The New York City Department of Environmental Protection (NYC DEP) conducts monitoring of a very large set of parameters within their watersheds, reservoirs, conveyances, and related facilities with real-time and near real-time sensor arrays as well as grab samples and laboratory analyses. These data help to guide operations, assess management actions, and fulfill regulatory requirements. Monitoring data related to water quality and streamflow, along with watershed and reservoir models, are used to varying degrees in evaluations of engineering, water supply operation and design, and program assessments. The most relevant example is the Operations Support Tool (OST), a combined water quantity/water quality model that simulates water availability and quality throughout the NYC water supply system and is used to inform decisions about system operation and planning.
NYC DEP’s monitoring and modeling programs are not officially a part of the Watershed Protection Program, yet the study Statement of Task (Box 1-1) appropriately asks the Committee to review these programs. The total annual cost of monitoring and modeling, including the OST, is about $6.7 million per year ($5 million for monitoring, $1.1 million for OST, and $0.6 million for modeling). These programs, which constitute approximately 7 percent of the total annual cost of the watershed protection effort, play a very significant role in ongoing evaluation of the entire Watershed Protection Program. This chapter reviews the monitoring and modeling programs and makes recommendations for improving these critical support efforts. This chapter has four major sections: (1) monitoring and statistical assessments based on the monitoring data, (2) deterministic modeling of the watershed, (3) modeling of individual reservoirs, and (4) the OST.
NYC DEP operates an extensive monitoring program covering the entire NYC watershed and water delivery system, described in the Watershed Water Quality Monitoring Plan (the Plan; NYC DEP, 2016a). In addition, the high-level findings from that monitoring system are documented in annual NYC DEP Watershed Water Quality reports (with NYC DEP, 2018, 2019 being the most recent two reports). These NYC DEP reports contain detailed descriptions of the monitoring network and the vast array of contaminants sampled by that network.
Monitoring and assessment activities are ideally conducted as an integrated program, rather than monitoring standing alone as separate and distinct, for two reasons. One is that the ultimate value of monitoring is only realized when it is coupled with assessments in which raw data are converted to information that can form the basis of actions. The second reason is that assessment is a crucial step in any adaptive management process. Findings of the assessment process, whether statistically robust or only general indications, should be the trigger to modify future monitoring activities and future watershed management activities. As elaborated on below, strengthening the linkages between monitoring and assessment and embracing new approaches to data analysis and the display of results could be beneficial to NYC DEP, regulatory agencies, and stakeholders.
This section defines four broad objectives of an integrated monitoring and assessment program, followed by a high-level overview of the monitoring system. The section comments on the monitoring and assessment efforts (as described in NYC DEP, 2018) and suggests ways NYC DEP could enhance the information content and usefulness of their monitoring results. The goal of the section is to relate how the assessments, based on the data, can better guide the Watershed Protection Program in the future and enhance understanding of the whole system for the water supply managers, regulators, and stakeholders. Having a strong data analysis and assessment program, with a wide range of tools and products, is vital to the continued success of the Watershed Protection Program. The products need to include those aimed at regulatory agencies as well as the interested public, decision makers (including those who decide about future investments of public funds), and the science community.
Information Objectives of a Water Quality Monitoring and Assessment Program
Before describing the NYC DEP monitoring and assessment program, it is worth defining four possible objectives for a source-water monitoring and assessment program for a water utility.
The first objective of any monitoring system is to warn of potential or existing water quality problems. This objective encompasses both short-term forecasts (hours to several days) as well as “nowcasts” that describe conditions in near real time. Near-term assessments are critical to safe operation of the system by supporting operational adjustments (e.g., temporarily avoiding delivery of potentially low-quality water) and for instituting enhanced monitoring to track possible sources of these poor conditions with an aim toward temporary or permanent resolution of the problems. For the warning function, timeliness is key to an effective program while accuracy is of less concern. The program should focus on identifying major threats to the system and accept some percentage of false alarms. The timeliness issue points to the value of surrogate measures of water quality. Surrogate measures are variables that can be sensed in real time, statistically related to the water-quality variables of greatest interest, and telemetered to provide input to a warning system that alerts utility managers about the need to undertake immediate responses.
Documenting the history of water quality is a second objective of any monitoring system. This includes quantifying the transport, storage, and mobilization of a wide range of contaminants throughout the system. Here, the goal is to accurately describe the concentrations and fluxes of constituents of concern (particularly nutrients, particles, pathogens, and algae) and the watershed conditions (such as discharge and temperature) that drive them. Accuracy is important for this objective; timeliness is not critical. This description of the system is particularly important for improving the capabilities of system managers to understand and predict system responses. For example, such descriptions are needed to answer questions about how algal blooms in a reservoir respond to the inputs of nutrients, turbidity, and water as well as weather factors such as temperature and sunlight. Analyses of the data could include “after action” reviews of important events such as extreme storms, extreme droughts, or temporary failure of system components such as wastewater treatment plants (WWTPs) in the watershed. The description of actual conditions in the watershed requires that data analysts make their best effort to estimate conditions on each day in the period being monitored and make statistical inferences about the unsampled days based on the available data. These descriptions are vital to the evaluation of process-based models used in planning and management, such as the watershed model called the Soil and Water Assessment Tool (SWAT) and the model of reservoir processes CE-QUAL-W2—both discussed in more detail later.
Trend analysis is another objective of monitoring and entails describing the evolving behavior of watersheds in delivering contaminants to downstream locations (typically the reservoirs). The focus is on time scales of a decade or more with an emphasis on the role that changes in land use, land management practices, watershed protection strategies, and climate may be playing as drivers of water quality. It is a probabilistic assessment of the variables that are crucial to watershed protection, in contrast with the description objective, which is focused on providing an accurate day-to-day record of actual conditions. Examples of trend analysis questions include: What has been the change in the 95th percentile of the distribution of daily average turbidities at some location in the watershed over the previous two decades? How has the mean annual flux of total phosphorus (TP) to a reservoir changed over the last two decades?
Such characterizations are valuable for several reasons. First, they can help managers evaluate the progress being made toward achieving watershed protection goals. For example, deterministic watershed models may suggest that actions in the watershed should have reduced loadings by 20 percent. Trend assessment is the empirical check on that prediction, revealing what reduction actually happened. Second, they may help identify surprises, such as an unanticipated increase in nutrient fluxes even though actions and modeling of their consequences suggested that reductions should have taken place. Finally, they can provide feedback to the deterministic models being used to help design and prioritize effective watershed protection strategies. They do this through an iterative process of adjusting the coefficients or structure of deterministic models and then determining the extent to which the hindcasts of the models reproduce simulated outcomes that match the observed trends.
Support for Predictive Modeling
Predictive modeling uses monitoring data to improve the understanding of linkages between the driving variables and water-quality outcomes. Driving variables can include watershed land uses, land-use practices (e.g., cropping systems and waste disposal systems), and climatic factors such as the probability distributions of air temperatures and precipitation amounts. The water quality outcomes can be probability distributions of concentrations or fluxes of water-quality variables in the streams and reservoirs. Improved understanding is accomplished through analysis methods that use the observed weather, hydrology, land use, and water quality data to estimate the parameters of a geospatially referenced model that statistically relates the inputs to the relevant water-quality outputs. Such models can help predict future water quality response to the combination of changes in climate, land use, land-use practice, and best management practice (BMP) implementation that are expected in the coming years or decades. These models are needed to prioritize management actions and investments to help ensure that the best quality of water is delivered by the NYC water supply system.
Overview of NYC DEP Monitoring and Assessment Program
The overall NYC DEP monitoring and assessment program can be categorized as follows, which is described as a “pyramid.” Compliance monitoring is described by NYC DEP as the pinnacle (highest priority) of the monitoring system. Activities further down the list form a “base which provides the context for interpreting the compliance information.” The elements of the program in priority order, as described by NYC DEP (NYC DEP, 2016a) are:
- Compliance monitoring required to meet all federal, state, and local regulations to ensure the safety of the water supply for public health;
- Monitoring to assess the watershed protection and improvement strategies required by the filtration avoidance determination (FAD). This includes evaluation of the status and trends in the water supply and evaluation of the effectiveness of the watershed protection policies;
- Monitoring to support current and future predictions to ensure operational decisions in the short term and policies over the long term. In particular, these data are needed to support the Modeling Program which is aimed at prediction of future conditions at time scales of hours to decades;
- Surveillance to ensure delivery of the best quality water (i.e., rapid identification of problems in support of short-term operating decisions).
The program has been reviewed internally about every five years since 1997; the most recent of these reviews is reflected in the Plan (NYC DEP, 2016a). The evolution of the monitoring program reflects emerging issues, new monitoring technologies, and improvements in scientific understanding of water quality. The Plan correctly notes that a monitoring program is not simply the sum total of what is measured, where it is measured, how often it is measured, and how it is collected, its quality is ensured, and it is stored. Rather, it includes the objectives of the monitoring, the data analysis methods used, and the reporting procedures whereby clients of the monitoring program are informed (those clients being water consumers and the general public, regulators, and operators of the water supply system). The Plan makes a particular point of citing a seminal paper in the history of water quality monitoring called “The ‘Data-Rich but Information-Poor’ Syndrome in Water Quality Monitoring” (Ward et al., 1986). That paper emphasizes that monitoring systems must be based on the information goals of the program and that ongoing analysis of the data must be conducted alongside the data collection process. In short, data collection must be seen as a means to an end (creating information) rather than an end in itself. The Committee strongly endorses this perspective.
The monitoring program has many component activities. First, the monitoring system includes three types of sites focused on water quantity. These include 25 meteorological stations, about 70 snow survey sites, and about 55 stream gages. Second, the monitoring and assessment of both water quantity and water quality parameters occurs in the following locations in the watershed: rivers, reservoirs, WWTPs, and key points in the engineered water delivery system. At these locations, measurements of physical variables (such as flow or temperature), chemical concentrations (both dissolved and total), particulate material (either as a concentration or a turbidity value), and biological parameters (such as chlorophyll a, harmful algal concentrations, or benthic organisms) are taken. Table 12-1 summarizes the sites at which water-quality grab samples are taken in 2018. This report focuses almost entirely on the watershed sites, but the distribution system sites are included for completeness.
In addition to this extensive system of grab-sample monitoring sites, NYC DEP collects data on a near-continuous basis with sensors located in several types of environments at a set of robotic monitoring sites, known as RoboMon. These include stream-monitoring sites, reservoir profiling buoys, fixed-depth buoys and also an under-ice buoy operated during the winter season. These systems typically measure temperature, turbidity, and specific conductivity at 15-minute intervals, and the data are transmitted to NYC DEP facilities in real time. The 19 RoboMon System sites collectively record about 1.3 million measurements per year.
TABLE 12-1 Summary for 2018 of the Number of Grab Samples Collected, Water Quality Analyses Performed, and Number of Sites at which Sampling was Conducted
|System||Number of Samples||Number of Analyses||Number of Sites|
SOURCE: NYC DEP (2019, Table 1.1).
