ter regulations dealing with these common sources of contaminants and the large amounts of runoff that may be produced. Simple SCM guidance manuals can be used to select and size any needed controls for these sites, based on the areas of concern at the facility. For these facilities, simple visual inspections with no monitoring requirements may be appropriate to ensure compliance with the ba- sic stormwater regulations. A regionally calibrated stormwater quality model can be used to evaluate these basic stormwater conditions and to calculate the expected benefits of control measures. Periodic random monitoring of sites in this category should be conducted to verify the small magnitude of discharges from these sites and the performance of SCMs. Medium-Risk Industrial Operations For “medium-intensity” industry facilities, site inspections and modeling should be supplemented with suitable outfall monitoring to ensure compliance. As noted in Box 4-2, there can be a tremendous amount of variability in indus- trial runoff characteristics. However, the dataset described in that example was a compilation of data from many different types of facilities, with no separation by industrial type. Even different facilities in a single industrial group may have highly variable runoff characteristics. However, a single facility has much less variability, and reasonable monitoring strategies can be developed for compli- ance purposes. As noted in Box 4-4, about 40 samples were expected to be needed for each site in that example. With typical permit periods of five years, this would require that less than ten samples per year (more than the three sam- ples per year currently obtained at many locations) be collected in order to de- termine the EMC for the site for comparison to allowable discharge conditions. Obviously, the actual number of samples needed is dependent on the variability of the runoff characteristics and the allowable error, as described elsewhere. After about 10 to 15 storms have been monitored for a site, it would be possible to better estimate the total number of samples actually needed based on the data quality objectives. If the monitoring during the permit period indicated exces- sive stormwater discharges, then the SCMs are obviously not adequate and would need improvement. The permit for the next five-year period could then be modified to reflect the need for more stringent controls, and suitable fines accessed if the facility was not in compliance. It is recommended that absolute compliance not be expected in the industrial permits, but that appropriate benchmarks be established that allow a small fraction of the monitored events to exceed the goals. This is similar to discharge permit requirements for combined sewers, and for air quality regulations, where a certain number of excessive pe- riods are allowed per year.

298 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES High-Risk Industrial Facilities For “high-risk” industrial sites of the most critical nature, especially if non- compliance may cause significant human and environmental health problems, visual inspections and site modeling should be used in conjunction with moni- toring of each event during the permit period. Because of the potential danger associated with noncompliance, the most stringent and robust controls would be required, and frequent monitoring would be needed to ensure compliance. If noncompliance was noted, immediate action would be needed to improve the discharge conditions. This is similar to industrial and municipal NPDES moni- toring requirements for point sources. MODELING TO LINKING SOURCES OF POLLUTION TO EFFECTS IN RECEIVING WATERS Stormwater permitting is designed to regulate dischargers, develop informa- tion, and reduce the level of stormwater pollutants and impact on receiving wa- terbodies. An important assumption is that the level of understanding of the stormwater system, through a combination of monitoring and modeling, is suffi- cient to associate stormwater discharges with receiving waterbody impacts. Impairment of waterbodies can occur for a variety of physical, chemical, and biological reasons, often with a complex combination of causes. The ambient water quality of a receiving waterbody, which may result in a determination of impairment, is itself a function of the total mass loading of pollutant; dilution with stream discharge or standing waterbody volume; the capacity of the aquatic ecosystem to assimilate, transform, or disperse the pollutant; and transport out of the waterbody. In addition to the chemical and physical attributes of the water, impairment may also be characterized by degraded biologic structure or geo- morphic form of the waterbody (e.g., channel incision in urban areas). Interac- tions between multiple pollutant loadings, long turnover and residence times, saturation effects, and cascading feedbacks with biological communities com- plicate the apparent response of waterbodies to pollutant discharge. This is par- ticularly important when considering cumulative watershed effects, in which interactions between stressors and long-term alteration of watershed conditions may contribute to threshold responses of a waterbody to continued loading or alteration. Under these conditions, simple “loading-response” relations are often elusive and require consideration of historical and local watershed conditions. As an example, pollutant loading at high stream flow or into strong tidally flushed systems may be advected downstream or into the coastal ocean without building up significant concentrations, while pollutant loading at low flow may not be effectively transported and dispersed and may build up to harmful con- centrations. In the former case the pollutant may be rapidly transported out of the local waterbody, but may impact a more distant, downstream system. In addition, certain pollutants, such as inorganic nitrogen, may be discharged into

MONITORING AND MODELING 299 surface waters and subsequently transformed and removed from the water col- umn into vegetation or outgassed (e.g., volatilized or denitrified) into the atmos- phere under certain ecosystem conditions. Sediment and other pollutants may be stored for long time periods in alluvial or lacustrine deposits, and then remo- bilized long after the initial loading into a stream reach or standing waterbody in response to extreme climate events, land-use change, reservoir management, or even reductions in the pollutant concentrations in the water column. Conse- quently, long lags may exist between the actual discharge of the sediment (and any pollutants adsorbed or otherwise stored within the deposits) and their con- tribution to waterbody impairment. Therefore, understanding the fate of pollut- ants, particularly nonconservative forms, may require consideration of the full ecosystem cycling and transport of the material over long time periods. Impairment of waterbodies can be assessed on the basis of biological indi- cators, as discussed in Chapter 2. As organisms and communities respond to multiple stressors, it is not always clear what the direct or indirect effects of any specific pollutant discharge is, or how that may be exacerbated by correlated or interacting activity in the watershed. The association of specific types of im- pairment with surrounding land use implicitly accounts for these interactions but does not provide a mechanistic understanding of the linkage sufficient to specify effective remedial activity. However, much progress has been made in deter- mining toxic effects of certain contaminants on different aquatic species assem- blages (see, e.g., Shaver et al., 2007) and on quantifying impacts of land use on flow duration curves, EMCs, and loading rates for a number of pollutants (Maestre and Pitt, 2005). For the latter effort, it has been shown that there is large variability within land-use categories, both as a function of specific SCMs and of innate differences due to historical legacies, climate, and hydrogeology. A protocol linking pollutants in stormwater discharges to ambient water quality criteria should be based on conservation of mass, in which the major inputs, outputs, transformations, and stores of the pollutant can be quantified. Indeed, these are the components of hydrologic and watershed models used to simulate the fate and transport of stormwater and its pollutants. SCMs that im- prove ambient water quality criteria are designed to act on one or more of these mass balance terms. A number of these measures act to reduce the magnitude of a stormwater source (e.g., porous pavement), while others are designed to ab- sorb or dissipate a pollutant within a hydrologic flowpath downstream from a source (e.g., rain garden, detention pond, stream restoration). The latter requires some consideration of the flowpath from the source to the receiving waterbody. Therefore, determining the major sources, sinks, and transformations of the pol- lutant should be the first step in this procedure. For a number of pollutants there may be very few potential sources, while for others there may be multiple sig- nificant sources. The spatial diversity of these sources and sinks may also range from uniform distribution to “hot spot” patterns that are difficult to detect and quantify. Many stormwater models work effectively with sources, but are not structured to follow the transport or transformation of pollutants from source to waterbody along hydrologic flowpaths. Figure 4-14 shows the drainage area of Jordan Lake, an important regional

300 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES FIGURE 4-14 The drainage area to Jordan Lake, a major drinking water reservoir in the Triangle area of North Carolina, is under nutrient-sensitive rules, requiring reductions in total nitrogen and phosphorus. Drainage flowlines and catchment areas are from NHDplus, and are shaded according to their percentage of industrial and commercial land cover from the NLCD. The area outlined in black is a small urban catchment, detailed in Figure 4-15, and comprised of a wooded central region, surrounded by residential and institutional land + use. SOURCE: Data from the NHD .

MONITORING AND MODELING 301 drinking water source in the Triangle area of North Carolina. Catchment areas are shaded to relate the percentage of industrial and commercial land cover, ac- cording to the National Land Cover Database (NLCD). Figure 4-15 shows a small tributary within the Jordan Lake watershed in Chapel Hill (outlined in Figure 4-14) with a high-resolution image of all impervious surfaces overlain on the topographically defined surface flowpath network. Each of the distributed sources of stormwater is routed through a flowpath consisting of other pervious and impervious segments, within which additions, abstractions, and transforma- tions of water and pollutants occur depending on weather, hydrologic, and eco- system conditions. The cumulative delivery and impact of all stormwater sources include the transformations occurring along the flowpaths, which could include specific SCMs such as detention or infiltration facilities or simply infil- tration or transformations in riparian areas or low-order streams. The riparian area may be bypassed depending on stormwater concentration or piping, and it may have various levels of effectiveness on reducing pollutants depending on geomorphic, ecosystem, and hydrologic conditions. The ability of a stormwater model to capture these types of effects is a key property influencing its ability to associate a stormwater source with a waterbody outcome. FIGURE 4-15 A small urban catchment in the Lake Jordan watershed of North Carolina with distributed sources of impervious surface (buildings and roads) stormwater arranged within the full surface drainage flowpath system. Stormwater from each source is routed down surface and subsurface flowpaths to the nearest tributary and out the drainage network, with additions and abstractions of water and pollutants along each flowpath + segment. SOURCE: Data from the NHD .

302 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES This section discusses the fundamentals of stormwater modeling and the capabilities of commonly used models. Much of this information is captured in a summary table at the end of the section (Table 4-7). The models included are the following: The Rational Method, or Q = C*I*A, where Q is the peak discharge for small urban catchments, A is the catchment area, I is the rainfall intensity, and C is a rainfall-runoff coefficient. The Simple Method, which classifies stormwater generation and impact regimes by the percent impervious cover TR-20 and TR-55 The Generalized Watershed Loading Function (GWLF) Program for Predicting Polluting Particle Passage through Pits, Pud- dles, and Ponds (P8) Model for Urban Stormwater Improvement Conceptualization (MU- SIC) Stormwater Management Model (SWMM) Source Loading and Management Model (WinSLAMM) Soil and Water Assessment Tool (SWAT) Hydrologic Simulation Program–Fortran (HSPF) Western Washington Hydrologic Model Chesapeake Bay Watershed Model (CBWM) Fundamentals of Stormwater Models Stormwater models are designed to evaluate the impacts of a stormwater discharge on a receiving waterbody. In order to do this, the model must have the capability of describing the nature of the source term (volumes, constitu- ents), transport and transformation to the receiving waterbody, and physical, chemical, and biological interaction with the receiving water body and ecosys- tem. No model can mechanistically reproduce all of these interactions because of current limitations in available data, incomplete understanding of all proc- esses, and large uncertainties in model and data components. Computer re- sources, while rapidly advancing, still limit the complexity of certain applica- tions, especially as spatial data become increasingly available and it is tempting to model at ever-increasing resolution and comprehensiveness. Therefore, mod- els must make a set of simplifying assumptions, emphasizing more reliable and available data, while attempting to retain critical processes, feedbacks, and in- teractions. Models are typically developed for a variety of applications, ranging from hydraulic design for small urban catchments to urban and rural pollutant loading at a range of watershed scales.

MONITORING AND MODELING 303 An evaluation of the current state of stormwater modeling should say much about our ability to link pollutant sources with effects in receiving waters. Both stormwater models and models supporting the evaluation of SCM design and effectiveness are based on simulating a mass budget of water and specific pol- lutants. The detail of mass flux, transformation, and storage terms vary depend- ing on the scale and purpose of the application, level of knowledge regarding the primary processes, and available data. In many cases, mechanisms of transfor- mation may be either poorly understood or may be dependent on detailed inter- actions. As an example, nitrogen-cycle transformations are sensitive to very short temporal and spatial conditions, termed “hot spots” and “hot moments” relative to hydrologic flowpaths and moisture conditions (McClain et al., 2003). Stormwater runoff production and routing are common components of these models. All models include an approach to estimate the production of stormwa- ter runoff from one or more zones in the watershed, although runoff routing from the location(s) of runoff production to a point or waterbody is not always included explicitly. Major divisions between approaches are found in the repre- sentation of the watershed “geography” in terms of patterns and heterogeneity, and in runoff production and routing. Some stormwater models do not consider the effects of routing from a runoff source to a local waterbody directly, but may attempt to reproduce net impacts at larger scales through the use of unit hydro- graph theory to estimate peak flows, and delivery ratios or stormwater control efficiency factors to estimate export to a waterbody. There are a number of different approaches and paradigms used in stormwa- ter models that include varying degrees of watershed physical, biological, and chemical process detail, as well as spatial and temporal resolution and the repre- sentation of uncertainty in model estimates. A number of researchers have writ- ten about the nature of watershed models (e.g., Beven, 2001; Pitt and Vorhees, 2002). At present, many hydrologic and stormwater models have become so complex, with multiple choices for different components, that standard descrip- tions apply only to specific components of the models. The following discus- sion is generalized; most models fit the descriptions only to certain degrees or only under specific conditions in which they are operated. Lumped Versus Distributed Approaches Central to the design of watershed models is the concept of a “control vol- ume,” which is a unit within which material and energy contents and balances are defined, with boundaries across which material and energy transport occurs. Control volumes can range from multiple subsurface layers and vegetation can- opy layers bounded in three dimensions to a full watershed. Lumped models ignore or average spatial heterogeneity and patterns of watershed conditions, representing all control volumes, and the stores, sources, and sinks of water and pollutants in a vertically linked set of conceptual components, such as surface interception, unsaturated and saturated subsurface zones, and a single stream or

304 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES river reach. For example, SWAT or HSPF are conceptually lumped at the scale of subwatersheds (e.g., the level of geography in Figure 4-14) and do not show any spatial patterns at higher resolutions (e.g., Figure 4-15) than these units. While multiple land-use/soil combinations may be represented, these models do not represent the connectivity of the land segments (e.g., which land segments drain into which land segments) and assume all unique land segment types drain directly to a stream. Distributed models include some scheme to represent spatial heterogeneity of the watershed environment pertinent to stormwater generation, including land cover, soils, topography, meteorological inputs, and stream reach properties dis- tributed through a set of linked control volumes. Control volumes representing land elements, including vertically linked surface and subsurface stores, are connected by a representation of water and pollutant lateral routing through a network of flowpaths that may be predefined or set by the dynamics of surface, soil, and saturated zone water storage. The land elements may be grid cells in a regular lattice, or irregular elements (e.g., triangles) with the pattern adapted to variations in land surface characteristics or hydraulic gradients. A number of models are intermediate between lumped and distributed, with approaches such as lumping at the subwatershed scale, incorporating statistical distributions of land element types within subwatersheds but without explicit pattern representation, or lumping some variables and processes (such as groundwater storage and flux), while including distributed representation of topography and land cover. Thus, within the model SLAMM (Pitt and Vorhees, 2002), the catchment is described in sufficient detail to summarize the break- down of different drainage sequences. As an example, roof area will be broken down to the proportion that drains to pervious areas and to directly connected impervious areas. An important distinction is that there is no routing of the out- put of one land element into another, such that there is no drainage sequence that may significantly modify the stormwater runoff from its source to the stream. Implicitly, all land elements drain directly into a stream, although a loss rate or delivery ratio can be specified. The choice of a more lumped or distributed model is often dependent on available data and overall complexity of the model. Simpler, lumped models may be preferred in the absence of sufficient data to effectively parameterize a distributed approach, or for simplicity and computational speed. However, fully lumped models may be limited in their ability to represent spatial dependency, such as the development and dynamics of riparian zones, or the effects of SCM patterns and placement. As there is typically an irreducible level of spatial het- erogeneity in land surface characteristics down to very small levels below the resolution of individual flow elements, we note that all models lump at some scale (Beven, 2000).

MONITORING AND MODELING 305 Mechanistic Versus Conceptual Process Representation Mechanistic, or process-based, approaches attempt to reproduce key storm- water transport and transformation processes with more physically, chemically, or biologically based detail, while conceptual models represent fluxes between stores and transformations with aggregate, simplified mathematical forms. No operational models are built purely from first principles, so the distinction be- tween mechanistic and conceptual process basis is one of degree. The level of sampling necessary to support detailed mechanistic models, as well as remaining uncertainty in physicochemical processes active in heteroge- neous environments typically limits the application of first-principle methods. The development or application of more mechanistic approaches is currently limited by available measurements, which require both time and resources to adequately carry out. Unfortunately, modeling and monitoring have often been mutually exclusive in terms of budgets, although it is necessary for both to be carefully planned and integrated. A new generation of sensors and a more rig- orous and formal sampling protocol for existing methods will be necessary to advance beyond the current practice. At present, most operational hydrologic and transport models are based on a strong set of simplifying assumptions regarding active processes and/or the spa- tial variation of sources, sinks, and stores in the watershed. Runoff production can be computed by a range of more mechanistic to more conceptual or empiri- cal methods. More mechanistic methods include estimation of infiltration ca- pacities based on soil hydraulic properties and moisture conditions, excess run- off production, and hydraulic routing over land surfaces into and through a stream-channel network. More conceptual approaches use a National Resources Conservation Service (NRCS) curve number approach (see Box 4-5) and unit hydrograph methods to estimate runoff volume and time of concentration. Pol- lutant concentrations or loads are often estimated on the basis of look-up tables using land use or land cover. Land use- or land cover-specific EMC or unit area loading for pollutants can be developed directly from monitoring data or from local, regional, or national databases. The NSQD statistically summarizes the results of a large number of stormwater monitoring projects (as discussed previ- ously in this chapter). The effects of SCM performance (typically percent re- moval) can be estimated from similar databases (e.g., www.bmpdatabase.org). A set of models, such as SWAT, incorporate fairly detailed descriptions of nu- trient cycling as an alternative to using EMC, requiring more detailed inputs of soil, crop, and management information. Unfortunately, the detailed biogeo- chemistry of this and similar models is typically not matched by the hydrology, which remains lumped at individual Hydrologic Response Unit (HRU) levels using NRCS curve number methods, although options exist to incorporate more mechanistic infiltration excess runoff.

306 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES BOX 4-5 NRCS Technical Release 55 NRCS methods to estimate runoff volumes and flows have been popular since the early 1950s (Rallison, 1980). Fundamentally they can be broken into the separation of runoff from the rainfall volume (Curve Number Method), the pattern of runoff over time (di- mensionless unit hydrograph), and their application within computer simulation models. In the late 1970s these components were packaged together in a desktop hydrology method known as Technical Release 55 (TR-55). TR-55 became the primary model used by the majority of stormwater designers, and there is considerable confusion over the terms used to describe what aspects of the NRCS methods are in use. The NRCS Curve Number Method was first derived in the 1950s for prediction of run- off from ungauged agricultural areas. It relates two summation ratios, that of runoff to rain- fall and that of moisture retained to maximum potential retention. Two statistically based relations were developed to drive the ratio, the first of which is based on a “curve number” which depicts the soil type, land cover, and initial moisture content. The second or initial abstraction is defined as the volume of losses that occur prior to the initiation of runoff, and is also related to the curve number. Data were used to derive curve numbers for each soil type and cover as shown in Figure 4-16 (Rallison, 1980). FIGURE 4-16 Development of curve number from collected data. SOURCE: Reprinted, with permission, from Rallison (1980). Copyright 1980 by the American Society of Civil Engineers.

MONITORING AND MODELING 307 The Curve Number method is a very practical method that gives “average” runoff re- sults from a watershed and is used in many models (WIN TR-55, TR-20, SWMM, GWLF, HEC-HMS, etc.). Caution has to be exercised when using it for smaller urbanizing storm events. For example, past practice was to average curve numbers for developments for pavement and grass based on percent imperviousness. While this works well for large storms, for smaller storms it gives erroneous answers through violation of the initial ab- straction relationship. Current state manuals (MDE, 2000; PaDEP, 2006) do not allow paved- and unpaved-area curve numbers to be averaged. When applied to continuous simulation models (such as in SWMM or GWLF), it requires an additional method to re- cover the capacity to remove runoff because the soil capacity to infiltrate water is restored over time. The NRCS Dimensionless Unit Hydrograph has also evolved over many years and simply creates a temporal pattern from the runoff generated from the curve number method. This transformation is based upon the time of concentration, defined as the length of time the water takes to travel from the top to the bottom of the watershed. The dimen- sionless curve ensures that conservation of mass is maintained. The main purpose of this method is to estimate how long it takes the runoff generated by the curve number to run off the land and produce discharge at the watershed outlet. The NRCS curve number and dimensionless unit hydrograph were first incorporated in the Soil Conservation Service (SCS) TR-20 hydrologic computer model developed in the 1960s. As most stormwater professionals did not have access to mainframes, SCS put together TR-55, which created a hand or calculator method to apply the curve number and dimensionless unit hydrograph. In order to create this hand method, many runs were gen- erated using TR-20 to develop patterns for different times of concentration. The difficulty with using the original TR-55 in the modern era is that the simplifications to the hydrograph development do not allow the benefits of SCMs to be easily accounted for. The use of the term TR-55 has been equated with the curve number method; this has created confusion, especially when it is included in municipal code. Further clouding the issue, there are two types of TR-55 computer models available. One is based on the origi- nal, outdated, simplified hand method, and the other (Win TR-55) returns to the more ap- propriate application of the curve number and dimensionless hydrograph methods. In ei- ther case, the focus of these models is on single event hydrology and cannot easily incor- porate or demonstrate the benefits of the wide range of structural and nonstructural SCMs. Note that the curve number and dimensionless unit hydrograph methods are incorporated in many continuous flow models, including SWMM and GWLF, as the basis of runoff gen- eration and runoff timing.