The NYC DEP monitoring program is evolving toward using advanced statistical modeling techniques to turn the large body of collected data into a set of information products for detecting and understanding water quality changes in the watershed. The Committee applauds this direction and encourages them to evolve toward increased data interpretation and more communication of trends (or lack thereof) observed. At present, NYC DEP does not seem to be using these new statistical methods in its annual summaries but is apparently using some of them in communications with various stakeholders. Monitoring program summary products, if properly designed and presented, can be very useful for communicating the success of watershed protection activities and emerging topics of concern. They can also help win support for needed, but unpopular, actions that are being taken. This section gives examples of other monitoring and assessment programs in the United States that have effectively communicated monitoring results.
Two important foundations of the entire monitoring program are (1) laboratory accreditation (using National Environmental Laboratory Accreditation Conference standards and an outside accrediting body), and (2) a modern Laboratory Information Management System that provides for efficient and accurate input of all data types into a single data management system. The Laboratory Information Management System also provides needed interfaces from that system for operational, regulatory, and scientific uses of the data. NYC DEP has made major commitments to these systems as a foundation for all elements of the monitoring process.
The compliance aspects of the monitoring system are designed to be responsive to the regulations of the U.S. Environmental Protection Agency (EPA), the New York State Department of Health, the New York State Department of Environmental Conservation, and NYC DEP. The measurements are of the following variables: total or fecal coliform (described in Chapter 11), turbidity, temperature, pH, free chlorine residual, and discharge (flow rate). All of these involve frequencies ranging from continuous to several times per week. A few locations require less-frequent sampling for pathogens, lead, copper, a wide range of inorganic chemicals including nitrate, disinfection by-products, organic chemicals, and asbestos. Basins designated as “phosphorus-restricted” (see Chapter 3) have additional monitoring requirements.
Water Quantity Monitoring
The Plan includes a discussion of the networks used to collect data on meteorological variables, snow water equivalent, and streamflow. These extensive networks have operated over many decades and generally report data in near real time. They are critical for water supply system operations (including their use in the OST) and valuable in analysis systems designed to interpret water quality data. NYC DEP does not appear to engage (or contract with others for this purpose) in studying these meteorological or hydrologic variables from a trends perspective. Analyses in the NASEM (2018) report on OST showed that potentially important trends in streamflow have taken place in recent decades. Evaluating the role that these ongoing hydrologic changes may be playing in water quality is a topic that should be considered. The NYC DEP has expressed a strong interest in climate change, as it may affect the reliability and quality of water delivered to the customers. They have done several model-based studies of climate change impacts, but have not put substantial emphasis on research studies of recent changes in the climate and hydrologic system in the watershed. If NYC DEP actively engaged (either alone or in partnership with universities and/or federal agencies) in the analysis of trends already happening in these water quantity variables, it would help keep the OST up to date, given that OST depends on historical records to generate its short-term ensemble forecasts. It will also help NYC DEP determine if the water-related trends that are emerging are similar to what the current generation of climate models would generate as hindcasts, given the trajectory of greenhouse forcing that has occurred to date. Such an analysis would help NYC DEP evaluate the potential usefulness of these climate models for estimating hydrologic changes over the next few decades.
Groundwater does not seem to be included in NYC DEP’s overall monitoring strategy. Groundwater levels, monitored and telemetered daily, can be a useful indicator of watershed hydrologic conditions and a useful surrogate for estimating the groundwater contribution to stream flow. Runoff that follows a surface-water pathway to the stream may have a very different chemical signature than water entering the stream from groundwater. Knowing the status of groundwater quantity and quality in a particular watershed could enhance the understanding of water quality variations in streams and reservoirs. In addition, hydrologic forecasting, particularly for drought conditions, might benefit from having groundwater data to constrain the watershed model being used. Direct observation of nutrient concentrations in selected shallow groundwater wells can be a valuable contribution to the understanding of pathways and lag times for nutrient transport and can help set appropriate expectations for the timelines for BMPs to realize their potential for water quality improvement of streams and reservoirs. A modest investment in monitoring groundwater quantity and quality is warranted.
The remaining four subsections discuss aspects of the monitoring and assessment program as they relate to the four purposes described at the beginning of this chapter: warning, descriptions, trend analysis, and predictive modeling. It focuses on ideas to improve the value of these activities in terms of furthering the goals of the Watershed Protection Program.
The monitoring program places high priority on knowledge of the hydrologic flow system starting with meteorological inputs, snowpack information, streamflow, and flows through the engineered system. NYC DEP makes extensive use of in situ sensors in reservoirs for temperature, pH, specific conductivity, and turbidity. The reservoirs generally have water-profiling buoys that provide water quality information over the range of depths and now include under-ice buoys in Ashokan Reservoir. This under-ice capability is important because high-turbidity events can happen during the winter when traditional monitoring systems using surface-based sampling would be unable to assess the potential threat of highly turbid water. In addition, six river monitoring locations, upstream of some of the reservoirs, have in situ sensors that monitor every 15 minutes for at least turbidity and temperature. These sensor data are telemetered to central locations where NYC DEP operations staff can view these conditions in near real time (delays of only a few minutes). This information facilitates management decisions that can maximize the use of less-turbid water in other parts of the system and thereby avoid delivery of highly turbid water through the Catskill Aqueduct to downstream parts of the system. Samples are also collected at these locations to facilitate frequent calibration of the instruments. The extensive use of in situ real-time monitoring is a very positive aspect of the overall monitoring and assessment program.
NYC DEP could begin to use the real-time turbidity sensor information along with real-time streamflow and perhaps other real-time variables to make nearly continuous estimates of suspended-sediment concentrations (perhaps for each of several particle size classes) and total phosphorus concentrations at the site where the turbidity sensor is located. The idea of using turbidity as a statistical surrogate for suspended sediment concentration or total phosphorus concentration is well known and used by NYC DEP and the U. S. Geological Survey (USGS) among others (Rasmussen et al., 2009). The questions of interest in such statistical experiments include the evaluation of the error properties of such estimates, their usefulness as predictors of reservoir turbidity a few hours or few days into the future, and their usefulness as a means of providing real-time inputs to deterministic reservoir water quality models such as the W2 model that is a part of OST. These types of estimates are likely to be much more accurate than those that come from the use of infrequent water samples and a statistical model of concentration as a function of time, season, and discharge. Thus, these estimates can serve some of the other objectives such as program evaluation. Moreover, given that mitigation efforts are underway in the various watersheds to control turbidity and/or phosphorus, it is important that there be regular (once every year or two) checking to see if the surrogate regression relationships are still valid or if they need to be updated because of changes in sources of the materials.
NYC DEP might also explore some of the new technologies that have recently been developed for real-time measurement of suspended sediment concentration and particle size distributions. Acoustics and acoustics
combined with turbidity show great promise in advancing the estimation of suspended-sediment concentration in rivers (Agrawal et al., 2019; Landers et al., 2015). Laser diffraction is another technology being used in the field to more quickly obtain particle-size distribution and volumetric-sediment concentration (Czuba et al., 2015).
Real-time in situ instrumentation can provide an indication of changing water-quality conditions that may affect drinking-water treatment processes. Some sensors, such as dissolved oxygen, pH, and fluorescence, can be indicative of algal activity and dissolved organic matter. These measurements, properly calibrated to the reservoir of interest, may show increases in biomass and changes in algal community composition and the potential for trihalomethane production. Best practices for sensor applications for harmful algal blooms are still being determined, but there is a lot of potential (Bertone et al., 2018). Data collected by these sensors may be used to develop statistical models to estimate the probability of occurrence, and in some cases concentration, of algal toxins and taste-and-odor compounds (Foster et al., 2019; Francy et al., 2016; Graham et al., 2017). These models are site specific and may change over time because of changes in water quality or shifts in algal community composition (Graham et al., 2017). In lakes and reservoirs, day of year, wind speed and direction, pH, dissolved oxygen, and algal fluorescence are often used as explanatory variables, while in rivers, day of year, flow, algal fluorescence, and turbidity are often used as explanatory variables. In addition to sensors, imaging technologies are also being used as indicators of changing water-quality conditions. These technologies range from simple cameras (Foster et al., 2019) to in situ flow cytometry that captures images of individual organisms1 and can provide near-real-time notification of the presence of potentially harmful algae. NYC DEP could explore these newly emerging in situ monitoring technologies as warning signals for trihalomethane production, taste and odor, and the potential for harmful algal blooms.
The goal of description is to create the most accurate possible time history of the concentration and flux of various constituents of interest such as suspended sediment (or turbidity), phosphorus, nitrogen, and carbon through the system. Such after-the-fact descriptions can be used to test the accuracy of models, particularly for the reservoirs. Some of the kinds of questions that one might consider would be these. (1) How does turbidity at the outlet of Ashokan Reservoir respond to the time history of suspended sediment input along with variables such as streamflow, air temperature, wind, and ice conditions? (2) How does total phosphorus concentration in Cannonsville Reservoir respond to the time history of dissolved and suspended phosphorus inputs as well as flow conditions, temperature, solar insolation, and other environmental variables? (3) How do variables such as chlorophyll a or trihalomethane formation potential in a reservoir respond to the inputs of nitrogen, phosphorus, carbon, water, and sediment, as well as air and water temperature? NYC DEP has several simulation models intended for such goals.
The Plan (NYC DEP, 2016a) has a chapter on the role of modeling to meet the FAD-related goals. It has an extensive description of the monitoring data that are used to support water quality modeling in the reservoirs. The data appear to be used both for quantifying inputs to the reservoir (particularly fine particulates—approximated by turbidity measurements) and for calibration or verification of the various models.
The Plan describes how daily loads of various constituents are calculated, as follows: “Daily loads will be calculated by multiplying concentration by mean daily flow. Linear interpolation will be used to estimate analyte concentration between sampling days. The product of mean daily flow and estimated concentration from linear interpolation will be the estimated daily load.” These daily loads (flux) are among the most important inputs. The models are aimed at estimating the spatial and temporal distribution of these pollutants in a given reservoir and at their outflow. Daily flux estimates are also crucial to the calibration and verification of models (such as the SWAT model) of how watersheds respond to weather conditions (rain, snow, temperature), land use, BMPs, and point sources in the watershed. Yet, the linear interpolation approach described here does not take full advantage of the data that NYC DEP has collected. Much better estimation methods are widely available.
The deficiency of the interpolation approach can be described with a simple thought experiment. Consider suspended sediment as the analyte of interest. Suppose for March of some specific year there is one sample value from near the beginning of March and the next is near the end of March. Further suppose that both were collected on days when discharge was fairly low, but for a few days between the two sample values there was a storm event with much greater inflow. Experience suggests that the suspended sediment concentration on those high-flow days was probably much higher than the concentration on the two sampled days. But, using the Plan’s method, the estimated concentration on those high-flow days will be assumed to be a weighted average (a linear interpolation) of the concentrations observed on the two sampled days. The flux calculation for the days between those two samples is made up of these interpolated concentration values, each multiplied by the discharge value for that day. It is almost certain that in this case the monthly estimate will be much too low.
A better approach is to use the whole body of sample data (not just these two points) to build a general statistical model for this sampling site. The model would describe how concentration varies with discharge and perhaps with other variables related to time of year, the recent history of discharge, a possible long-term trend, and a random error component. Even more complex models are possible and useful where the data are sufficient to support that. Such an approach, which captures much more of the information content of the data set, can produce much more accurate flux estimates than the interpolation method they use. Methods (with well documented software) that follow this general type of approach include regression-based methods such as Load Estimator (Runkel et al., 2004) or weighted regressions on time, discharge, and season (WRTDS) (Hirsch et al., 2010). Recently, a new method has been proposed and tested that uses these underlying models plus the autoregressive properties of the residuals to improve these daily estimates (Lee et al., 2019; Zhang and Hirsch, 2019). NYC DEP should consider using these approaches to estimate inputs to some of their deterministic water quality models. Also, generalized additive models are another category of very useful tool for making such temporal or spatial inferences from a large data set (Murphy et al., 2019). Studies by Lee et al. (2016, 2019), indicate that these regression-based methods are far more accurate than interpolation for many analytes. NYC DEP could make more effective use of their large water quality data sets to obtain more accurate estimates of daily fluxes of nutrients and sediment.
The Plan provides some general discussion of how the NYC DEP uses monitoring data for trend detection. They note that, given the inherent variability of water quality data, trend detection needs to be based on a fixed data collection program for at least five years. In fact, five years is not sufficient for the analysis of trends, unless those trends are of a very large magnitude and very abrupt. Furthermore, trends large enough to be discerned with only five years of data are not likely in the WOH watershed, given the relatively low intensity of land use and low population density, and also given that the point sources have already been upgraded to a high level of treatment. Standard protocols for trend analyses that consider trends at multiple time scales, no shorter than ten years and as much 30 years, would be more appropriate for development and use.
Trend analysis in water quality is difficult because of the profound role that interannual variation in precipitation can have on water quality conditions. In general, meaningful assessments of water quality trends need to be made on records long enough to contain multiple examples of wet conditions and dry conditions. It is unlikely that this criterion can be met in a period of less than about a decade. Fortunately, NYC DEP has been persistent in its data collection and has many data sets that span two or more decades. They are beginning to use new statistical tools designed to enhance the signal-to-noise ratio in the data and so provide more meaningful assessments of the trends in the watershed.
When considering nutrient records from streams, there is more value in analysis of flux trends rather than concentration trends, as discussed in Chapter 4. Reservoirs tend to respond to the mass loading of nutrients (Jarvie et al., 2017; Stamm et al., 2014) over months to years. Trend analyses on stream concentration data alone (where equal weight is given to concentrations on high flow days and low flow days) are not suitable metrics because they are not meaningful predictors of reservoir concentrations, which are the primary driver
of eutrophication. Simply put, what matters is the mass of nutrients entering the reservoir and not the concentration of nutrients entering the reservoir.
Examples from Other Programs
A few regional- to national-scale examples that enable one to perceive differences in flux across space and time are offered. The first example comes from the Chesapeake Bay restoration program and shows ten-year trends in orthophosphorus across a set of monitored watersheds. The results, based on analysis of all of the orthophosphorus data collected at each site from 2007 through 2016, are shown in Figure 12-1. Such a depiction makes it clear which areas of the watershed are seeing increases in orthophosphorus and which are seeing decreases, and it shows these changes on a scale of changing yields, so that the bars on the graph depict the changes on a per-unit-area basis. This type of analysis has been useful to the Chesapeake Bay Program in identifying where changes in strategy are needed. The method used here is designed to “filter out” the year-to-year variation in flux from year-to-year variations in streamflow. These kinds of analyses and graphical displays of results would help track and compare water quality results across the NYC watersheds and over time.
NYC DEP is already using one such analysis (WRTDS; Hirsch et al., 2010), and open-source software for its application already exists (Hirsch and DeCicco, 2015). One particular value of this type of analysis is that these empirical trend results can be compared to trends simulated by watershed models in which the change in management practices is considered in the model runs. In the Chesapeake Bay, the use of conservation tillage in some agricultural areas (in watersheds such as the Conestoga in Pennsylvania and the Choptank in Maryland) in the watershed water quality model hindcasted substantial declines in phosphorus yield over this period of time, but the trend analysis of the data for these watersheds show significant increases in phosphorus yields. The discovery of this discrepancy led to further examination by the watershed modeling team and a change in the underlying process representation in the watershed model (see Kleinman et al., 2019). This “surprise” factor is an important reason for empirical trend analyses. It is an important ingredient to test and improve models used in planning.
Also from the Chesapeake Bay watershed is an example of a unified analysis of the total phosphorus yields over a period of approximately 30 years. As shown in Figure 12-2, the circles represent the estimates of yields for each individual year. These meet the “Description” objective in the four-part system of objectives. These individual-year values are strongly influenced by the precipitation in each year, but the solid curve represents the flow-normalized yield, representing the changes over time, without the confounding effect of the year-to-year variations in weather. This type of illustration helps the viewer to see, at a glance, which watersheds are high yield and which are low yield and at the same time depicts the magnitude and direction of the trends in yields.
Another graphic product shown in Figure 12-3 comes from a nationwide study of the loadings of nutrients to either the oceans or the Great Lakes (Oelsner and Stets, 2019). This one graphic helps the viewer quickly assess the range in yields across many watersheds, but also the direction of the trends in those yields and a measure indicating where those trends are considered to be statistically likely. It also places the observed trends in context (e.g., small change in a large value, or large change in a small value). This kind of a depiction of yield values and their trends could be useful for an analysis of priorities among the several watersheds in the NYC water supply system.
Finally, results from a study of soluble reactive phosphorus flux from major rivers draining to the western basin of Lake Erie, an area with acute problems from harmful algal blooms in some recent years, are shown in Figure 12-4. These analyses are based on a variation of the WRTDS trend analysis method known as generalized flow normalization (Choquette et al., 2019), which can be used to isolate the portion of the trend that is a result of changes in the discharge versus concentration relationship and the portion of the trend that is due to the trends in discharge itself.
This analysis shows clearly that, at the first five sites shown here (the Raisin [RAIS], Maumee [MAUW], Sandusky [SAND], Honey Creek [HONE], and Rock Creek [ROCK]), the trend of an increase in soluble reactive
phosphorus (SRP) is statistically considered to be highly likely. The figure also indicates the extent of that change from 1995 to 2015, and shows the relative contributions to the overall trend in each watershed from increases in discharge and increases in concentration. In these watersheds, the upward trend in discharge has been an important contributor to the overall increase in SRP flux, accounting for approximately one-third of the total increase at each site.
These graphical approaches have some potential application to the display of trend analysis results for NYC DEP. Visual displays of trend information are important tools for management and communication of water quality issues. These are just a few examples of possible approaches, but all of them depict multiple dimensions of the data in each plot; in doing so, they capture the most important trends in time or space, while providing insight into the complexity of various situations. Finding the best way to display the results of water quality assessment information is not simple, but there are many texts (e.g., Tufte, 2001) and classes that can help analysts find effective ways to do it. The large investment that NYC DEP has made in data collection, when analyzed with the right tools, could enhance understanding of the system and help convey the water quality changes that have happened during the past few decades.
Suggested Improvements to NYC DEP Trends Analysis
Published NYC DEP trends analyses have yet to benefit from the types of analyses and data display discussed above, as shown by the examples presented in Box 12-1 and Appendix B. These examples are highlighted because they show how potentially unsubstantiated conclusions can be drawn when rigorous analyses are not undertaken.
As described in detail in Box 12-1, the analysis of SRP and TP data from the Cannonsville watershed found in Hoang et al. (2019) led to the authors’ conclusions that both point and nonpoint-source programs had helped to reduce dissolved phosphorus in the Cannonsville Reservoir from 1993 to 2014. This conclusion was not based on any attempt to statistically model the behavior of SRP and TP in the watershed. Rather it was derived from the results of a watershed model, SWAT-Hillslope or SWAT-HS (discussed later in this chapter), that attempted to account for how agricultural BMPs would lead to reductions in pollutant loading to streams and eventually to reservoirs. The authors also concluded that the TP record was chaotic and not interpretable. The Committee’s analysis found that by using an appropriate statistical model, it became possible to meaningfully interpret these data.
One of the values of carrying out formal analyses of water quality trends is the potential for surprise. In the Hoang et al. (2019) paper, the following statement is made about model results for the period 1993-2015: “The combined effects of both types of watershed protection programs (WWTP upgrades and nonpoint source controls) are 39 and 43 percent load reductions for soluble and particulate phosphorus, respectively.” That would be about a 42 percent reduction in TP. Using a portion of the TP trend analysis presented in Chapter 4 (results shown in Figure 4-11) the Committee’s analysis suggests an increase of 77 percent over this same time frame. Because of the high variability of TP, the uncertainty about the magnitude of this change is quite large. Expressed as a 90 percent confidence interval, the change over the 1993–2015 period runs from a decrease of 14 percent to an increase of 150 percent. Nonetheless, there is a rather profound contradiction between model results (a large decrease) and a statistical analysis of the empirical data (likely a large increase). This surprise should have led to a much more critical analysis of the SWAT-HS model.
Such discrepancies between observed trends and model simulations are not unique to the NYC watershed. Previous studies in areas such as the Upper Mississippi River and the Chesapeake Bay watersheds have similarly found large discrepancies between modeled changes in water quality and observed changes in water quality, as described in Box 12-2. This discrepancy does not mean that simulation models such as SWAT-HS are of no value in planning for watershed protection. The model may become highly useful, but only after it is modified to be consistent with the temporal pattern of the observations (as seen in Figure 4-11). NYC DEP can build from the Hoang et al. (2019) study by asking why the SWAT-HS model failed to simulate the observed behavior. Was it poor input information, or an inadequate representation of the processes, or failure to consider legacy sources of phosphorus? NYC DEP should examine how SWAT-HS might be improved to more closely replicate the observed decadal-scale changes or consider if some entirely different model formulation would be more appropriate.
Substantial data regarding NYC water supplies and watersheds (observational, derived, and modeling data) should be documented and made available publicly. This will stimulate data analysis and research at a high level, it will minimize needed external investments in targeted studies, it will maximize the value of rate-payer dollars, and, importantly, it will lead to better understanding of the system. The Committee encourages the NYC DEP to provide access to their data in order to facilitate exploration of these rich data sets by external researchers and thereby contribute to expanded knowledge of the changing water quality of the watershed. An easy way to provide such access to these data sets is to enter them into the National Water Quality Portal,2 a partnership of the USGS and EPA. The portal makes the data discoverable and available to potential users. Making their data easily available to all scientists regardless of affiliation can lead to research conducted by (and funded by) others that can materially add to the understanding of the water quality of the watershed. Also, the ability to combine data from the NYC DEP data set with data from other watersheds increases the potential for developing new insights. It might be particularly valuable to compare water quality trends observed in the NYC watersheds with other nearby watersheds with similar landscape and land use attributes but without the investment in BMPs that has taken place in the NYC watersheds. This is another way to gain insight on the question of whether these investments have made a difference in water quality to date.
Support for Predictive Modeling
The final reason for collecting monitoring data is to have a set of observations that can be used to calibrate and test models that are used to evaluate strategies for achieving water quality goals such as reductions of phosphorus loads into a reservoir. There are many forms that such models can take and they range from those that are strongly based on fine-scale process models (such as SWAT-HS) to those that are strongly based on statistical concepts and simpler process conceptualizations (such as SPARROW [SPAtially Referenced Regression On Watersheds] described in the next section of this chapter). Regardless of the type of model that is used, observations are critical. The observations can be used to drive the calibration process either through a formal parameter optimization or through trial-and-error parameter estimation. Assuming that the water quality goal under consideration is related to time-averaged loads to a reservoir (averaged over months, seasons, or years), the data needed for parameter estimation and performance evaluation must take the form of loads computed over an appropriate averaging time. This means that the data collected at the network of monitoring sites needs to include streamflow and constituent concentrations and it needs to cover a period of several years in order to make sure that it covers a reasonable range of hydrologic conditions.