308 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES Deterministic Versus Stochastic Methods Deterministic models are fully determined by their equation sets, initial and boundary conditions, and forcing meteorology. There are no components that include random variation. In a stochastic model, at least one parameter or vari- able is drawn from a probability distribution function such that the same model set-up (initial and boundary conditions, meteorology, parameter sets) will have randomly varying results. The advantage of the latter approach is the ability to generate statistical variability of outcomes, reflecting uncertainty in parameters, processes, or any other component. In fact, any deterministic model can be op- erated in a stochastic manner by sampling parameter values from specified probability distributions. It is recognized that information on the probability distribution of input pa- rameters may be scarce. For situations with limited information on parameter values, one option is to assume a uniform distribution that brackets a range of values of the parameter reported in the literature. This would at least be a start in considering the impacts of the variability of model inputs on outputs. A thor- ough discussion on methods for incorporating uncertainty analysis into model evaluation is provided in Chapter 14 of Ramaswami et al. (2005). It should be noted that the ability to generate probability distribution information on storm- water outcomes requires a potentially large number of model runs, which may be difficult for detailed mechanistic and distributed models that have large com- putational loads. Continuous Versus Event-Based Approaches Another division between modeling approaches is the time domain of the simulation. Event-based models limit simulation time domains to a storm event, covering the time of rainfall and runoff generation and routing. Initial condi- tions need to be estimated on the basis of antecedent moisture or precipitation conditions. For catchments in which runoff is dominated by impervious sur- faces, this is a reasonable approach. In landscapes dominated by variable source area runoff dynamics in which runoff is generated from areas that actively ex- pand and contract on the basis of soil moisture conditions, a fuller accounting of the soil moisture budget is required. Furthermore, event-based modeling is in- appropriate for water quality purposes because it will not reproduce the full dis- tribution of receiving water problems. Continuous models include simulation of a full time domain composed of storm and inter-storm periods, thus tracking soil moisture budgets up to and including storm events. Outfall Models After beneficial use impairments are recognized, cause-and-effect relation-

MONITORING AND MODELING 309 ships need to be established and restorative discharge goals need to be devel- oped. Models are commonly used to calculate the expected discharges for dif- ferent outfalls affecting the receiving water in a community. All of the models shown in Table 4-7 can calculate outfall discharge quantities, although some may only give expected average annual discharge. Models calculate these dis- charges using a variety of processes, but all use an urban hydrology component to determine the runoff quantity and various methods to calculate the quality of the runoff. The runoff quantity is multiplied by the pollutant concentration in the outfall to obtain the mass discharges of the different pollutants. The outfall mass discharge from the various outfalls in the area can then be compared to identify the most significant outfalls that should be targeted for control. The most common hydrology “engines” in simple stormwater models are the NRCS curve number method or a simple volumetric runoff coefficient—Rv, the ratio of runoff to rainfall—for either single rainfall events or the total annual rainfall depth. Runoff quality in the simple models is usually calculated based on published EMCs for similar land uses in the same geographical area. More complex models may use build-up and wash-off of pollutants from impervious surfaces in a time series or they may derive pollutant concentrations from more detailed biogeochemical cycling mechanisms, including atmospheric deposition and other inputs (e.g., fertilizer). Some models use a combination of these proc- esses depending on the area considered, and others offer choices to the model user. Again, these processes all need local calibration and verification to reduce the likely uncertainty associated with the resultant calculated discharge condi- tions. Source Area When the outfalls are ranked according to their discharges of the pollutants of importance, further detailed modeling can be conducted to identify sources of the significant pollutants within the outfall drainage area. Lumped parameter models cannot be used, as the model parameters vary within the drainage area according to the different source areas. Distributed area models can be used to calculate contributions from different source areas within the watershed area. This information can then be used to rank the land uses and source area contri- butions. In-stream responses can be calculated if the land-area models are linked to appropriate receiving-water models. Need for Coupling Models As urban areas become increasingly extensive and heterogeneous, including a gradient of dense urban to forest and agricultural areas, linkage and coupling of models to develop feedback and interactions (e.g., impacts of urban runoff hydraulics with stream scour and sedimentation, mixed with agricultural nutrient

310 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES and sediment production on receiving waterbodies) is a critical area that requires more development. In general, stormwater models were designed to track and predict discharges from sources by surface water flowpaths into receiving wa- terbodies, such that infiltration was considered to be a loss (or retention) of wa- ter and its constituents. To fully evaluate catchment-scale impacts of urbaniza- tion on receiving waterbodies, the infiltration term needs to be considered a source term for the groundwater, and a groundwater component or model needs to be coupled to complete the surface–subsurface hydrologic interactions and loadings to the waterbody. Finally, each of the models may or may not incorporate explicit considera- tion of SCM performance based on design, implementation and location within the catchment. As discussed in the next chapter, SCM models can range from simple efficiency factors (0–1 multipliers on source discharge) to more detailed treatment of physical, chemical, and biological transport and transformations. Linking to Receiving-Water Models Specific problems for urban receiving waters need to be identified through comprehensive field monitoring and modeling. Monitoring can identify current problems and may identify the stressors of importance (see Burton and Pitt [2002] for tools to evaluate receiving water impairments). However, monitoring cannot predict conditions that do not yet exist and for other periods of time that are not represented at the time of monitoring. Modeling is therefore needed to gain a more comprehensive understanding of the problem. In small-scale totally urbanized systems, less complex receiving-water models are needed. However, as the watershed becomes more complex and larger with multiple land uses, the receiving-water models also need to become more complex. Complex receiv- ing-water models need to include transport and transformations of the pollutants of concern, for example. Examples of models shown on the comparison table that include receiving-water processes are MUSIC and HSPF. Other models (such as WinSLAMM) provide direct data links to external receiving-water models. Calibration and verification of important receiving-water processes that are to be implemented in a model can be very expensive and time consuming, and still result in substantial uncertainty. Model Calibration and Verification Calibration is the process where model parameters are adjusted to minimize the difference between model output and field measurements, with an aim of keeping model parameters within a range of values reported in the literature. Model verification, similar to model validation, is used to mean comparison between calibrated model results using part of a data set as input and results from application of the calibrated model using a second (independent) part of

MONITORING AND MODELING 311 the data set as input. Oreskes et al. (1994) present the viewpoint that no model can really be verified; at best, verification should be taken to mean that a model is consistent with a physical system under a given set of comparison data. This is not synonymous with saying that the model can reliably represent the real system under any set of conditions. In general, the water quantity aspects of stormwater modeling are easier to calibrate and verify than the water quality aspects, in part because there are more water quantity data available and because chemical transformations are more complex to simulate. A thorough discussion of the broad topic of model evaluation is provided by several excellent texts on this subject, including Schnoor (1996) and Ramaswami et al. (2005). Models in Practice Today Table 4-7 presents a set of models used for stormwater evaluation that range in complexity from first-generation stormwater models making use of simple empirical land cover/runoff and loading relations to more detailed and informa- tion-demanding models. The columns in Table 4-7 provide an abbreviated de- scription of some of the attributes of these models—common usage, typical ap- plication scales, the degree of model complexity, some data requirements (for the hydrologic component), whether the model addresses groundwater, and whether the model has the ability to simulate SCMs. Models capable of simulat- ing a water quality component require EMC data, with some models also having a simple build-up/wash-off approach to water quality simulation (e.g., SWMM, WinSLAMM, and MUSIC) and others simulating more complex geochemistry (e.g., SWAT and HSPF). The set of columns in Table 4-7 is not meant to be exhaustive in describing the models, which is why websites are provided for comprehensive model descriptions and data requirements. In addition to the models listed in Table 4-7, a representative set of emerg- ing research models that are not specifically designed for stormwater, but may offer some advantages for specific uses, are also described below. In general, it is important that models that integrate hydrologic, hydraulic, meteorologic, wa- ter quality, and biologic processes maintain balance in their treatment of process details. Both model design and data collection should proceed in concert and should be geared toward evaluating and diagnosing the consistency of model or coupled model predictions and the uncertainty attached to each component and the integrated modeling system. The models should be used in a manner that produces both best estimates of stormwater discharge impacts on receiving wa- terbodies, as well as the level of uncertainty in the predictions. The Rational Method is a highly simplified model widely used to estimate peak flows for in sizing storm sewer pipes and other low level drainage path- ways. The method assumes a constant rainfall rate (intensity), such that the run- off rate will increase until the time at which all of the drainage area contributes to flow at its outlet (termed the time of concentration). The product of the drainage area and rainfall intensity is considered to be the input flow rate to the

312 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES drainage area under consideration; the ratio of the input flow rate to an outflow discharge rate is termed the runoff coefficient. Runoff coefficients for a variety of land surface types and slopes have been compiled in standard tables (see e.g., Chow et al., 1988). The outflow is determined by multiplying inflow (rainfall intensity times drainage area) by the runoff coefficient for the land-surface type. As pointed out by Chow et al. (1988), this method is often criticized owing to its simplified approach, so its use is limited to stormwater inlet and piping designs. The Simple Method estimates stormwater pollutant loads for urban areas, and it is most valuable for assessing and comparing the relative stormwater pol- lutant load changes of different land use and stormwater management scenarios. It requires a modest amount of information, including the subwatershed drainage area and impervious cover, stormwater pollutant concentrations (as defined by the EMC), and annual precipitation. The subwatershed can be broken up into specific land uses, such that annual pollutant loads are calculated for each type of land use. Stormwater pollutant concentrations are usually estimated from local or regional data, or from national data sources. The Simple Method esti- mates pollutant loads for chemical constituents as a product of annual runoff volume and pollutant concentration, as L = 0.226 R x C x A, where L = annual load (lbs), R = annual runoff (inches), C = pollutant concentration (mg/l), and A = area (acres). Of slightly increased complexity are those models initially developed dec- ades ago by the Soil Conservation Service, now the NRCS of the U.S. Depart- ment of Agriculture (USDA). NRCS Technical Releases (TR) 20 and 55 are widely used in many municipalities, despite the availability of more rigorous, updated stormwater models. Box 4-5 provides an overview of the NRCS TR-55 assumptions and approaches. A number of watershed models that are used for stormwater assessment are lumped, conceptual forms, with varying levels of process simplification and spatial patterns aggregated at the subwatershed level, with aspatial statistical distribution of land types as described above. The GWLF model (Haith and Shoemaker, 1987) is an example of this type of approach, using simple land use- based EMC with NRCS curve number estimates of runoff within a watershed context. GWLF is a continuous model with simplified upper- and lower-zone subsurface water stores, and a simple linear aquifer to deliver groundwater flow. EMCs are assigned or calibrated for subsurface and surface flow delivery, while sediment erosion and delivery are computed with the use of the Universal Soil Loss Equation and delivery coefficients. The methods are easily linked to a Geographical Information System (GIS), which provides land-use composition at the subwatershed level and develops estimates of runoff and loading that are typically used to estimate annual loading. AVGWLF links GWLF with Arc- View and is used as a planning- or screening-level tool. A recent example of AVGWLF for nutrient loading linked to a simple stream network nutrient decay model for the development of a TMDL for a North Carolina water supply area is given in Box 4-6.

MONITORING AND MODELING 313 BOX 4-6 The B. Everett Jordan Lake GWLF Watershed Model Development Jordan Lake is a regionally important water supply reservoir at the base of the 1,686- square-mile Haw watershed in North Carolina (see Figure 4-17). It is considered a nutrient- sensitive waterbody. Officials are now in the process of implementing watershed goals to reduce nitrogen and phosphorus, with the reduction goals differentiated by geographic location within the basin. In support of the development of these rules as part of a TMDL effort, the North Carolina Division of Water Quality commissioned a water quality modeling study (Tetra Tech, 2003). The modeling effort was needed to support the evaluation of nutrient reduction strategies in different parts of the watershed relative to Jordan Lake, which requires both a model of nutrient loading, as well as river transport and transformation. Given data and resource restrictions, a more detailed model was not considered feasible. As GWLF does not support nutrient transformations in the stream network, the model was used in conjunction with a method to decay nutrient source loading by river transport distance to the lake. A spreadsheet model was designed to take as input GWLF estimates of seasonal loads for 14-digit hydrologic unit code (HUC) subbasins of the Haw, and to reduce the loads by river miles between the subwatershed and Jordan Lake. The GWLF loading model was calibrated to observations in small subwatersheds within the Haw using HRUs developed from soil and NLCD land classes, updated with additional information from county GIS parcel databases and the 2000 Census. This information was used to estimate subwatershed impervious surface cover, fertilizer inputs, runoff curve numbers, soil water capacity, and vegetation cover to adjust evapotranspiration rates. Wastewater disposal (sewer or septic) was estimated on the basis of urban service boundaries. GWLF was used to provide loading estimates, using limited information on soil and groundwater nutrient concentrations, and calibrated delivery ratios. In-stream loss was based on a first-order exponential decay function of river travel time to Jordan Lake, with the decay coefficient generated by estimates of residence time in the river network, and upstream/downstream nutrient loads following non-linear regression methods used in SPARROW (Alexander et al., 2000). Further adjustments based on im- poundment trapping of sediment and associated nutrient loads were carried out for larger reservoirs in the Haw. The results provided estimates of both loading and transport efficiency to Jordan Lake, with estimates of relative effectiveness of sectoral loading reductions in different parts of the watershed. FIGURE 4-17 14 digit HUCs draining to Jordan Lake in the Haw River watershed of North Carolina. SOURCE: Tetra Tech (2003).

314 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES P8 (Program for Predicting Polluting Particle Passage through Pits, Puddles, and Ponds) is a curve number-based model for predicting the generation and transport of stormwater runoff pollutants in urban watersheds, originally devel- oped to help design and evaluate nutrient control in wet detention ponds (Palm- strom and Walker, 1990; http://wwwalker.net/p8/). Continuous water-balance and mass-balance calculations are performed and consist of the following ele- ments: watersheds, devices, particle classes, and water quality components. Continuous simulations use hourly rainfall and daily air temperature time series. The model was initially calibrated to predict runoff quality typical of that meas- ured under NURP (EPA, 1983). SCMs in P8 include detention ponds (wet, dry, extended), infiltration basins, swales, and buffer strips. Groundwater and base- flows are also included in the model using linear reservoir processes. MUSIC is a part of the Catchment Modelling Toolkit (www.toolkit.net.au) developed by the Cooperative Research Center for Catchment Hydrology in Australia (Wong et al., 2001). The model concentrates on the quality and quan- tity of urban stormwater, including detailed accounting of multiple SCMs acting within a treatment train and life-cycle costing. It employs a simplified rainfall– runoff model (Chiew and McMahon, 1997) based on impervious area and two moisture stores (shallow and deep). TSS, total nitrogen, and total phosphorus are based on EMCs, sampled from lognormal distributions. The model does not contain detailed hydraulics required for routing or sizing of SCMs, and it is de- signed as a planning tool. EPA’s SWMM has the capability of simulating water quantity and quality for a single storm event or for continuous runoff. The model is commonly used to design and evaluate storm, sanitary, and combined sewer systems. SWMM accounts for hydrologic processes that produce runoff from urban areas, including time-varying rainfall, evaporation, snow accumulation and melting, depression storage, infiltration into soil, percolation to groundwater, interflow between groundwater and the drainage system, and nonlinear reservoir routing of overland flow. Spatial variability is modeled by dividing a study area into a collection of smaller, homogeneous subcatchment areas, each containing its own fraction of pervious and impervious sub-areas. Overland flow can be routed between sub-areas, between subcatchments, or between entry points of a drainage system. SWMM can also be used to estimate the production of pollutant loads associated with runoff for a number of user-defined water quality constituents. Transport processes include dry-weather pollutant buildup over different land uses, pollutant wash-off from specific land uses, direct contribution of rainfall deposition, and the action of such SCMs as street cleaning, source control, and treatment in storage units, among others. Watershed models such as SWAT (Arnold et al., 1998) or HSPF (Bicknell et al., 1997, 2005) have components based on similar land-use runoff and load- ing factors, but also incorporate options to utilize detailed descriptions of inter- ception, infiltration, runoff, routing, and biogeochemical transformations. Both models are based on hydrologic models that were developed prior to the avail- ability of detailed digital spatial information on watershed form and use concep-

MONITORING AND MODELING 315 tual control volumes that are not spatially linked. HRUs are based on land use, soils, and vegetation (and crop) type, among other characteristics, and are con- sidered uniformly distributed through a subbasin. Within each HRU, simplified representations of soil upper and lower zones, or unsaturated and saturated com- ponents, are vertically integrated with a conceptual groundwater storage-release component. There is no land surface routing and all runoff from a land element is considered to reach the river reach, with some delivery ratio if appropriate for sediment and other constituents. Like GWLF, the models are typically not de- signed to estimate loadings from individual dischargers, but are used to help guide and develop TMDL for watersheds. SWAT and HSPF are integrated within the EPA BASINS system (http://www.epa.gov/waterscience/basins) with GIS tools designed to use available spatial data to set up and parameterize simu- lations for watersheds within the United States. Examples of combining one of these models, typically designed for larger-scale applications (such as the area shown in Figure 4-14) with more site-specific models such as SLAMM or SWMM, are given in Box 4-7. BOX 4-7 Using SWAT and WinSLAMM to Predict Phosphorus Loads in the Rock River Basin, Wisconsin Wisconsin Administrative Code NR 217 states that wastewater treatment facilities in Wisconsin must achieve an effluent concentration of 1 mg/L for phosphorus. Alternative limits are allowed if it can be demonstrated that achieving the 1 mg/L limit will not “result in an environmentally significant improvement in water quality” (NR 217.04(2)(b)1). In re- sponse to NR 217, a group of municipal wastewater treatment facilities formed the Rock River Partnership (RRP) to assess water quality management issues (Kirsch, 2000). The RRP and the Wisconsin Department of Natural Resources funded a study to seek water quality solutions across all media, and not just pursue additional reductions from point sources. A significant portion of the study required a modeling effort to determine the mag- nitude of various nutrient sources and determine potential reductions through the imple- mentation of global SCMs. The Rock River Basin covers approximately 9,530 square kilometers and lies within the glaciated portion of south central and eastern Wisconsin (Figure 4-18). The Rock River and its numerous tributaries thread their way through this landscape that spreads over 10 counties inhabited by more than 750,000 residents. There are 40 permitted municipalities in the watershed, representing 4 percent of the land area, and they are served by 57 sew- age treatment plants. Urban centers include Madison, Janesville, and Beloit as well as smaller cities such as Waupun, Watertown, Oconomowoc, Jefferson, and Beaver Dam. Although the basin is experiencing rapid growth, it is still largely rural in character with agri- culture using nearly 75 percent of the land area. Crops range from continuous corn and corn–soybean rotations in the south to a mix of dairy, feeder operations, and cash cropping in the north. The basin enjoys a healthy economy with a good balance of agricultural, in- dustrial, and service businesses. continues next page

316 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES BOX 4-7 Continued The focus of the modeling was to construct an intermediate-level macroscale model to better quantify phosphorus loads from point and nonpoint sources throughout the basin. The three goals of the modeling effort were to (1) estimate the average annual phosphorus load, (2) estimate the relative contribution of phosphorus loads from both nonpoint (urban and agricultural) and point sources, and (3) estimate changes in average annual phospho- rus loads from the application of global SCMs and point source controls. SWAT was selected for the agricultural analysis and WinSLAMM was selected to de- velop phosphorus loads for the urban areas. WinSLAMM was selected to make estimates of stormwater loads, because it is already calibrated in Wisconsin for stormwater volumes and pollutant concentrations. Outputs of phosphorus loads from WinSLAMM were used as input to SWAT. One output of SWAT was a total nonpoint phosphorus load based on agri- cultural loads calculated in SWAT and stormwater loads estimated by WinSLAMM. SWAT was calibrated with data from 23 USGS gauging stations in the Rock River Ba- sin. Hydrology was balanced first on a yearly basis looking at average annual totals, then monthly to verify snowfall and snowmelt routines, and then daily. Daily calibration was conducted to check crop growth, evapotranspiration, and daily peak flows. Crop yields predicted by SWAT were calibrated to those published in the USDA Agricultural Statistics. Under current land-use and management conditions, the model predicted an average annual load of approximately 1,680,000 pounds of total phosphorus for the basin with 41 percent from point sources and 59 percent from nonpoint sources. Less than 10 percent of the annual phosphorus load is generated by the urban areas in the watershed. Evaluation of various SCM scenarios shows that with implementation of NR 217 (applicable point source effluent at 1 mg/L) and improvement in tillage practices and nutrient management practices, total phosphorus can be reduced across the basin by approximately 40 percent. It is important to note that the nonpoint management practices that were analyzed were limited to two options: modifications in tillage practices, and adoption of recommended nutrient application rates. No other management practices (i.e., urban controls, riparian buffer strips, etc.) were simulated. Urban controls were not included because the urban areas contributed a relatively small percentage of the total phosphorus load. Thus, load- ings depicted by SWAT under these management scenarios do not necessarily represent the lowest attainable loads. Results suggest that a combination of point and nonpoint con- trols will be required to attain significant phosphorus reductions. The CBWM is a detailed watershed model that is extended from HSPF as a base, but includes additional components to incorporate stormwater controls at the land segment level. HSPF is operated for a number of subbasins, and each subbasin model includes different land segments based on land cover and soil units as aspatial, lumped distribution functions, but also includes representation of SCMs and (large) stream routing. Model implementation at the scale of the full Chesapeake Bay watershed requires fairly coarse-grained land partitioning. A threshold of 100 cfs mean annual flow is used to represent streams and rivers, and the one-to-one mapping of land segment to river reach produces large, het- erogeneous land segments as the basic runoff-producing zones. SCMs are im- plemented either at the field or runoff production unit as distinct land segment types in terms of management or land cover, or as “edge-of-field” reductions of runoff or pollutant loads. The latter are assigned as static efficiency factors irre- spective of flow conditions or season, with all SCMs within a land segment in- tegrated into a single weighted efficiency value.

MONITORING AND MODELING 317 FIGURE 4-18 Rock River Basin, Wisconsin. SOURCE: Reprinted, with permission, from Kirsch (2000). Copyright 2000 by American Society for Biological and Agricultural Engi- neers. SLAMM is designed for complex, urban catchments and is used as a plan- ning tool to assess both stormwater and pollutant runoff production and the ca- pability of specific stormwater control strategies to reduce stormwater dis- charges from urban sources. It is specifically designed to capture the most sig- nificant distributed and sequential drainage effects of variable source areas in urban catchments (Pitt and Vorhees, 2002) and is based on detailed descriptions of the catchment composition, including both type and relative position (drain- age sequence) of land elements. The model is dependent on high-resolution classification or description of the catchment that has become increasingly available in urban areas over the past two decades, and comprehensive field as- sessment of runoff and pollutant loading from different urban land elements. SLAMM uses continuous simulation for some aspects, such as the build up of street pollutant loads between storms, while using event-based simulation for runoff. The description of build-up and wash-off is a critical component in ur- ban stormwater models applied to areas with substantial impervious surfaces and is a good example of the need to match detailed and rigorous field sampling in

318 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES order to adequately describe and represent dominant processes. Details of measurement and model representation for build-up and wash-off of contami- nants are given in Box 4-8. Potential New Applications of Coupled Distributed Models The advent of high-resolution digital topographic and land-cover data over the past two decades has fueled a significant shift in runoff modeling towards “spatially explicit” simulations that distinguish and connect runoff producing elements in a detailed flow routing network. While models developed prior to the availability of high-resolution data or based on older paradigms developed in the absence of this information required spatial and conceptual lumping of con- trol volumes, more recently developed distributed models may contain control volumes linked in multiple vertical layers (soil and aquifer elements) and later- ally from a drainage divide to the stream, including stream-channel and riparian segments. A set of models has been developed and applied to stormwater gen- eration using this paradigm that can be applied at the scale of residential neighborhoods, resolving land cover and topography at the parcel level. These models also vary in terms of their emphasis, with some models better represent- ing coupled surface water–groundwater interactions, water, carbon and nutrient cycling, or land–atmosphere interactions. Boyer et al. (2006) have recently re- viewed a set of hydrologic and ecosystem models in terms of their ability to simulate sources, transport, and transformation of nitrogen within terrestrial and aquatic ecosystems. Data and information requirements are typically high, and the level of process specificity may outstrip the available information necessary to parameterize the integrated models. However, an emphasis is placed on pro- viding mechanistic linkage and feedbacks between important surface, subsur- face, atmospheric, and ecosystem components. Examples of these models in- clude the Distributed Hydrology Soil Vegetation model (DHSVM, Wigmosta et al., 1994); the Regional Hydro-Ecologic Simulation System (RHESSys, Band et al., 1993; Tague and Band, 2004); ParFlow-Common Land Model (CLM, Max- well and Miller, 2007); the Penn State Integrated Hydrologic Model (PIHM, Qu and Duffy, 2007); the Soil Moisture Distribution and Routing (SMDR) model (Easton et al., 2007); and that of Xiao et al. (2007). One advantage of integrating surface and subsurface flow systems within any of these model structures is the ability to incorporate different SCMs by specifying characteristics of specific locations within the flow element networks linked to the subsurface drainage. Examples can include alteration of surface detention storage and release curves to simulate detention ponds, or soil depth, texture, vegetation, and drainage release for rainfall gardens. The advantage of this approach is the tight coupling of these SCM features with the connected surface and subsurface drainage systems, allowing the direct incorporation of the SCM as sink or source terms within the flowpath network. Burgess et al. (1998) effectively demonstrated that suburban lawns can become the major