Because such models are typically used to evaluate management strategies that have time scales of decades, it is highly valuable for the observed data to cover time scales of decades and that information on the drivers of water quality also cover decades. To have some confidence in whatever modeling approach is taken, the model should be run in hindcast mode, driving it with observed atmospheric inputs (temperature, precipitation, etc.) and the history of land use/land cover, BMP implementation, and point-source discharges. Selecting a model to use for evaluating future strategies must be based, in part, on hindcast results. The question is, do the hindcasted loads track reasonably well with the historical trends in loads. This is more challenging than what is typically done, which is to evaluate models on how well they reproduce discharge and concentration time series for individual storm events or individual seasons. NYC DEP is well situated to make good use of their data for this purpose because the data (1) are long term, (2) they cover discharge as well as concentration data, and (3) they cover the full range of seasons and hydrologic conditions. However, it does not appear from the studies that NYC DEP has published that the monitoring data have been used in this manner. Rather, they tend to be used to simply demonstrate a match of observed and predicted concentrations at the short-term scale (hours to days). The Committee urges NYC DEP to make more use of their monitoring data for parameter estimation and evaluation of models in relation to long-term trend.
This shift of modeling perspective implies that NYC DEP will need to pay very close attention to documenting changes in land use/land cover, application of BMPs, and changes in point-source loads. In many respects these data are just as important to the building of predictive models as are the water quality observations themselves. Having accounting tools to summarize the changes in inputs and outputs of nutrients (see discussion on net anthropogenic nitrogen input and net anthropogenic phosphorus input methods in Box 12-3) can be very helpful in organizing the key data on these drivers of water quality change.
NYC DEP does not appear to be using methods that allow direct estimation of water quality based on the characteristics of their individual watersheds and reservoirs and the observed load, or trends in load, computed from their monitoring data. Properly used, these methods should make it possible to project future changes in water quality based on anticipated or proposed changes in land use, management practices, and climate. The next section, on watershed modeling, delves into some of the modeling options. Regardless of which approaches are used, they all depend on high-quality monitoring data and also on the synthesis of those monitoring data through some of the approaches described in the previous sections on Description and Trend Analysis.
As stated in the Executive Summary of the Multi-Tiered Water Quality Modeling Program Annual Status Report (NYC DEP, 2016b),
The New York City Department of Environmental Protection has maintained a program of water quality modeling for its water supply system for over 20 years. The general goal of this program is to develop and apply quantitative tools, supporting data, and data analyses in order to evaluate effects of land use change, watershed management, reservoir operations, ecosystem health, and climate change on water supply quantity and quality.
Hence, NYC DEP has developed a multi-tiered modeling system that includes short-term and longer-term weather predictions; watershed models that simulate runoff generation and basic runoff water quality; reservoir models that simulate water balances and fate and transport of heat, suspended solids and other contaminants; and operational models that balance water availability, domestic supply needs, environmental flows, water quality, and treatability (see Figure 12-5). This work is conducted by the Water Quality Modeling Group, which is part of the Water Quality Science and Research, Directorate of Water Quality within the Bureau of Water Supply.
NYC DEP has relied almost exclusively on deterministic models, arguably the most common approach for water quality modeling, to inform decision making. Water quality modelers generally have employed this approach since it supports the development and application of increasingly detailed models with small space/time scales and elaborate ecologic description. This often results in the impression that detailed management questions can be satisfactorily addressed using these models. A weakness of this approach is that, after some point, increasing model detail often translates into increased model prediction uncertainty. Further, the complexity of these models can hinder the estimation of prediction error. For reservoir modeling and for OST, deterministic modeling has proven to be successful for the NYC DEP. However, as proposed in the following section on watershed models, a new approach that includes both statistical and deterministic models may provide the best information to guide and evaluate watershed management.
NYC DEP’s Current Suite of Watershed Models
Watershed models simulate runoff generation and the loading of pollutants from the land to nearby waterbodies. The NYC DEP has used several different watershed-scale (pollutant loading) models since the 1990s, when they first began to develop total maximum daily load (TMDL) calculations for reservoirs that were impaired due to phosphorus loading. In their TMDL reports, NYC DEP variously applied the “Reckhow Model” (or “Nested Reckhow Model” or “Reckhow Land Use Model” which are all essentially the same; Reckhow et al., 1992) and the Generalized Watershed Loading Function (GWLF) model (Haith, 1985; Haith and Shoemaker, 1987) to assess allowable phosphorus loading to achieve compliance with the TMDLs, based on the existing guidance value of 20 µg/L. The Reckhow Model is essentially a lumped-parameter model that relies on the selection of phosphorus export coefficients (unit area loads) to model land use-specific loadings. The reports from the 1990s concluded with the statement that more complex models would be applied in the future for TMDLs.
Since then, the NYC DEP has worked with two watershed models: the GWLF model/Variable Source Loading Function (VSLF) model (Easton et al., 2008a; Haith, 1985; Haith and Shoemaker, 1987; Schneiderman et al., 2007), and SWAT and its variants SWAT-Variable Source Area (SWAT-VSA) and SWAT-HS (Arnold et al., 1998; Easton et al., 2008b; Hoang et al., 2019; Mukundan et al., 2015; Pradhanang et al., 2013). As discussed in a subsequent section, NYC DEP should consider moving toward hybrid models, such as SPARROW which uses a data-driven approach to gain a better handle on uncertainty (see Table 12-2).
Generalized Watershed Loading Function
GWLF, a lumped (spatially averaged) model, and it derivative, VSLF, a semi-distributed model, provide daily estimates of streamflow, nutrients, dissolved organic carbon (DOC), and total suspended solids (TSS). GWLF is essentially a compromise between export coefficients and mechanistic models, by incorporating simple mechanistic details into the model. Schneiderman et al. (2002) developed and applied a revised version of GWLF to the watershed of Cannonsville Reservoir, implemented in Vensim (Ventana Systems Inc., 1999). Schneiderman et al. (2002) and NYC DEP (2006, 2007) calibrated this version of GWLF for Pepacton, Ashokan, West Branch, Schoharie, Neversink, and Rondout watersheds. Schneiderman et al. (2007) applied VSLF to the Cannonsville Reservoir watershed to assess the spatial and temporal evolution of critical source areas (e.g., areas that produce disproportionate runoff and nutrient and/or sediment loads).
In all of these applications, NYC DEP employed the common practice for calibration and validation of water quality models by splitting the time series of water quality data into a few years for calibration and the following few years for validation. When the length of the time series is limited (e.g., five to ten years), the notable differences between the calibration and validation datasets are likely to be natural forcing functions such as precipitation and temperature, and the resulting streamflow. It is unlikely, however, that the forcing functions that are the principal focus of the model application, such as changes in (1) land use/land cover, (2) the probability distributions of climate variables, and/or (3) point-source pollutant discharges, will change very much. To the extent that pollutant loads to a waterbody change over the time period of the model calibration, it may largely be due to interannual variations in weather. Unfortunately, in the absence of a data set focusing on land use or management changes, model validation may be inconclusive. Hence, when cause-effect mechanisms are not characterized by field data, the application and use of mathematical models should be based on best professional judgment that model equations reflect the actual processes. It is also important to consider the extent to which the model has been successful as a decision support tool in other applications (e.g., the Chesapeake Bay Program).
The applications of GWLF by NYC DEP (e.g., NYC DEP, 2011) have focused on watershed-wide evaluations of implemented and proposed watershed management activities (such as BMP implementation) and land use/land cover changes on reservoir water quality. For example, by linking GWLF to a reservoir model, NYC DEP (2011) simulated the effect of FAD programs and land use changes within the Cannonsville and Pepacton watersheds on reservoir water quality. Much of the focus of this analysis was on how three different modeled scenarios (baseline and two FAD programs, driven by meteorological records) would affect nutrient loads over time. Results from this application support use of the GWLF model for long-term dissolved phosphorus load
TABLE 12-2 Summary Comparison of Watershed Models
|Model||Model Type||Temporal Scale||Spatial Dimension||Response Variables|
|Reckhow Model||Export coefficients||Annual||Subwatershed||Phosphorus, nitrogen|
|GWLF/VSLF||Simple mechanistic||Daily||Subwatershed||Streamflow, nutrients, DOC, TSS|
|SWAT||Complex mechanistic||Daily||Field-subwatershed||Streamflow, runoff, nutrients, sediments, etc.|
|SPARROW||Nonlinear regression on simple mechanistic equations||Annual or seasonal||Small to regional watershed||Nutrients, pesticides, suspended sediments, organic carbon, fecal bacteria|
NOTES: “Model type” refers to the mathematical structure or parameter estimation approach, “temporal scale” refers to the time-step used in the model for each computed result, “spatial dimension” refers to the scale at which the watershed may be broken down into individual modeled sections, and “response variables” refers to the types of variables that may be predicted using the model. DOC = dissolved organic carbon; GWLF/VSLF = Generalized Watershed Loading Function/Variable Source Loading Function; SPARROW = SPAtially Referenced Regression On Watersheds; SWAT = Soil and Water Assessment Tool; TSS = total suspended solids.
estimation from a large watershed but say little about how effectively the GWLF model can predict the effect of relatively small land management changes on water quality response. Other assessments of land use/land cover change and watershed management activities using GWLF (NYC DEP, 2006, 2007; Schneiderman et al., 2002) have had mixed results. That is, model predictions of streamflow matched observations quite well, but there was a wide discrepancy between predictions of pollutant load and observations; in general, such results are not unusual for these types of models. Because GWLF is not subject to error analysis, it is possible that the uncertainty in GWLF predictions of pollutant load may be large. As an alternative to GWLF, the USGS SPARROW model (described in detail below) is fitted using nonlinear regression, and therefore it does provide an error analysis.
Soil and Water Assessment Tool
The SWAT model (Arnold et al., 1998) is widely used for runoff and pollutant transport assessment in agricultural systems. NYC DEP and partners have worked extensively with SWAT, both developing new model routines to better capture relevant regional hydrologic and water quality processes (Easton et al., 2008b, Hoang et al., 2017; White et al., 2011) and for evaluating water quality management and/or climate impacts (Hoang et al., 2019; Pradhanang et al., 2013). Building on the modified GWLF model (VSLF, Schneiderman et al., 2007), Easton et al. (2008b) modified SWAT to include variable source area (VSA) hydrology, by relating soil moisture and the propensity of runoff generation to terrain attributes. Not only did this modification allow for better prediction of runoff source areas, but it also improved phosphorus predictions. White et al. (2011) further modified SWAT-VSA to include a subfield-level water balance, which replaced the standard curve number method of runoff prediction. This model provides predictive skill equal to or better than the curve number method, but with a more physically based approach, which reduced parametrization needs of the model, thus making it easier to apply to data-scarce regions. Hoang et al. (2017) developed the latest realization of the SWAT model (SWAT-HS) for the watershed by introducing an aquifer that allows lateral transfer of subsurface water across the landscape based on the SWAT-VSA concept. This modification better accounts for water movement between modeling units, thus improving prediction of where saturated, runoff-generating areas occur. Compared to GWLF, SWAT has the benefit of scalability; that is, the model can be parametrized at scales from the subfield to the basin level, should such input data exist. This scalability is due to the use of the
hydrologic response unit (HRU) as the finest-scale unit at which calculations are performed. HRUs are created by spatial overlays of soil type, land use, and in the case of SWAT-VSA and SWAT-HS, topographic index values; thus, the scale of these input data define the scale of the HRU. Together, these modifications to SWAT are a concerted effort to ensure that areas that contribute disproportionate nonpoint source pollution loads to water bodies are correctly predicted, thus allowing evaluation of management efforts to improve water quality.