MONITORING AND MODELING 319 BOX 4-8 Build-up and Wash-off of Contaminants from Impervious Surfaces The accumulation and wash-off of street particulates have been studied for many years (Sartor and Boyd, 1972; Pitt, 1979, 1985, 1987) and are important considerations in many stormwater models, such as SWMM, HSPF, and SLAMM, that require information pertaining to the movement of pollutants over land surfaces. Accumulation rates are usu- ally obtained through trial and error during calibration, with little, if any, actual direct meas- urements. Furthermore, those direct measurements that have been made are often mis- applied in modeling applications, resulting in unreasonable model predictions. Historically, streets have been considered the most important directly connected im- pervious surface. Therefore, much early research was directed toward measuring the processes on these surfaces. Although it was eventually realized that other surfaces can also be significant pollutant sources (see Pitt et al., 2005a,b for reviews), additional re- search to study accumulation and wash-off for these other areas has not been conducted, such that the following discussion is focused on street dirt accumulation and wash-off. Accumulation of Particulates on Street Surfaces The permanent storage component of street surface particulates is a function of street texture and condition and is the quantity of street dust and dirt that cannot be removed naturally by rain or wind, or by street cleaning equipment. It is literally trapped in the tex- ture of the street. The street dirt loading at any time is this initial permanent loading plus the accumulation amount corresponding to the exposure period, minus the resuspended material removal by wind and traffic-induced turbulence. One of the first research studies to attempt to measure street dirt accumulation was conducted by Sartor and Boyd (1972). Field investigations were conducted between 1969 and 1971 in several cities throughout the United States and in residential, commercial, and industrial land-use areas. Figure 4-19 is a plot of the 26 test area measurements collected from different cities, but separated by the three land uses. The data are the accumulated solids loading plotted against the number of days since the street had been cleaned by the municipal street cleaning operation or a “significant” rain. There is a large amount of vari- ability. The street cleaning and this rain were both assumed to remove all of the street dirt; hence, the curves were all forced through zero loading at zero days. A more thorough study was conducted in San Jose, California by Pitt (1979), during which the measured street dirt loading for a smooth street was also found to be a function of time. As shown in Figure 4-20, both accumulation rates and increases in particle size of the street dirt increase as time between street cleaning lengthens. However, it is also evi- dent that there is a substantial residual loading on the streets immediately after the street cleaning, which differs substantially from the assumption of Sartor and Boyd that rains re- duce street dirt to zero. The San Jose study also investigated the role of different street textures, which re- sulted in very different street dirt loadings. Although the accumulation and deposition rates are quite similar, the initial loading values (the permanent storage values) are very different, with greater amounts of street dirt trapped by the coarser (oil and screens) pavement. Street cleaning and rains are not able to remove this residual material. The early, uncor- rected Sartor and Boyd accumulation rates that ignored the initial loading values were al- most ten times the corrected values that had reasonable “initial loads.” continues next page

320 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES BOX 4-8 Continued FIGURE 4-19 Accumulation curves developed during early street cleaning research. SOURCE: Sartor and Boyd (1972). FIGURE 4-20 Street dirt accumulation and particle size changes on good asphalt streets in San Jose, California. SOURCE: Pitt (1979).

MONITORING AND MODELING 321 Finally, it was found that, at very long accumulation periods relative to the rain fre- quency, the wind losses (fugitive dust) may approximate the deposition rate, resulting in very little increases in loading. In Bellevue, Washington, with inter-event rain periods aver- aging about three days, steady loadings were observed after about one week (Pitt, 1985). However, in Castro Valley, California, the rain inter-event periods were much longer (rang- ing from about 20 to 100 days), and steady loadings were never observed (Pitt and Shaw- ley, 1982). Taking many studies into account (Sartor and Boyd 1972—corrected; Pitt, 1979, 1983, 1985; Pitt and Shawley, 1982; Pitt and Sutherland, 1982; Pitt and McLean, 1986), the most important factors affecting the initial loading and maximum loading values have been found to be street texture and street condition, and not land use. When data from many locations are studied, it is apparent that smooth streets have substantially less loadings at any ac- cumulation period compared to rough streets for the same land use. Very long accumula- tion periods relative to the rain frequency result in high street dirt loadings. However, dur- ing these conditions the wind losses of street dirt (as fugitive dust) may approximate the deposition rate, resulting in relatively constant street dirt loadings. Wash-off of Street Surface Pollutants Wash-off of particulates from impervious surfaces is dependent on the available sup- ply of particulates on the surface that can be removed by rains, the rain energy available to loosen the material, and the capacity of the runoff to transport the loosened material. Ob- servations of particulate wash-off during controlled tests have resulted in empirical wash-off models. The earliest controlled street dirt wash-off experiments were conducted by Sartor and Boyd (1972) to estimate the percentage of the available particulates on the streets that would wash off during rains of different magnitudes. Sartor and Boyd fitted their data to an exponential curve, as shown in Figure 4-21 (accumulative wash-off curves for several parti- cle sizes). The empirical equation that they developed, N = No e-kR, is only sensitive to the total rain depth up to the time of interest and the initial street dirt loading. FIGURE 4-21 Street dirt wash-off during high-intensity rain tests. SOURCE: Sartor and Boyd (1972). continues next page

322 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES BOX 4-8 Continued There are several problems with this approach. First, these figures did not show the total street dirt loading that was present before the wash-off tests. Most modelers have assumed that the asymptotic maximum shown was the total “before-rain” street dirt loading; that is, the No factor has been assumed to be the total initial street loading, when in fact it is only the portion of the total street load available for wash-off (the maximum asymptotic wash-off load observed during the wash-off tests). The actual total street dirt loadings were several times greater than the maximum wash-off amounts observed. STORM and SWMM now use an availability factor (A) for particulate residue as a calibration procedure in order to reduce the wash-off quantity for different rain intensities (Novotny and Chesters, 1981). Second, the proportionality constant, k, was found by Sartor and Boyd to be slightly de- pendent on street texture and condition, but was independent of rain intensity and particle size. The value of this constant is usually taken as 0.18/mm, assuming that 90 percent of the particulates will be washed from a paved surface in one hour during a 13 mm/h rain. However, Alley (1981) fitted this model to watershed outfall runoff data and found that the constant varied for different storms and pollutants for a single study area. Novotny exam- ined “before” and “after” rain-event street particulate loading data using the Milwaukee NURP stormwater data (Bannerman et al., 1983) and found almost a three-fold difference between the proportionality constant value for fine (<45 m) and medium-sized particles (100 to 250 m). Jewell et al. (1980) also found large variations in outfall “fitted” values for different rains compared to the typical default value. They stressed the need to have local calibration data before using the exponential wash-off equation, as the default values can be very misleading. The exponential wash-off equation for impervious areas is justified, but wash-off coefficients for each pollutant would improve its accuracy. The current SWMM5 version discourages the use of accumulation and wash-off functions due to lack of data, and the misinterpretation of available data. It turns out that particle dislodgement and transport characteristics at impervious areas can be directly measured using relatively simple wash-off tests. The Bellevue, Washington, urban runoff project (Pitt, 1985) included about 50 pairs of street dirt loading observations close to the beginnings and ends of rains to determine the differences in loadings that may have been caused by the rains. The observations were affected by rains falling directly on the streets, along with flows and particulates originating from non-street areas. When all the data were considered together, the net loading difference was about 10 to 13 g/curb-m removed, which amounted to a street dirt load reduction of about 15 percent. Large reduc- tions in street dirt loadings for the small particles were observed during these Bellevue rains. Most of the weight of solid material in the runoff was concentrated in fine particle sizes (<63 m). Very few wash-off particles greater than 1,000 m were found; in fact, street dirt loadings increased for the largest sizes, presumably due to settled erosion mate- rials. Urban runoff outfall particle size analyses in Bellevue (Pitt, 1985) resulted in a me- dian particle size of about 50 m; similar results were obtained in the Milwaukee NURP study (Bannerman et al., 1983). The results make sense because the rain energy needed to remove larger particles is much greater than for small particles. In order to clarify street dirt wash-off, Pitt (1987) conducted numerous controlled wash-off tests on city streets in Toronto. The experimental factors examined included rain intensity, street texture, and street dirt loading. The differences between available and total street dirt loads were also related to the experimental factors. The runoff flow quantities were also carefully monitored to determine the magnitude of initial and total rain water losses on impervious surfaces. The test setup was designed and tested to best represent actual rainfall conditions, such as rain intensities (3 mm/h) and peak rain intensities (12 mm/h). The kinetic energies of the “rains” during these tests were therefore comparable to actual rains under investigation. Figure 4-22 shows the asymptotic wash-off values ob- served in the tests, along with the measured total street dirt loadings. The maximum as- ymptotic values are the “available” street dirt loadings (No). As can be seen, the measured

MONITORING AND MODELING 323 FIGURE 4-22 Wash-off plots for high rain intensity, dirty street, and smooth street test, showing the total street dirt loading. SOURCE: Pitt (1987). total loadings are several times larger than these “available” loading values. For example, the asymptotic available total solids value for the high-intensity rain–dirty street–smooth 2 street test was about 3 g/m while the total load on the street for this test was about 14 g/ 2 m , or about five times the available load. The differences between available and total loadings for the other tests were even greater, with the total loads typically about ten times greater than the available loads. The total loading and available loading values for dis- solved solids were quite close, indicating almost complete wash-off of the very small parti- cles. The availability factor (the ratio of the available loading, N0, to the total loading) de- pended on the rain intensity and the street roughness, such that wash-off was more effi- cient for the higher rain energy and smoother pavement tests. The worst case was for a low rain intensity and rough street, where only about 4.5 percent of the street dirt would be washed from the pavement. In contrast, the high rain intensities on the smooth streets were more than four times more efficient in removing street dirt (20 percent removal). A final important consideration in calculating wash-off of street dirt during rains is the carrying capacity of the flowing water to transport sediment. If the calculated wash-off is greater than the carrying capacity (such as would occur for relatively heavy street dirt loads and low to moderate rain intensities), then the carrying capacity is limiting. For high rain intensities, the carrying capacity is likely sufficient to transport most or all of the wash-off 2 material. Figure 4-23 shows the maximum wash-off amounts (g/m ) for the different tests conducted on smooth streets plotted against the rain intensity (mm/h) used for the tests (data from Sartor and Boyd, 1972, and Pitt, 1987). Wash-off limitations for rough streets would be more restrictive. continues next page

324 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES BOX 4-8 Continued FIGURE 4-23 Maximum wash-off capacity for smooth streets (based on measurements of Sartor and Boyd, 1972; Pitt, 1987). If the predicted wash-off, using the previous “standard” wash-off equations, is smaller than the values shown in this figure, then those values can be used directly. However, if the predicted wash-off is greater than the values shown in this figure, then the values in the figure should be used. Accumulation and Wash-off Summary This discussion summarized street particulate wash-off observations obtained during special wash-off tests, along with associated street dirt accumulation measurements. The objectives of these tests were to identify the significant rain and street factors affecting particulate wash-off and to develop appropriate wash-off models. The controlled wash-off experiments identified important relationships between “available” and “total” particulate loadings and the significant effects of the test variables on the wash-off model parameters. Past modeling efforts have typically ignored or misused this relationship to inaccurately predict the importance of street particulate wash-off. The available loadings were almost completely washed off streets during rains of about 25 mm (as previously assumed). How- ever, the fraction of the total loading that was available was at most only 20 percent of the total loading, and averaged only 10 percent, with resultant actual wash-offs of only about 9 percent of the total loadings. In many model applications, total initial loading values (as usually measured during field studies) are used in conjunction with model parameters as the available loadings, resulting in predicted wash-off values that are many times larger than observed. This has the effect of incorrectly assuming greater pollutant contributions originating from streets and less from other areas during rains. This in turn results in inaccurate estimates of the effec- tiveness of different source area urban runoff controls. Although streets can be important sources of runoff and stormwater pollutants, their significance varies greatly depending on the land use and rainfall pattern. They are much more important sources in areas having relatively mild rains (e.g., the Pacific Northwest), where contaminants from other potential sources are not effectively transported to the storm drainage system.

MONITORING AND MODELING 325 source of stormwater in seasonally wet conditions (Seattle), while Cuo et al. (2008) have explored the modification of DHSVM to include detention SCMs. Xiao et al. (2007) explicitly integrated and evaluated parcel scale SCM design and efficiency into their model. Wang et al. (2008) integrated a canopy inter- ception model with a semi-distributed subsurface moisture scheme (TOP- MODEL) to evaluate the effectiveness of urban tree canopy interception on stormwater production, utilizing a detailed spatial dataset of urban tree cover. Band et al. (2001) and Law (2003) coupled a water-, carbon-, and nitrogen- cycling model to a distributed water routing system modified from DHSVM to simulate nitrogen cycling and export in a high-spatial-resolution representation of forested and suburban catchments. While these models have the potential to directly link stormwater generation with specific dischargers, the challenge of scaling to larger watersheds remains. SMDR (Easton et al., 2007) has recently been used to integrate rural and urban stormwater production, including dis- solved phosphorus source and transport in New York State. Alternatives to mass budget-based models include fully statistical ap- proaches such as simple regressions based on watershed land use and population (e.g., Boyer et al., 2002); nonlinear regression using detailed watershed spatial data and observed loads to estimate retention parameters and loading of nutri- ents, sediment, and other pollutants (e.g., Smith et al., 1997; Brakebill and Pre- ston, 1999; Schwarz et al., 2006); and Bayesian chain models (e.g., Reckhow and Chapra, 1999; Borsuk et al., 2001). These models have the advantage of being data-based, and therefore capable of assimilating observations as they become available to update water quality probabilities, but also lack a process basis that might support management intervention. A major debate exists within the literature as to the relative advantages of detailed process-based models that may not have inadequate information for parameterization, and the more empiri- cal, data-based approaches. Limitations in Extending Stormwater Models to Biological Impacts The mass budget approach may be successful in developing the physical and chemical characteristics of the receiving waterbody in terms of the flow (or stage) duration curve, the distribution of concentrations over time, and the inte- grated pollutant storage and flux (load) terms. However, the biological status of the waterbody requires a link between the physical and chemical conditions, primary productivity, and trophic system interactions. Progressing from aquatic ecosystem productivity to trophic systems includes increasingly complex eco- logical processes such as competition, herbivory, predation, and migration. To date, mechanistic linkage between flow path hydraulics, biogeochemistry, and the ecological structure of the aquatic environment has not been developed. Instead, habitat suitability for different communities is identified through em- pirical sampling and analysis, with the implicit assumption that, as relative

TABLE 4-7 Example Mathematical Models That Have Been or Can Be Used in Stormwater Modeling 326 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES

MONITORING AND MODELING 327 Note: CN, curve number

328 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES habitat suitability changes, transitions will occur between species or assem- blages. These methods may work well at the base of the trophic system (algae, phytoplankton) and for specific conditions such as DO limitations on fish com- munities, but the impacts of low to moderate concentrations of pollutants on aquatic ecosystems may still be poorly understood. A critical assumption in these and similar models (e.g., ecological community change resulting from physical changes to the watershed or climate) is the substitution of space for time. More detailed understanding of the mechanisms leading to a shift in eco- logical communities and interactions with the physical environment is necessary to develop models of transient change, stability of the shifts, and feedback to the biophysical environment. Given these limitations, it should be noted that statistical databases on spe- cies tolerance to a range of aquatic conditions have been compiled that will al- low the development of habitat suitability mapping as a mechanism for (1) tar- geting ecosystem restoration, (2) determining vulnerable sites (for use in appli- cation of the Endangered Species Act), and (3) assessing aquatic ecosystem im- pairment and “best use” relative to reference sites. *** Stormwater models have been developed to meet a range of objectives, in- cluding small-scale hydraulic design (e.g., siting and sizing a detention pond), estimation of potential contributions of stormwater pollutants from different land covers and locations using empirically generated EMC, and large water- shed hydrology and gross pollutant loading. The ability to associate a given discharger with a particular waterbody impairment is limited by the scale and complexity of watersheds (i.e., there maybe multiple discharge interactions); by the ability of a model to accurately reproduce the distribution function of dis- charge events and their cumulative impacts (as opposed to focusing only on de- sign storms of specific return periods); and by the availability of monitoring data of sufficient number and design to characterize basic processes (e.g., build- up/wash-off), to parameterize the models, and to validate model predictions. In smaller urban catchments with few dominant dischargers and significant impervious area, current modeling capabilities may be sufficient to associate the cumulative impact of discharge to waterbody impairment. However, many im- paired waterbodies have larger, more heterogeneous stormwater sources, with impacts that are complex functions of current and past conditions. The level of sampling that would be necessary to support linked model calibration and verifi- cation using current measurement technologies is both time-consuming and ex- pensive. In order to develop a more consistent capability to support stormwater permitting needs, there should be increased investment in improving model paradigms, especially the practice and methods of model linkage as described above, and in stormwater monitoring. The latter may require investment in a new generation of sensors that can sample at temporal resolutions that can adjust to characterize low flow and the dynamics of storm flow, but are sufficiently

MONITORING AND MODELING 329 inexpensive and autonomous to be deployed in multiple locations from distrib- uted sources to receiving waterbodies of interest. Finally, as urban areas extend to encompass progressively lower-density development, the interactions of sur- face water and groundwater become more critical to the cumulative impact of stormwater on impaired waterbodies. EPA needs to ensure continuous support and development of their water quality models and spatial data infrastructure. Beyond this, a set of distributed watershed models has been developed that can resolve the location and position of parcels within hydrologic flow fields; these are being modified for use as ur- ban stormwater models. These models avoid the pitfalls of lumping, but they require much greater volumes of spatial data, provided by current remote sens- ing technology (e.g., lidar, airborne digital optical and infrared sensors) as well as the emerging set of in-stream sensor systems. While these methods are not yet operational or widespread, they should be further investigated and tested for their capabilities to support stormwater management. CONCLUSIONS AND RECOMMENDATIONS This chapter addresses what might be the two weakest areas of the storm- water program—monitoring and modeling of stormwater. The MS4 and par- ticularly the industrial stormwater monitoring programs suffer from (1) a paucity of data, (2) inconsistent sampling techniques, (3) a lack of analyses of available data and guidance on how permittees should be using the data to improve stormwater management decisions, and (4) requirements that are difficult to relate to the compliance of individual dischargers. The current state of stormwa- ter modeling is similarly limited. Stormwater modeling has not evolved enough to consistently say whether a particular discharger can be linked to a specific waterbody impairment, although there are many correlative studies showing how parameters co-vary in important but complex and poorly understood ways (see Chapter 3). Some quantitative predictions can be made, particularly those that are based on well-supported causal relationships of a variable that responds to changes in a relatively simple driver (e.g., modeling how a runoff hydrograph or pollutant loading change in response to increased impervious land cover). However, in almost all cases, the uncertainty in the modeling and the data, the scale of the problems, and the presence of multiple stressors in a watershed make it difficult to assign to any given source a specific contribution to water quality impairment. More detailed conclusions and recommendations about monitoring and modeling are given below. Because of a ten-year effort to collect and analyze monitoring data from MS4s nationwide, the quality of stormwater from urbanized areas is well characterized. These results come from many thousands of storm events, systematically compiled and widely accessible; they form a robust dataset of utility to theoreticians and practitioners alike. These data make it possible to

330 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES accurately estimate the EMC of many pollutants. Additional data are available from other stormwater permit holders that were not originally included in the database and from ongoing projects, and these should be acquired to augment the database and improve its value in stormwater management decision-making. Industry should monitor the quality of stormwater discharges from certain critical industrial sectors in a more sophisticated manner, so that permitting authorities can better establish benchmarks and technology- based effluent guidelines. Many of the benchmark monitoring requirements and effluent guidelines for certain industrial subsectors are based on inaccurate and old information. Furthermore, there has been no nationwide compilation and analysis of industrial benchmark data, as has occurred for MS4 monitoring data, to better understand typical stormwater concentrations of pollutants from various industries. The absence of accurate benchmarks and effluent guidelines for critical industrial sectors discharging stormwater may explain the lack of enforcement by permitting authorities, as compared to the vigorous enforcement within the wastewater discharge program. Industrial monitoring should be targeted to those sites having the greatest risk associated with their stormwater discharges. Many industrial sites have no or limited exposure to runoff and should not be required to under- take extensive monitoring. Visual inspections should be made, and basic con- trols should be implemented at these areas. Medium-risk industrial sites should conduct monitoring so that a sufficient number of storms are measured over the life of the permit for comparison to regional benchmarks. Again, visual inspec- tions and basic controls are needed for these sites, along with specialized con- trols to minimize discharges of the critical pollutants. Stormwater from high- risk industrial sites needs to be continuously monitored, similar to current point source monitoring practices. The use of a regionally calibrated stormwater model and random monitoring of the lower-risk areas will likely require addi- tional monitoring. Continuous, flow-weighted sampling methods should replace the tradi- tional collection of stormwater data using grab samples. Data obtained from too few grab samples are highly variable, particularly for industrial monitoring programs, and subject to greater uncertainly because of experimenter error and poor data-collection practices. In order to use stormwater data for decision mak- ing in a scientifically defensible fashion, grab sampling should be abandoned as a credible stormwater sampling approach for virtually all applications. It should be replaced by more accurate and frequent continuous sampling methods that are flow weighted. Flow-weighted composite monitoring should continue for the duration of the rain event. Emerging sensor systems that provide high tem- poral resolution and real-time estimates for specific pollutants should be further investigated, with the aim of providing lower costs and more extensive monitor- ing systems to sample both streamflow and constituent loads.