NYC DEP believes that the SWAT model offers “the promise of increased accuracy in simulating both current conditions and in the evaluation of changes in land use and climate change” (NYC DEP, 2016b). While distributed-parameter models such as SWAT provide more space/time detail than lumped-parameter models such as GWLF, it is possible that this additional detail comes at the expense of increased prediction uncertainty, or increased effort needed to adequately parameterize the model.
It is clearly desirable for modelers to present information on model prediction uncertainty so that decision makers and stakeholders are better off than they are in the absence of knowledge of this uncertainty. Several investigators have attempted to apply uncertainty analysis to the pollutant loading models described above. Hong and Swaney (2013) developed a Visual Basic program that applies Monte Carlo simulation for GWLF applications. Partial uncertainty analyses using generalized likelihood uncertainty estimation (Beven and Binley, 1992) applied to SWAT have been presented by Benaman (2003) and Tolson and Shoemaker (2008). Yang et al. (2007) used Markov chain Monte Carlo on a SWAT application. Yet none of these uncertainty analyses were comprehensive, in that uncertainty in all parameters, model inputs, and model structure was not included.
Reckhow (2014) presents a decision analytic approach for using model prediction uncertainty that could serve as a mental framework or be implemented by the NYC DEP. This approach is based on the expected value of sample information, which can be calculated for a model prediction accompanied by uncertainty analysis (e.g., SPARROW; see next section). For example, the watershed modeling group might identify several subwatersheds that are predicted to be higher contributors of pollutant loading and considered to be good candidates for BMP implementation. Using the expected value of sample information to evaluate the degree to which additional monitoring data might reduce model prediction error of the impact of these BMPs, the NYC DEP would have a stronger basis (lower prediction error) for selecting particular subwatersheds for management interventions. This becomes a decision analytic approach when the cost of each BMP is compared to the cost of the corresponding additional monitoring.
Future of Watershed Modeling Within NYC DEP
The primary need for watershed modeling in support of the Watershed Protection Program is to inform decisions on actions that the subprograms (e.g., the Watershed Agricultural Program and Watershed Forestry Program) might consider to further protect water quality. The SPARROW model of the USGS could assist greatly in achieving this goal.
SPARROW is a statistical modeling approach to quantify nutrient sources and delivery to individual reaches within the watersheds. The SPARROW technique was developed by Smith et al. (1997), with contemporary code fully available and documented in SAS (Schwarz et al., 2006) and in R (Alexander and Gorman Sanisaca, 2019). It is a spatially explicit water quality modeling approach used to predict sources and fluxes of nutrients in surface waters (see Figure 12-6 for the model structure). Predictions from the steady-state model versions typically explain spatial variations in mean-annual, stream-nutrient fluxes in relation to hydrological and biogeochemical processes that have long-term impacts on the supply, loss, and transport of nutrients in terrestrial
and aquatic environments (Schwarz et al., 2006). Separating land and water components provides estimates of the rates of nutrient delivery from point and nonpoint sources to streams, rivers, lakes, reservoirs, and estuaries. Because the nutrient contribution from each source is tracked separately, the percent contribution from each source category (e.g., fertilizer, manure, WWTPs, atmospheric deposition) can be computed for each reach. In addition to loads and concentrations in streams and source share allocations, model predictions include stream and reservoir losses.
SPARROW expands on conventional regression methods by using a mechanistic model structure in correlating observed nutrient fluxes in streams with spatial data on nutrient sources, landscape characteristics, and stream properties. Model parameters are determined statistically using nonlinear parameter estimation techniques and mass-balance constraints on model inputs (sources) and outputs (Alexander and Gorman Sanisaca, 2019; Schwarz et al., 2006). Parameters are estimated by spatially correlating observed stream water-quality records with geographic data on climatic and watershed properties (e.g., precipitation, temperature, topography, vegetation, soils, and wetlands) as well as any relevant point and nonpoint sources and atmospheric deposition. Such parameter estimation with mass-balance constraints ensures that the calibrated model will not be more complex than can be supported by the data. This provides an objective statistical approach for evaluating alternative hypotheses about the effects of land use changes, watershed management, point-source controls, or climate change in controlling the seasonal patterns of surface water quality.
The estimation techniques used in SPARROW also have the advantage of providing measures of uncertainty in model parameters and predictions (Alexander and Gorman Sanisaca, 2019; Schwarz et al., 2006). By spatially referencing nutrient source locations and watershed attributes to surface water flow paths, defined according to a stream network, and imposing mass-balance constraints, the model has been shown to improve the accuracy of predictions of stream export and the interpretability of model coefficients over other approach-
SPARROW can address smaller-scale questions such as where are the hot spots of nutrient pollution problems, and where should mitigation and management efforts be best targeted. Model simulations are used to estimate long-term average values of water characteristics, such as the amount of nitrogen or phosphorus that is delivered downstream, on the basis of existing monitoring data. Further, the framework generates hypotheses about precursor sources of nutrient loadings to surface waters that can be used in conjunction with source information from the nutrient mass-balance accounting budgets (discussed below). SPARROW modeling results can help managers determine how to reduce loads of nutrients and design protection strategies, design strategies to meet regulatory requirements, predict changes in water quality that might result from management actions, and identify gaps and priorities in monitoring.
Most applications of SPARROW have been static analyses. That is, they consider a base year (say the year 2012) and develop a statistical estimate of fluxes at the monitored locations in the region being modeled. These statistical estimates of flux (i.e., mass loadings of constituents in streamflow) are time averages covering a period of about a decade surrounding the base year. They use methods such as Load Estimator or WRTDS to make these flux estimates and their uncertainty. Then, the explanatory variables are computed for the many subwatersheds. Some of these explanatory variables are landscape and river network variables that do not change over time and some of them are landscape and point-source variables that are set to values for each subwatershed that are based on knowledge of base-year conditions. In the past few years, four studies have been published that apply SPARROW in a dynamic manner (to locations other than the NYC watershed) to model the changes in flux over time. That is, these new dynamic versions of SPARROW are designed to model the changes in fluxes observed over a multi-decade period (e.g., base years of 1992, 2002, and 2012) and use explanatory variables that change over time based on point-source changes and land use and BMP changes over the period. The four approaches are documented by Ator et al. (2019), Chanat and Yang (2018), Strickling and Obenour (2018), and Wellen et al. (2012). The four papers all use somewhat different approaches to build their models of changing fluxes based on changing activities in the watershed. Each of them offers a possible approach that could be applied to nitrogen, phosphorus, or carbon transport in the NYC watersheds. Making use of both the spatial and temporal dimensions of water quality change will make these dynamic SPARROW formulations important tools for guiding water quality management.
SPARROW results have played valuable roles in assessing nutrient sources and evaluating possible nutrient source reduction strategies in the Chesapeake Bay Restoration and in the Mississippi River Basin efforts to reduce hypoxia in the Gulf of Mexico. Examples of the types of products that can be created for such purposes include “Response of Nitrogen Loading to the Chesapeake Bay to Source Reduction and Land Use Change Scenarios: A SPARROW-Informed Analysis (Miller et al., 2020), and “Regional Effects of Agricultural Conservation Practices on Nutrient Transport in the Upper Mississippi River Basin” (Garcia et al., 2016). Because NYC DEP has nutrient monitoring data that span several decades, they could create a dynamic SPARROW model for the NYC watershed, taking advantage of the broader SPARROW framework and data sets that already exist in the northeastern U.S. SPARROW model (Ator, 2019), adding in the NYC DEP monitoring data and more detailed information about land use/land cover, nutrient use, and point sources. NYC DEP could partner with the USGS or with university scientists to develop SPARROW applications that are relevant to the evaluation of nutrient reduction strategies in the NYC reservoir watersheds.
SPARROW models for nitrogen and phosphorus have been applied across the northeastern region including the NYC watershed (Hoos et al., 2013; Moorman et al., 2014). Figure 12-7 shows the estimated yields (estimated for a base year of 2002) of phosphorus to four of the NYC reservoirs and for each reservoir apportions that total yield into source shares: wastewater, urban land, fertilizer, manure, and background. For three of the four reservoirs, the background phosphorus yields are heavily dominated by background (the phosphorus that comes from the soil and decaying plant material of forested lands). The simulations also suggest that for the Cannonsville Reservoir, the largest source share after background is manure, which is about twice the magnitude of the fertilizer, urban (which is mostly septic systems), and wastewater sources. (Note that these results predate the completion of the WWTP upgrades and as such a newer analysis would probably find the wastewater share to be very small). Similar but expanded information is presented in Table 12-3, which shows SPAR-
ROW results for seven NYC reservoirs. The table illustrates the differences in categories such as wastewater, manure, and fertilizer across the watersheds and gives the magnitude of the total masses. These kinds of outputs can be very useful for informing decision makers about priorities for future action. They make it possible to compare across watersheds and across source types to identify the best targets for future control measures.
The Committee recommends that the NYC DEP make use of the SPARROW statistical modeling framework to quantify nutrient sources and delivery to individual reaches within the watersheds. A SPARROW model that would best serve NYC DEP would need to extend beyond the geographic domain of the NYC watersheds which means that it would need to build on the foundation of the most recent regional SPARROW model. Such a model could provide indications of which subwatersheds are the largest sources of nutrients to each of the NYC reservoirs. Then, coupled with more detailed process models that capture details of soil topographic setting, and land use (such as the SWAT model) they could identify hot spots that would define areas where investments in control measures may have the largest payoffs in terms of reductions of nutrient delivery to each reservoir. To run a SPARROW model the NYC DEP would need to carry out spatially detailed nutrient mass-balance accounting budgets (see Box 12-3). These nutrient accounting budgets provide a structure for tracking the complex set of changes in nutrient movements into and off the landscape as management practices change (e.g., precision feed management, shifts between dairy and beef production, manure-to-energy conversion projects). SPARROW modeling results and simulations, which build on these accounting budgets, can help managers to determine where and how to reduce loads of nutrients and design protection strategies, to design strategies to meet regulatory requirements, to predict changes in water quality that might result from management actions, and to identify gaps and priorities in monitoring.
Final Thoughts on How Watershed Models Should Be Used to Inform Decision Making
The NYC DEP modeling program has had demonstrated success with OST and with selected applications of reservoir models. The watershed modeling program also has demonstrated some success with watershed-wide nonpoint source modeling impacts on downstream reservoirs. However, the key role for watershed modeling should be to provide an evaluation of the expected impact of proposed changes in watershed management programs (e.g., agriculture and forest management), preferably with an estimate of the uncertainty in that evaluation. Unfortunately, the NYC DEP has not undertaken this type of analysis.
The NYC DEP watershed modeling program should become more vertically and horizontally integrated with other NYC DEP and watershed programs, particularly those associated with land acquisition and agriculture. Decisions by the NYC DEP concerning what land in the watersheds might be targeted for land acquisition (see Chapter 7) or for agricultural BMP implementation (Chapter 5) should be informed by the watershed modeling.
To identify areas of the watershed producing large nutrient and sediment loads the Committee recommends that SPARROW (or dynamic SPARROW or other statistically based model) be used. In such applications, SPARROW results could be presented by subwatersheds and then ranked according to their relative contribution of pollutant loads or by their delivery ratios. Once these subwatersheds are identified and ranked, SWAT and/or some other appropriate quasi-mechanistic model can be used to identify areas of those subwatersheds where pollutant loads are originating from and perhaps the impact of targeted BMP implementation. The use of SPARROW for watershed-scale modeling rather than SWAT is preferred because of SPARROW’s ability to account for uncertainty and its ease of application for all of NYC’s watersheds. The recommendation of SWAT rather than SPARROW for targeted small-scale land use/land cover analyses is justified because SWAT can more mechanistically model most of the common agricultural BMPs.
The Committee believes that the watershed modeling program of NYC DEP needs to grow from its present staffing and funding level and that this growth should be aimed at achieving a balance between scientists with a primarily statistical perspective and those with a primarily process-based perspective. Building and updating the data sets to support such models needs to be a priority for the modeling group. Collaboration with outside
TABLE 12-3 Sources and Loadings of Total Phosphorus in Selected NYC Water Supply Reservoirs
|Total phosphorus source shares (percent contribution and yield)|
|Total P yield
|Total phosphorus source shapes (mass loads)|
|Total P load P
groups such as the USGS and university researchers will be crucial to enhancing the modeling program but it must have a central core of NYC DEP staff who are deeply involved in the development of new tools as well. NYC DEP needs to maintain a critical mass of in-house expertise to make watershed modeling a working tool of the Watershed Protection Program.
Reservoir water quality models are mathematical representations of aquatic systems that simulate physical, chemical and biological properties and processes in lakes and reservoirs. Water quality modeling of the source water supply has been conducted by the NYC DEP for decades to assist in complying with the regulatory requirements of the FAD. In particular, meeting the turbidity standard of 5 nephelometric turbidity units (NTUs) in the Kensico Reservoir withdrawal has been a key element to management of the water supply and ensuring compliance with the SWTR. Reservoir models are core components of the OST, which is used to help achieve the turbidity objective and inform management of the City’s water supply and its water quality.
The OST includes CE-QUAL-W2 models developed for Schoharie, Ashokan, Kensico, and Rondout reservoirs. CE-QUAL-W2 (Cole and Wells, 2002, 2015) are two-dimensional hydrodynamic and water quality models, colloquially called W2. Recent efforts to extend the CE-QUAL-W2 models to the Neversink, Cannonsville, and Pepacton reservoirs further increase the comprehensiveness of model coverage for reservoirs in the WOH watersheds. The CE-QUAL-W2 models simulate two-dimensional (2-D) longitudinal-vertical transport, stratification/mixing, and water quality. The model was originally developed at the U.S. Army Corps of Engineers Waterways Experiment Station, has been extensively refined over time, and is now managed by Dr. Scott Wells at Portland State University (Wells, 2018). The model is suited for run-of-the-river reservoirs where the primary gradients are in the longitudinal (down-river) and vertical directions; the model has been used for over 450 lakes and reservoirs, nearly 300 rivers, and numerous estuaries and other waterbodies (Wells, 2018). In addition to water budget, heat budget, and turbidity, CE-QUAL-W2 can predict concentrations of dissolved oxygen (DO), nutrients, phytoplankton, and other parameters. The recent NASEM (2018) review of the OST (NASEM, 2018) provides comprehensive descriptions of the CE-QUAL-W2 models developed for Schoharie, Ashokan, Kensico, and Rondout reservoirs. The report indicates that NYC DEP reservoir models have generally been shown to reliably predict temperature and turbidity levels, although they have not been used to predict other water quality parameters.
Prior to the development of the W2 model for Rondout Reservoir, turbidity delivered from the Delaware Basin via Rondout Reservoir was specified at historical median values. The Water Quality Modeling Group recently developed a probabilistic CE-QUAL-W2 turbidity model for the Rondout Reservoir (Figure 12-8) to provide greater insight into turbidities delivered from this important component of NYC’s source water supply (NYC DEP, 2016b).
Since the CE-QUAL-W2 model is a deterministic model based on temporal hydrologic, meteorological, and water quality boundary conditions, it provides a single predicted set of outputs that is dependent upon the representativeness of the input data and the 2-D approximation. Although deterministic models can be calibrated to historical data and used to hindcast when conditions can be known quite accurately, reliably forecasting future conditions can be quite challenging. The NYC DEP modeling team addressed these challenges in an early application by adopting a probabilistic approach using 47 different National Weather Service hydrologic forecasts to generate a series of model predictions. Meteorological data from 1987 to 2012 provided 25 different data sets to drive the model, yielding a total of 1,175 simulations (NYC DEP, 2016b). Model output from 100 simulations for the November 13–December 1, 2015 time period demonstrates the median, 10 percent and 90 percent levels of turbidity withdrawn from Rondout Reservoir (Figure 12-9A). Observed turbidity levels over this period were below historical median values, often by substantial margins, and were typically within
the 10th-90th percentile bands (Figure 12-9B). The approach represents a sophisticated use of CE-QUAL-W2 and provides significant improvement in estimates of turbidity in withdrawals from Rondout Reservoir compared with historical median values used previously.
General Lake Models for Cannonsville and Neversink Reservoirs
Other models have been developed that can simulate protozoa, bacteria and other constituents of potential concern to the NYC water supply. For example, the General Lake Model (GLM)-Aquatic Ecodynamics (AED) model (Hipsey et al., 2013, 2014) and Aquatic Ecosystem Model (AEM3D) (Hodges and Dallimore, 2016) are 1-D (vertical) and 3-D models, respectively, that include a wide range of components and reactions. The GLM-AED model is an open-source model developed by the University of Western Australia and the Global Lake Ecological Observatory Network, while AEM3D is a derivative of the Estuary, Lake and Coastal Ocean Model - Computational Aquatic Ecosystem DYnamics Model (ELCOM-CAEDYM) model (Hipsey, 2014; Hodges and Dallimore, 2014). Both models are increasingly used by researchers, water resource managers, and others to better understand and manage lakes and reservoirs. The Water Quality Modeling Group at the NYC DEP has recently developed GLM models of the Cannonsville and Neversink reservoirs and completed water-balance and temperature-stratification simulations for the 2007-2008 period, with the longer-term objective to simulate organic carbon dynamics and formation of disinfection byproduct precursors provided by the more comprehensive biogeochemical capabilities of GLM-AED (compared with CE-QUAL-W2).
With inflow, outflow, and meteorological data as inputs, the GLM models predicted very similar trends in temperature for both reservoirs, with strong thermal stratification throughout the spring and summer, drawdown over the summer and fall, weakening of stratification in the fall, and mixing in early winter (NYC DEP, 2016b). Error analysis for temperature yielded root-mean-squared errors of 1.77–2.56 °C across the calibration and validation datasets, confirming that the models adequately represent the temperature characteristics and stratification-mixing dynamics of the lakes. Water column measurements at several locations across each of the reservoirs yielded very similar temperature profiles, demonstrating that the 1-D approximation is reasonable for temperature, although given that shortwave, longwave, latent and sensible heat fluxes are all surface phenomena, the validity of the 1-D assumption for turbidity, DOC, and DBP-precursors remains to be established.
Other NYC DEP Water Quality Modeling Efforts
The NYC DEP water quality modeling group has proven to be quite adept at identifying the correct level of model complexity to match the scientific, operations and/or management questions at hand. For example, utilization of the 1-D Upstate Freshwater Institute Lake Stratification Model No. 4 in long-term (39-year) simulations across a range of meteorological and hydrological conditions allowed simulation of the potential water quality benefits in Cannonsville Reservoir derived from projected reductions of nutrients from both point sources and nonpoint sources (Figure 12-10). Summer epilimnetic concentrations of chlorophyll a and total phosphorus for each of the 39 years of the long-term simulation, displayed as distributions with concentrations on the x-axis and number of years at a concentration bin on the y-axis, were favorably shifted to lower values with watershed interventions. For example, predicted summer-average chlorophyll a concentrations were lowered from a range of 9.5–13 µg/L and median value near 11 µg/L under 1990 land use conditions to 6–9.5 µg/L and median value of approximately 7 µg/L with 2006-2009 land use and implementation of point source and watershed BMPs (Figure 12-10, left column). Similar reductions in total phosphorus concentrations were also predicted with land use changes and implementation of point source and watershed programs (Figure 12-10, right hand column). While recognizing the substantial inherent uncertainty in these simulation results, the modeling nonetheless illustrates possible insight offered by application of watershed and reservoir water quality models, e.g., into potential long-term improvements in water quality that might be achieved under different land-use and nutrient control strategies.
Application of the 3-D Environmental Fluid Dynamics Code (Hamrick, 1992) allowed evaluation of a 1,700-ft diversion wall proposed to reduce short-circuiting and delivery of high levels of turbidity to the east basin Ashokan intake. The model allowed prediction of fine-scale hydrodynamic processes and demonstrated that the diversion wall would be ineffective at preventing high turbidity from reaching the intake (Figure 12-11).
Future of Water Quality Modeling Within NYC DEP
The NYC DEP Water Quality Modeling Group continues to make essential contributions to the mission of the NYC DEP. Nevertheless, some gaps exist between modeling and measurements. Opportunities also exist to enhance predictive capabilities.
One such area involves the comprehensiveness of assessments of predictions, observations, and operations. That is, while specific studies and analyses have been highlighted in available reports and presentations, it is unclear if a system-level analysis has been conducted to identify sources of uncertainty and the relative contributions to uncertainty in water quality. For example, turbidity levels in Kensico Reservoir outflow are a primary decision node. It is not clear, based on available documents and presentations, how closely predictions
from OST match observed turbidity levels in Kensico Reservoir withdrawal (Figure 4-8). Is there evidence of any systematic bias in reservoir model predictions? Are there specific meteorological, hydrologic, watershed or reservoir conditions where strong discrepancies between observations and predictions are often found? Such an analysis would help guide future efforts, not only with respect to modeling, but potentially also for monitoring, maintenance of current programs, and implementation of new protection efforts.
As noted in the National Academies’ report on the OST (NASEM, 2018), an area where substantial improvements in turbidity prediction would be expected is in better understanding of turbidity-flow relationships within the Catskill and Delaware watersheds. While turbidity generally increases with increasing flow, tremendous variability exists (e.g., Figure 4-7 in this report). Empirical regressions have been developed by NYC DEP that capture central tendencies in turbidity-flow relationships, but they fail to adequately capture the inherent variability present. As a result, uncertainty is not fully represented in model forecasts of turbidity based upon ensemble forecasts of streamflow and simple empirical regressions of turbidity-runoff. As stated in NASEM (2018), model formulations used for simulating turbidity loads to the reservoir should include uncertainty in discharge-based regression models. Reductions in uncertainty through incorporation of seasonal and antecedent conditions and other factors in turbidity-flow relationships would be expected to improve forecasts of turbidity.