MONITORING AND MODELING 331 Flow monitoring and on-site rainfall monitoring need to be included as part of stormwater characterization monitoring. The additional information associated with flow and rainfall data greatly enhance the usefulness of the much more expensive water quality monitoring. Flow monitoring should also be correctly conducted, with adequate verification and correct base-flow sub- traction methods applied. Using regional rainfall data from locations distant from the monitoring location is likely to be a major source of error when rainfall factors are being investigated. The measurement, quality assurance, and main- tenance of long-term precipitation records are both vital and nontrivial to stormwater management. Whether a first flush of contaminants occurs at the start of a rainfall event depends on the intensity of rainfall, the land use, and the specific pol- lutant. First flushes are more common for smaller sites with greater impervi- ousness and thus tend to be associated with more intense land uses such as commercial areas. Even though a site may have a first flush of a constituent of concern, it is still important that any SCM be designed to treat as much of the runoff from the site as possible. In many situations, elevated discharges may occur later in an event associated with delayed periods of peak rainfall intensity. Stormwater runoff in arid and semi-arid climates demonstrates a seasonal first-flush effect (i.e., the dirtiest storms are the first storms of the season). In these cases, it is important that SCMs are able to adequately handle these flows. As an example, early spring rains mixed with snowmelt may occur during peri- ods when wet detention ponds are still frozen, hindering their performance. The first fall rains in the southwestern regions of the United States may occur after extended periods of dry weather. Some SCMs, such as street cleaning targeting leaf removal, may be more effective before these rains than at other times of the year. Watershed models are useful tools for predicting downstream impacts from urbanization and designing mitigation to reduce those impacts, but they are incomplete in scope and typically do not offer definitive causal links between polluted discharges and downstream degradation. Every model simulates only a subset of the multiple interconnections between physi- cal, chemical, and biological processes found in any watershed, and they all use a grossly simplified representation of the true spatial and temporal variability of a watershed. To speak of a “comprehensive watershed model” is thus an oxy- moron, because the science of stormwater is not sufficiently far advanced to determine causality between all sources, resulting stressors, and their physical, chemical, and biological responses. Thus, it is not yet possible to create a proto- col that mechanistically links stormwater dischargers to the quality of receiving waters. The utility of models with more modest goals, however, can still be high—as long as the questions being addressed by the model are in fact relevant and important to the functioning of the watershed to which that model is being applied, and sufficient data are available to calibrate the model for the processes

332 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES included therein. EPA needs to ensure that the modeling and monitoring capabilities of the nation are continued and enhanced to avoid losing momentum in un- derstanding and eliminating stormwater pollutant discharges. There is a need to extend, develop, and support current modeling capabilities, emphasizing (1) the impacts of flow energy, sediment transport, contaminated sediment, and acute and chronic toxicity on biological systems in receiving waterbodies; (2) more mechanistic representation (physical, chemical, biological) of SCMs; and (3) coupling between a set of functionally specific models to promote the link- age of source, transport and transformation, and receiving water impacts of stormwater discharges. Stormwater models have typically not incorporated in- teractions with groundwater and have treated infiltration and recharge of groundwater as a loss term with minimal consideration of groundwater contami- nation or transport to receiving waterbodies. Emerging distributed modeling paradigms that simulate interactions of surface and subsurface flowpaths pro- vide promising tools that should be further developed and tested for applications in stormwater analysis. REFERENCES Adams, B., and F. Papa. 2000. Urban Stormwater Management Planning with Analytical Probabilistic Methods. New York: John Wiley and Sons. Alexander, R. B., R. A. Smith, and G. E. Schwarz. 2000. Effect of stream channel size on the delivery of nitrogen to the Gulf of Mexico. Nature 403:758–761. Alley, W. M. 1981. Estimation of impervious-area washoff parameters. Water Resources Research 17(4):1161–1166. Arnold, J. G., R. Srinivasan, R. S. Muttiah, and J. R. Williams. 1998. Large area hydrologic modeling and assessment Part I: Model development. Journal of the American Water Resources Association 34(1):73e89. Band, L. E., P. Patterson, R. R. Nemani, and S. W. Running. 1993. Forest eco- system processes at the watershed scale: 2. Adding hillslope hydrology. Agricultural and Forest Meteorology 63:93–126. Band, L. E., C. L. Tague, P. Groffman, and K. Belt. 2001. Forest ecosystem processes at the watershed scale: hydrological and ecological controls of ni- trogen export. Hydrological Processes 15:2013–2028. Bannerman, R., K. M. Baun, P. E. Bohn, and D. A. Graczyk. 1983. Evaluation of Urban Nonpoint Source Pollution Management in Milwaukee County, Wisconsin. PB 84-114164. Chicago: EPA. Bannerman, R. T., D. W. Owens, R. B. Dodds, and N. J. Hornewer. 1993. Sources of pollutants in Wisconsin stormwater. Water Science and Tech- nology 28(3–5):241–259. Batroney, T. 2008. The Implications of the First Flush Phenomenon on Infiltra-

MONITORING AND MODELING 333 tion BMP Design. Master’s Thesis. Water Resources and Environmental Engineering, Villanova University, Villanova, PA, May. Beven, K. J. 2000. On the future of distributed modelling in hydrology. Hy- drological Processes 14:3183–3184. Beven, K. J. 2001. Dalton Medal Lecture: How far can we go in distributed hydrological modelling? Hydrology and Earth System Sciences 5(1):1–12. Bicknell, B. R., J. C. Imhoff, J. L. Kittle, Jr., A. S. Donigian, Jr., and R. C. Jo- hanson. 1997. Hydrological Simulation Program—Fortran, User's manual for version 11. EPA/600/R-97/080. Athens, GA: EPA National Exposure Research Laboratory. Bicknell, B. R., J. C. Imhoff, J. L. Kittle, Jr., T. H. Jobes, and A. S. Donigian, Jr. 2005. HSPF Version 12.2 User’s Manual. AQUA TERRA Consultants, Mountain View, CA. In cooperation with Office of Surface Water, Water Resources Discipline, U.S. Geological Survey, Reston, VA, and National Exposure Research Laboratory Office of Research and Development, EPA, Athens, GA. Borsuk, M. E., D. Higdon, C. A. Stow, and K. H. Reckhow. 2001. A Bayesian hierarchical model to predict benthic oxygen demand from organic matter loading in estuaries and coastal zones. Ecological Modeling 143:165–181. Boyer, E. W., C. L. Goodale, N. A. Jaworski, and R.W. Howarth. 2002. An- thropogenic nitrogen sources and relationships to riverine nitrogen export in the northeastern USA. Biogeochemistry 57:137–169. Boyer, E. W., R. B. Alexander, W. J. Parton, C. Li, K. Butterbach-Bahl, S. D. Donner, R. W. Skaggs, and S. J. Del Grosso. 2006. Modeling denitrifica- tion in terrestrial and aquatic ecosystems at regional scales. Ecological Ap- plications 16:2123–2142. Brakebill, J. W., and S. E. Preston. 1999. Digital Data Used to Relate Nutrient Inputs to Water Quality in the Chesapeake Bay Watershed, Version 1.0. U.S. Geological Survey Water Open-File Report 99-60. Brown, T., W. Burd, J. Lewis, and G. Chang. 1995. Methods and procedures in stormwater data collection. In: Stormwater NPDES Related Monitoring Needs. H. C. Torno (ed.). Reston, VA: Engineering Foundation and ASCE. Burgess, S. J., M. S. Wigmosta, and J. M. Meena. 1998. Hydrological effects of land-use change in a zero-order catchment. Journal of Hydrological En- gineering ASCE 3:86–97. Burton, G. A., Jr., and R. Pitt. 2002. Stormwater Effects Handbook: A Tool Box for Watershed Managers, Scientists, and Engineers. Boca Raton, FL: CRC Press, 911 pp. Chiew, F. H. S., and T. A. McMahon. 1997. Modelling daily runoff and pollut- ant load from urban catchments. Water (AWWA Journal) 24:16–17. Chow, V. T., D. R. Maidment, and L. W. Mays. 1988. Applied Hydrology. McGraw Hill. Clark, S. E., C. Y. S. Siu, R. E. Pitt, C. D. Roenning, and D. P. Treese. 2008. Peristaltic pump autosamplers for solids measurement in stormwater runoff.

334 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES Water Environment Research 80. doi:10.2175/106143008X325737. Cross, L. M., and L. D. Duke. 2008. Regulating industrial stormwater: state permits, municipal implementation, and a protocol for prioritization. Jour- nal of the American Water Resources Association 44(1):86–106. Cuo, L., D. P. Lettenmaier, B. V. Mattheussen, P. Storck, and M. Wiley. 2008. Hydrologic prediction for urban watersheds with the Distributed Hydrol- ogy-Soil-Vegetation-Model. Hydrological Processes, DOI: 10.1002/hyp.7023 Deletic, A. 1998. The first flush load of urban surface runoff. Water Research 32(8):2462–2470. Easton, Z. M., P. Gérard-Marchant, M. T. Walter, A. M. Petrovic, and T. S. Steenhuis. 2007. Hydrologic assessment of an urban variable source wa- tershed in the northeast United States. Water Resources Research 43:W03413, doi:10.1029/2006WR005076. EPA (U.S. Environmental Protection Agency). 1983. Results of the Nation- wide Urban Runoff Program. PB 84-185552. Washington, DC: Water Planning Division. Gibbons, J., and S. Chakraborti. 2003. Nonparametric Statistical Inference, 4th edition. New York: Marcel Dekker, 645 pp. Haith, D. A., and L. L. Shoemaker. 1987. Generalized watershed loading func- tions for stream flow nutrients. Water Resources Bulletin 23(3):471–478. Horwatich, J., R. Bannerman, and R. Pearson. 2008. Effectiveness of Hydro- dynamic Settling Device and a Stormwater Filtration Device in Milwaukee, Wisconsin. U.S. Geological Survey Investigative Report. Middleton, WI: USGS. Jewell, T. K., D. D. Adrian, and D. W. Hosmer. 1980. Analysis of stormwater pollutant washoff estimation techniques. International Symposium on Ur- ban Storm Runoff. University of Kentucky, Lexington, KY. Kirsch, K. J. 2000. Predicting Sediment and Phosphorus Loads in the Rock River Basin Using SWAT. Presented at the ASAE International Meeting, Paper 002175. July 9–12. St. Joseph, MI: ASAE. Law, N. L. 2003. Sources and Pathways of Nitrate in Urbanizing Watersheds. Ph.D. Dissertation Proposal, Department of Geography, University of North Carolina at Chapel Hill. Lee, H., X. Swamikannu, D. Radulescu, S. Kim, and M. K. Stenstrom. 2007. Design of stormwater monitoring programs. Journal of Water Research, doi:10.1016/j.watres.2007.05.016. Maestre, A., and R. Pitt. 2005. The National Stormwater Quality Database, Version 1.1. A Compilation and Analysis of NPDES Stormwater Monitor- ing Information. Washington, DC: EPA Office of Water. Maestre, A., and R. Pitt. 2006. Identification of significant factors affecting stormwater quality using the National Stormwater Quality Database. Pp. 287–326 In: Stormwater and Urban Water Systems Modeling, Monograph 14. W. James, K. N. Irvine, E. A. McBean, and R. E. Pitt (eds.). Guelph, Ontario: CHI.