Three-Dimensional Model Use
The CE-QUAL-W2 model has served as a valuable tool for the NYC DEP over the past two decades, and NYC DEP should be applauded for their sophisticated use of it in OST and in numerous other studies. As noted previously, NYC DEP has also applied 3-D models in a select number of cases, where understanding of finer-scale hydrodynamic processes was needed, although W2 remains the core model for OST and numerous other applications. Notwithstanding this, given the complex geometries of many of the reservoirs, dynamic meteorology of the region, and physical dimensions of reservoir inlet/outlet structures, dividing weirs, and other infrastructure, the coarse-scale, transverse-averaged discretization of a 2-D model such as CE-QUAL-W2 must necessarily average over larger volumes rather than be dictated by finer-scale hydrodynamic processes. As an example, the turbidity plume observed in Ashokan Reservoir in April 2005 demonstrates complex transport behavior not realistically represented using a 2-D laterally averaged model (Figure 12-12).
As another example, withdrawal turbidities from Rondout Reservoir were predicted from three sets of OST model runs for the period between July 11 and August 11, 2016 following a 7-inch storm on July 8-9. The model, used with forecasts in Position Analysis mode to assess the timing and magnitude of peak turbidity delivered to the Delaware Aqueduct, struggled to capture observed turbidity levels (NYC DEP, 2017). Out of 24 observed data points for this period, only four values were within the predicted 10th-90th percentile range, with differences between the observed values and the 10th-90th percentile range approaching 50 percent (Figure 12-13), whereas one would have expected about 19 values within the 10th-90th percentile range.
Simulations for the period between November 2 and 30, 2016, yielded somewhat better agreement, although predicted turbidity fell outside the 10th-90th percentile range for seven out of 23 sample dates. What these examples demonstrate is that the turbidities estimated by this approach vastly underestimate the variability of turbidity. For the Position Analysis approach to work properly, the model needs to project a realistic degree of uncertainty, which is not the case here. To what extent is the accuracy of predictions of turbidity diminished in the 2-D approximation? What other potential sources of error are not being adequately represented? The Committee attempts to answer this first question in Box 12-4.
Advances in computational power, memory, and parallelization make 3-D models based upon the Navier-Stokes equation practical for an increasing number of systems. Lattice-Boltzmann models are also increasingly being used to simulate hydrodynamics and transport in lakes, rivers and reservoirs (e.g., Kruger et al., 2017). Although 3-D models may remain too computationally demanding to be used in some long-term Monte Carlo-type applications, they provide a quantitative way to evaluate the compromises made using a 2-D approximation to real-world 3-D processes, thereby quantifying one potential source of error in 2-D model predictions and could help guide further enhancements in modeling and monitoring efforts. This 3-D analysis would be especially valuable for Ashokan Reservoir, given its complex morphology, unique infrastructure, and critical role in delivering water from the Catskill watershed, although the 2-D laterally averaged assumption introduces potential error in model predictions for other reservoirs as well.
Moreover, as observed for the analysis above, a 3-D representation does not necessarily require an excessively long execution time for simulating or forecasting conditions over modest time intervals (e.g., in the above simulation using relatively modest computational resources, a seven-day forecast could be completed in 8.5 minutes). With some balancing of grid-size, model performance and execution time, it is suggested that 3-D simulations may be viable for short-term (7 to 14 days) ensemble forecasting.
An important component of the overall NYC water supply is the OST. The OST was the subject of a recently published study by the National Academies (NASEM, 2018) and as such it is not a focus of this study. However, to understand the full range of actions and policies used by NYC DEP to achieve its watershed protection goals under the FAD, it is important that the reader have some familiarity with this analytical tool. The following is an overview of OST from the report:
“Surface water supply systems fundamentally involve transformation of precipitation inputs to drinking water by a system of watersheds, reservoirs, conduits, and treatment facilities. The NYC DEP’s Operations Support Tool (OST) informs management decisions by formally quantifying flow and quality changes as water moves though the Catskill, Delaware, and Croton systems to New York City. OST can be broadly described as a continuously evolving, state-of-the-art decision support system used by NYC DEP in water supply operations and planning (Porter et al., 2015). Its development was a direct outcome of the Catskill Turbidity Control Study—it was motivated by the need to mitigate fluctuations in source water turbidity in the Catskill system, which have the potential to cause exceedance of the 5 NTU maximum turbidity allowed for diversions from Kensico Reservoir in order to maintain filtration avoidance. In an idealized, simple system focused exclusively on this objective, the Ashokan Reservoir might be primarily managed to store higher-turbidity source water while water quality improves over time due to settling of suspended solids. The accounting of water routing for the NYC water system is not this simple, however, because tradeoffs between multiple objectives such as meeting water demands, controlling floods, and maintaining downstream releases from a large number of reservoirs must be concurrently considered and prioritized.
The OST decision support platform was developed because of the vast number of interdependent decisions that must be made to identify operational strategies that consider and appropriately balance multiple objectives at any point in time and in response to a wide range of possible future conditions. Although OST is not needed for operational responsiveness in some cases, it is intended to simulate decisions that closely approximate the decisions that a knowledgeable operator would make after considering and processing all of the available data, constraints, and possible operational controls. Manual completion of this task for the complex NYC water system may not allow for a complete evaluation of operational options; therefore, it is done by utilizing water quantity and water quality models that provide simulation of flows, reservoir levels, and water quality under a range of possible conditions for a future time period. OST is a flexible and modular platform—it accounts for water routing and provides the probability of predicted future operational conditions based on a wide range of near-real-time (meteorological, hydrologic, and water quality) data, several (deterministic and stochastic) modeled relationships, regulatory constraints, and operational insight.” (NASEM, 2018)
Fundamentally, OST is a distributed “ensemble forecasting tool” focused both on water quantity and water quality. It simulates outcomes in terms of flows, volumes, and turbidity levels of water at a large set of key locations in the entire NYC water supply system (such as reservoirs, streams, and aqueducts). In the future, OST will be enhanced to include other measures of water quality, but for now the water quality focus of OST is on turbidity (using the CE-QUAL-W2 model). The term “ensemble forecasting” is a class of forecasting techniques used in meteorology and hydrology where the forecast product is a conditional distribution of outcomes at specific locations and at specific time horizons. This term is in contrast to “point forecasts” where the forecast product is a single estimate of outcomes at each specific location and time horizon. For example, a point forecast provides a single trace of outcomes (such as streamflow or reservoir volume, or reservoir turbidity) into the future and might result in a statement that the best estimate of the volume of water and turbidity level in Ashokan Reservoir two weeks from today is 400 million cubic meters of water at a turbidity level of 15 NTU. In contrast, an ensemble forecast provides a large number (e.g., 50) of equally likely traces of outcomes and these might be summarized by statements about the probabilities of specific bad outcomes at the given time horizon. These bad outcomes could include turbidity levels at certain locations in the system that would require the addition of alum to achieve FAD objectives, failure to deliver sufficient water to customers, spillway flows at various reservoirs that would result in downstream flood damages, or releases to downstream reaches that violate habitat requirements.
Using OST, system managers can rapidly consider various short-term management actions and see how those actions would influence the probabilities of these specific bad outcomes. This can assist them in evaluating the various management options that are available to them over the coming days or weeks. These options are typically the rates of delivery of water from specific reservoirs to the aqueducts and to downstream river reaches.
It is crucial to understand that OST does not decide how the system is operated, but rather that it informs the managers of the potential consequences of various operational decisions. The OST is run repeatedly as watershed conditions change, incorporating the latest observations of precipitation, temperature, streamflow, reservoir contents, aqueduct flow rates, and turbidity levels throughout the system. OST always uses the latest ensemble forecasts of precipitation and temperature. During rapidly changing conditions (e.g., a storm or major snow-melt event) the OST might be run multiple times during a day as conditions and forecasts evolve. It is also used to evaluate strategies for dealing with the consequences of actions, such as taking various tunnels or gates out of service for repair or maintenance.
OST can operate in two modes, “Position Analysis Mode” (as described above), and in “Simulation Mode.” Position Analysis runs of OST typically focus on outcomes for a few months to a year into the future and initial conditions can be of great consequence. OST is a crucial part of meeting the “warning” function described earlier for the monitoring system. In Simulation Mode, OST runs last for many decades into the future (using historical weather and hydrologic conditions) as inputs. These runs are important for consideration of changes in engineering features of the system or potential significant changes in climate and/or land use.
OST does not control the quantity or quality of the water that flows off the landscape and into the system of reservoirs, but it does enable managers to make well-informed decisions about storing, releasing, and delivering water from the various parts of the system to achieve the best possible outcomes in terms of the quantity and quality of the water delivered to customers. As such, it has an important support function for the overall Watershed Protection Program. Due to its modular nature, other water quality models besides W2 can be “plugged into” OST, although this has yet to occur. The watershed models discussed earlier in this chapter have not yet been linked to OST; if they were, they would replace the ensemble streamflow forecasts that are currently used as input.
The NYC DEP deserves praise for their monitoring system design and for the accumulation of such data as streamflow, snowpack, turbidity, phosphorus, and nitrogen, which will become increasingly valuable as these
data sets grow. They should be reluctant to change monitoring locations or sampling protocols unless the arguments for change are highly compelling. The Committee urges NYC DEP to be unwavering in their attention to long-term monitoring, but also to develop a tradition of regularly subjecting those data to rigorous analysis, including the use of multiple models that range from dominantly statistical to dominantly process-based but strongly tied to the data.
Reporting on water quality should always include some formal statistical testing for trends over time. NYC DEP should apply WRTDS or other statistical modeling analyses to the many types of data they collect, with a focus on trends in flow-normalized flux of pollutants. These analyses have potential to provide a common basis to show the relevant trends for any given analyte across monitoring sites. Responding to this recommendation requires that NYC DEP develop formal statistical models and establish statistical protocols for the analysis of the many types of data that they collect. Statistically based tools can be used to provide projections of the consequences of future watershed protection activities and various land use changes on watershed fluxes. Sharing NYC DEP’s water quality data through a system such as the National Water Quality Portal will make it possible for other researchers to enhance their understanding of watershed water quality.
The New York City Department of Environmental Protection (NYC DEP) should clarify when report conclusions are based on deterministic model simulations and when they are based on observed data evaluated using a statistical model. NYC DEP needs to be cognizant of the limitations of deterministic models and use them for improved understanding and priority setting but avoid trying to use them to confirm program success. Deterministic models are important tools for the Watershed Protection Program, but they need to be part of a feedback process in which trends in observed data are compared to trends in the simulation outputs. If there are discrepancies between these results, the model should be modified to better correspond to the observed behavior of the system.
The New York City Department of Environmental Protection is urged to shift its modeling and reporting of progress in the Watershed Protection Program to a mass-balance approach. This requires a fundamental accounting strategy that estimates the sources and downstream disposition of constituents of concern. These strategies should have the geographic specificity to account for waters flowing to the major reservoirs and be divided by source types such as: manure or fertilizer application, septic tanks, urban stormwater, or other categories of point and nonpoint sources. These statistical approaches should start with nitrogen and phosphorus (where previous studies by others provide a strong foundation) and then move to carbon, fine sediment, and pathogens (which will likely prove more challenging to model). A mass balance approach is exemplified by nutrient accounting formulations such as NANI/NAPI and the statistically based watershed model SPARROW to produce maps of “delivered yield” of nutrients to downstream locations.