MONITORING AND MODELING 335 Maestre, A., R. E. Pitt, and D. Williamson. 2004. Nonparametric statistical tests comparing first flush with composite samples from the NPDES Phase 1 municipal stormwater monitoring data. Stormwater and urban water sys- tems modeling. Pp. 317–338 In: Models and Applications to Urban Water Systems, Vol. 12. W. James (ed.). Guelph, Ontario: CHI. Maestre, A., R. Pitt, S. R. Durrans, and S. Chakraborti. 2005. Stormwater qual- ity descriptions using the three parameter lognormal distribution. Stormwa- ter and urban water systems modeling. In: Models and More for Urban Wa- ter Systems, Monograph 13. W. James, K. N. Irvine, E. A. McBean, and R. E. Pitt (eds.). Guelph, Ontario: CHI. Mahler, B. J., P. C. VanMetre, T. J. Bashara, J. T. Wilson, and D. A. Johns. 2005. Parking lot sealcoat: An unrecognized source of urban polycyclic aromatic hydrocarbons. Environmental Science and Technology 39:5560– 5566. Maxwell, R. H., and M. Miller. 2005. Development of a coupled land surface and groundwater model. Journal Hydrometeorology 6:233–247. McClain, M. E., E. W. Boyer, C. L. Dent, S. E. Gergel, N. B. Grimm, P. M. Groffman, S. C. Hart, J. W. Harvey, C. A. Johnston, E. Mayorga, W. H. McDowell, and G. Pinay. 2003. Biogeochemical hot spots and hot mo- ments at the interface of terrestrial and aquatic ecosystems . Ecosystems 6:301–312. MDE (Maryland Department of the Environment). 2000. 2000 Maryland Stormwater Design Manual, Volumes I & II. Prepared by the Center for Watershed Protection and the Maryland Department of the Environment, Water Management Administration, Baltimore, MD. Novotny, V., and G. Chesters. 1981. Handbook of Nonpoint Pollution Sources and Management. New York: Van Nostrand Reinhold. Oreskes, N., Shrader-Frechette K., Belitz, K. 1994. Verification, Validation, and Confirmation of Numerical Models in the Earth Sciences. Science, New Series, 263(5147):641-646. PaDEP (Pennsylvania Department of Environmental Protection). 2006. Penn- sylvania Stormwater Best Management Practices Manual. Harrisburg, PA: Pennsylvania Department of Environmental Protection. Palmstrom, N., and W. Walker. 1990. The P8 Urban Catchment Model for Evaluating Nonpoint Source Controls at the Local Level. Enhancing States’ Lake Management Programs, USEPA. Pitt, R. 1979. Demonstration of Nonpoint Pollution Abatement Through Im- proved Street Cleaning Practices. EPA-600/2-79-161 Cincinnati, OH: EPA Office of Research and Development. Pitt, R. 1983. Urban Bacteria Sources and Control in the Lower Rideau River Watershed. Ottawa, Ontario: Ontario Ministry of the Environment. Pitt, R. 1985. Characterizing and Controlling Urban Runoff through Street and Sewerage Cleaning. EPA/600/S2-85/038. PB 85-186500. Cincinnati, OH: EPA. Storm and Combined Sewer Program. Risk Reduction Engineering Laboratory.

336 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES Pitt, R. 1986. The incorporation of urban runoff controls in the Wisconsin Pri- ority Watershed Program. Pp. 290–313 In: Advanced Topics in Urban Runoff Research. B. Urbonas and L. A. Roesner (eds.). New York: Engi- neering Foundation and ASCE. Pitt, R. 1987. Small Storm Urban Flow and Particulate Washoff Contributions to Outfall Discharges. Ph.D. Dissertation. Department of Civil and Envi- ronmental Engineering, University of Wisconsin–Madison. Pitt, R., and J. McLean. 1986. Toronto Area Watershed Management Strategy Study. Humber River Pilot Watershed Project. Toronto, Ontario: Ontario Ministry of the Environment. Pitt, R., and G. Shawley. 1982. A Demonstration of Nonpoint Source Pollution Management on Castro Valley Creek. Alameda County Flood Control and Water Conservation District (Hayward, CA) for the Nationwide Urban Runoff Program. Washington, DC: EPA, Water Planning Division. Pitt, R., and R. Sutherland. 1982. Washoe County Urban Stormwater Manage- ment Program. Reno, NV: Washoe Council of Governments. Pitt, R., and J. Voorhees. 1995. Source loading and management model (SLAMM). Pp. 225–243 In: Seminar Publication: National Conference on Urban Runoff Management: Enhancing Urban Watershed Management at the Local, County, and State Levels. March 30–April 2, 1993. EPA/625/R- 95/003. Cincinnati, OH: EPA Center for Environmental Research Informa- tion. Pitt, R., and J. Voorhees. 2002. SLAMM, the Source Loading and Management Model. Pp. 103–139 In: Wet-Weather Flow in the Urban Watershed. R. Field and D. Sullivan (eds.). Boca Raton, FL: CRC Press. Pitt, R., A. Maestre, H. Hyche, and N. Togawa. 2008. The updated National Stormwater Quality Database (NSQD), Version 3. Conference CD. 2008 Water Environment Federation Technical Exposition and Conference, Chi- cago, IL. Pitt, R. S., Clark, J. Lantrip, and J. Day. 1998. Telecommunication Manhole Water and Sediment Study; Vol. 1: Evaluation of Field Test Kits, 483 pp.; Vol. 2: Water and Sediment Characteristics, 1290 pp.; Vol. 3: Discharge Evaluation Report, 218 pp.; Vol. 4: Treatment of Pumped Water, 104 pp. Special Report SR-3841. Morriston, NJ: Bellcore, Inc. Pitt, R., A. Maestre, and R. Morquecho. 2003. Evaluation of NPDES Phase I municipal stormwater monitoring data. In: National Conference on Urban Stormwater: Enhancing the Programs at the Local Level. EPA/625/R- 03/003. Pitt, R. E., A. Maestre, R. Morquecho, and D. Williamson. 2004. Collection and examination of a municipal separate storm sewer system database. Stormwater and Urban Water Systems Modeling. Pp. 257–294 In: Models and Applications to Urban Water Systems, Vol. 12. W. James (eds.). Guelph, Ontario: CHI. Pitt, R., R. Bannerman, S. Clark, and D. Williamson. 2005a. Sources of pollut- ants in urban areas (Part 1)—Older monitoring projects. Pp. 465–484 and

MONITORING AND MODELING 337 507–530 In: Effective Modeling of Urban Water Systems, Monograph 13. W. James, K. N. Irvine, E. A. McBean, and R. E. Pitt (eds.). Guelph, On- tario: CHI. Pitt, R., R. Bannerman, S. Clark, and D. Williamson. 2005b. Sources of pollut- ants in urban areas (Part 2)—Recent sheetflow monitoring results. Pp. 485– 530 In: Effective Modeling of Urban Water Systems, Monograph 13. W. James, K. N. Irvine, E. A. McBean, and R. E. Pitt (eds.). Guelph, Ontario: CHI. Qu, Y., and C. J. Duffy. 2007. A semidiscrete finite volume formulation for multiprocess watershed simulation. Water Resources Research 43:W08419, doi:10.1029/2006WR005752. Rallison, R. K. 1980. Origin and evolution of the SCS runoff equation. Pp. 912–924 In: Proceedings of Symposium on Watershed Management, Boise, Idaho. New York: American Society of Civil Engineers. Ramaswami, A., J. B. Milford, and M. J. Small. 2005. Integrated Environ- mental Modeling: Pollutant Transport, Fate, and Risk in the Environment. New York: John Wiley and Sons. Reckhow, K. H., and S. C. Chapra. 1999. Modeling excessive nutrient loading in the environment. Environmental Pollution 100:197–207. Roa-Espinosa, A., and R. Bannerman. 1994. Monitoring BMP effectiveness at industrial sites. Pp. 467–486 In: Proceedings of Engineering Foundation Conference, Stormwater NPDES Related Monitoring Needs. H. C. Torno (ed.). Sartor, J., and G. Boyd. 1972. Water Pollution Aspects of Street Surface Con- taminants. EPA-R2-72-081. Schnoor, J. L. 1996. Environmental Modeling: Fate and Transport of Pollutants in Water, Air, and Soil. New York: John Wiley and Sons. Schwarz, G. E., A. B. Hoos, R. B. Alexander, and R. A. Smith. 2006. The SPARROW surface water-quality model: theory, application, and user documentation. U.S. Geological Survey Techniques and Methods Report, Book 6, Chapter B3. Selbig, W. R., and R. T. Bannerman. 2007. Evaluation of Street Sweeping as a Stormwater-Quality-Management Tool in Three Residential Basins in Madison, Wisconsin. USGS Scientific Investigations Report 2007-5156, 103 pp. Shaver, E., R. Horner, J. Skupien, C. May, and G. Ridley. 2007. Fundamentals of Urban Runoff Management: Technical and Institutional Issues, 2nd edi- tion. Madison, WI: EPA and North American Lake Management Society, 237 pp. Smith, R. A., G. E. Schwarz, and R. B. Alexander. 1997. Regional interpreta- tion of water-quality monitoring data. Water Resources Research 33(12):2781–2798. Soeur, C., J. Hubka, and G. Chang. 1995. Methods for assessing urban storm water pollution. Pp. 558 In: Stormwater NPDES Related Monitoring Needs. H. C. Torno (ed.). Reston, VA: Engineering Foundation and

338 URBAN STORMWATER MANAGEMENT IN THE UNITED STATES ASCE. Stenstrom, M. 2007. The Future of Industrial Stormwater Monitoring or, What Are Some Pitfalls and Ways We Can Improve Monitoring, for Industries and Others as Well. Presentation to the NRC Committee on Reducing Stormwater Discharge Contributions to Water Pollution. December 17, Ir- vine, CA. Stenstrom, M. K., and H. Lee. 2005. Final Report. Industrial Stormwater Monitoring Program. Existing Statewide Permit Utility and Proposed Modifications. Tague, C. L., and L. E. Band. 2004. RHESSys: Regional Hydro-Ecologic Simulation System—An object-oriented approach to spatially distributed modeling of carbon, water, and nutrient cycling. Earth Interactions 8:1–42. TetraTech. 2003. B. Everett Jordan Lake TMDL Watershed Model Develop- ment. NC DWQ Contract No. EW030318, Project Number 1-2. Wang, J., T. A. Endrenyi, and D. J. Nowak. 2008. Mechanistic simulation of tree canopy effects in an urban water balance model. Hydrological Proc- esses 44:75–85. WEF and ASCE. 1998. Urban Runoff Quality Management. WEF Manual Practice No. 23. ASCE Manual and Report on Engineering Practice No. 87. Reston, VA. Wigmosta, M. S., L. W. Vail, and D. A. Lettenmaier. 1994. A distributed hy- drology-vegetation model for complex terrain. Water Resources Research 30:1665–1679. Wong, T. H. F. 2000. Improving Urban Stormwater Quality–From Theory to Implementation. Journal of the Australian Water Association 27(6):28–31. Wong, T. H. F., H. P. Duncan, T. D. Fletcher, G. A. Jenkins, and J. R. Coleman. 2001. A Unified Approach to Modeling Urban Stormwater Treatment. Pa- per presented at the Second South Pacific Stormwater Conference, Auck- land, June 27–29, pp. 319–327. Wright, T., C. Swann, K. Cappiella, and T. Schueler. 2005. Urban Subwater- shed Restoration Manual No. 11: Unified Subwatershed and Site Recon- naissance: A User's Manual (Version 2.0). Ellicott City, MD: Center for Watershed Protection. Xiao, Q., E. G. McPherson, J. R. Simpson, S. L. Ustin and Mo. 2007. Hydro- logic processes at the urban residential scale. Hydrological Processes 21:2174-2188.

Next: 5 Stormwater Management Approaches »
Urban Stormwater Management in the United States Get This Book
×
Buy Paperback | $75.00
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

The rapid conversion of land to urban and suburban areas has profoundly altered how water flows during and following storm events, putting higher volumes of water and more pollutants into the nation's rivers, lakes, and estuaries. These changes have degraded water quality and habitat in virtually every urban stream system. The Clean Water Act regulatory framework for addressing sewage and industrial wastes is not well suited to the more difficult problem of stormwater discharges.

This book calls for an entirely new permitting structure that would put authority and accountability for stormwater discharges at the municipal level. A number of additional actions, such as conserving natural areas, reducing hard surface cover (e.g., roads and parking lots), and retrofitting urban areas with features that hold and treat stormwater, are recommended.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  6. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  7. ×

    View our suggested citation for this chapter.

    « Back Next »
  8. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!