The New York City Department of Environmental Protection should begin evolving their annual water quality report toward becoming a data analysis report. The whole watershed protection effort would greatly benefit from development of temporal and spatial representations that can be used on an ongoing basis to describe how the watershed is responding to the factors that influence it: changing land use, changing application of watershed protection strategies, changing climate, and more. These representations of the data should be designed to enhance understanding of watershed conditions and trends on the part of NYC DEP managers, watershed stakeholders, regulators, other scientists, and the public. Improving the annual report should involve monitoring and assessment staff as well as representatives of the subprograms within the Watershed Protection Program. Remaking the entire annual report in a single year is not realistic, but should take place over several years.
Watershed modeling should become a more integral part of the Watershed Protection Program. As described in previous chapters, watershed modeling can inform the Land Acquisition Program, agricultural BMP implementation, and the Stormwater Program. NYC DEP should aim to have a modeling and data
analysis team evenly balanced between deterministic and statistical models and approaches, who actively engage with the program areas. NYC DEP has fully and effectively integrated the use of models for short-term decision support (e.g., the OST and reservoir water quality models); it is urged to similarly embrace watershed models for long-term decision support for the Watershed Protection Program. The data and the tools are available; it is simply a matter of determination and staffing to make it a reality.
Uncertainty estimates are not comprehensively assessed for any of the currently applied NYC DEP watershed models. Uncertainty analysis can be undertaken for statistical models such as SPARROW, since the model fitting (typically regression analysis) task yields the necessary error terms. However, process-based models (such as SWAT) are parameterized through expert judgment, since available data are usually insufficient to use a statistical approach (such as regression) for parameter estimation. This means that model and parameter error terms are not easily estimated, so any error analysis will be incomplete. Modeling approaches such as the generalized likelihood uncertainty estimation can be useful for this partial error analysis.
Increased use of 3-D models could improve predictions of reservoir hydrodynamics and water quality. Numerical simulations conducted by the Committee demonstrate that the 2-D laterally averaged approximation introduces significant potential errors in predictions of turbidity, even for long narrow reservoirs where the 2-D approximation is reasonably expected to hold, when compared with a 3-D representation. Given the improved representation of hydrodynamic processes with a 3-D framework and continuing advances in computational power, more extensive use of 3-D models would be appropriate. For example, a comparison of 2-D and 3-D model results for Ashokan Reservoir could quantify the errors in predicted turbidity concentrations that are introduced by the 2-D approximation for this important and morphologically complex reservoir. Moreover, increases in computational power and improved numerical schemes may make 3-D simulations viable for some short-term ensemble forecasting.
Agrawal, Y. C., H. C. Pottsmith, D. Dana, and O. A. Mikkelsen. 2019. Super-Turbidity Meter: LISST-AOBS Combines Optical Turbidity with Acoustics, E-proceedings of the 38th IAHR World Congress, Panama City, Panama. https://www.sequoiasci.com/wp-content/uploads/2019/06/Agrawal_Panama_AOBS_full_Paper.pdf.
Alexander, R. B., A. H. Elliott, U. Shankar, and G. B. McBride. 2002. Estimating the sources and transport of nutrients in the Waikato River Basin, New Zealand. Water Resources Research 38:1268-1290.
Alexander, R. B., and L. Gorman Sanisaca. 2019. RSPARROW: An R system for SPARROW modeling [Software release]. U.S. Geological Survey. DOI: https://doi.org/10.5066/P9UAZ6FO.
Arnold, J. G., R. Srinivasan, R. S. Muttiah, and J. R. Williams. 1998. Large area hydrologic modeling and assessment part 1: Model development. Journal of the American Water Resources Association 34:73-89.
Ator, S. W. 2019. Spatially referenced models of streamflow and nitrogen, phosphorus, and suspended-sediment loads in streams of the Northeastern United States: U.S. Geological Survey Scientific Investigations Report 2019–5118. https://doi.org/10.3133/sir20195118.
Ator, S. W., A. M. García, G. E. Schwarz, J. D. Blomquist, and A. J. Sekellick. 2019. Toward explaining nitrogen and phosphorus trends in Chesapeake Bay tributaries, 1992–2012. Journal of the American Water Resources Association 55(5):1149-1168.
Benaman, J. 2003. Uncertainty and Sensitivity Analysis for Watershed Models: Hydrology and Sediment Transport Modeling on the Cannonsville Reservoir System. Ph.D. Dissertation. Cornell University.
Bertone, E., M. A. Burford, and D. P. Hamilton. 2018. Fluorescence probes for real-time remote cyanobacteria monitoring: A review of challenges and opportunities. Water Research 141:152-162.
Beven, K., and A. Binley. 1992. The future of distributed models: model calibration and uncertainty prediction. Hydrological Processes 6:279-298.
Boyer, E. W., and R. W. Howarth. 2008. Nitrogen fluxes from rivers to the coastal oceans. Pp. 1565-1584 In: Nitrogen in the Marine Environment, 2nd edition. D. G. Capone, D. A. Bronk, M. R. Mulholland, and E. J. Carpenter (eds.). San Diego, CA: Academic Press.
Boyer, E. W., R. W. Howarth, J. N. Galloway, F. J. Dentener, P. A. Green, and C. J. Vörösmarty. 2006. Riverine nitrogen export from the continents to the coasts. Global Biogeochemical Cycles 20:GB1S91. doi:10.1029/2005GB002537.
Boyer, E. W., C. L. Goodale, N. A. Jaworski, and R. W. Howarth. 2002. Anthropogenic nitrogen sources and relationships to riverine nitrogen export in the northeastern USA. Biogeochemistry 57:137-169.
Chanat, J. G., and G. Yang. 2018. Exploring drivers of regional water‐quality change using differential spatially referenced regression—a pilot study in the Chesapeake Bay watershed. Water Resources Research 54(10):8120–8145.
Choquette, A. F., R. M. Hirsch, J. C. Murphy, L. T. Johnson, and R. B. Confesor Jr. 2019. Tracking changes in nutrient delivery to western Lake Erie: Approaches to compensate for variability and trends in streamflow. Journal of Great Lakes Research 45(1):21-39.
Claassen, R., A. Cattaneo, and R. Johansson. 2008. Cost-effective design of agri-environmental payment programs: U.S. experience in theory and practice. Ecological Economics 65(4):737−752.
Cole, T. M., and S. A. Wells. 2002. CE-QUAL-W2: A Two-Dimensional, Laterally Averaged Hydrodynamic and Water Quality Model, Version 3.1. User manual. U.S. Army Corps of Engineers, Instruction Report EL-02.
Cole, T., and S. A. Wells. 2015. CE-QUAL-W2: A Two-Dimensional, Laterally Averaged, Hydro-Dynamic and Water Quality Model, Version 4.0. User Manual. Portland, OR: Department of Civil and Environmental Engineering, Portland State University.
Conservation Effects Assessment Project (CEAP). 2012. Assessment of the Effects of Conservation Practices on Cultivated Cropland in the Upper Mississippi River Basin. Washington, DC: USDA Natural Resources Conservation Service.
Czuba, J. A., T. D. Straub, C. A. Curran, M. N. Landers, and M. M. Domanski. 2015. Comparison of fluvial suspended-sediment concentration sand particle-size distributions measured with in-stream laser diffraction and in physical samples. Water Resources Research 51:320–340.
Easton, Z. M., M. T. Walter, and T. S. Steenhuis. 2008a. Combined monitoring and modeling indicate the most effective agricultural best management practices. Journal of Environmental Quality 37:1798–1809.
Easton, Z. M., D. R. Fuka, M. T. Walter, D. M. Cowan, E. M. Schneiderman, and T. S. Steenhuis. 2008b. Re-Conceptualizing the Soil and Water Assessment Tool (SWAT) model to predict runoff from variable source areas. Journal of Hydrology 348:279-291.
Foster, G. M., J. L. Graham, and L. R. and King. 2019. Spatial and temporal variability of harmful algal blooms in Milford Lake, Kansas, May through November 2016: U.S. Geological Survey Scientific Investigations Report 2018–5166. https://doi.org/10.3133/sir20185166.
Francy, D. S., A. M. Brady, C. D. Ecker, J. L. Graham, E. A. Stelzer, P. Struffolino, … & K. A. Loftin. 2016. Estimating microcystin levels at recreational sites in western Lake Erie and Ohio. Harmful Algae 58:23-34.
Galloway, J. N., F. J. Dentener, D. G. Capone, E. W. Boyer, R. W. Howarth, S. P. Seitzinger, G. P. Asner, C. Cleveland, P. Green, E. Holland, D. M. Karl, A. F. Michaels, J. H. Porter, A. Townsend and C. Vörösmarty. 2004. Nitrogen cycles: past, present and future. Biogeochemistry 70(2):153-226.
García, A. M., R. B. Alexander, J. G. Arnold, L. Norfleet, M. J. White, D. M. Robertson, and G. Schwarz. 2016. Regional effects of agricultural conservation practices on nutrient transport in the Upper Mississippi River Basin. Environmental Science and Technology 50(13):6991-7000.
Gelda, R. 2019. Reservoir Water Quality Modeling. Presentation at Meeting 3 of NASEM Committee to Review the NYC DEP Watershed Protection Program. West Harrison, NY. February 4-6.
Gelda, R. K., S. W. Effler, A. R. Prestigiacomo, F. Peng, A. J. P. Effler, B. A. Wagner, M. G. Perkins, D. M. O’Donnell, S. M. O’Donnell and D. C. Pierson. 2013. Characterizations and modeling of turbidity in a water supply reservoir following an extreme runoff event. Inland Waters 3(3):377-390.
Graham, J. L., G. M. Foster, T. J. Williams, A. R. Kramer, and T. D. Harris. 2017. Occurrence of cyanobacteria, microcys- tin, and taste-and-odor compounds in Cheney Reservoir, Kansas, 2001–16. U.S. Geological Survey Scientific Investigations Report 2017–5016. https://doi.org/10.3133/sir20175016.
Haith, D. A. 1985. An event-based procedure for estimating monthly sediment yields. Transactions of the American Society of Agricultural Engineers 28(6):1916-1920.
Haith, D. A., and L. L. Shoemaker. 1987. Generalized Watershed Loading Functions for stream flow nutrients. Water Resources Bulletin 23(3):471-478.
Hamrick, J. M. 1992. A three-dimensional environmental fluid dynamics computer code: Theoretical and computational aspects. Special Report No. 317. The College of William and Mary, Virginia Institute of Marine Science, VA.
Hipsey, M. R. 2014. Computational Aquatic Ecosystem Dynamics Model: CAEDYM. V.3.3 User Guide. Centre for Water Research, Univ. of Western Australia.
Hipsey, M. R., L. C. Bruce and D. P. Hamilton. 2013. Aquatic Ecodynamics (AED) Model Library: Science Manual. Draft v4. Univ. of Western Australia.
Hipsey, M. R., L. C. Bruce and D. P. Hamilton. 2014. GLM – General Lake Model: Model Overview and User Information. V. 2.0β. Univ. of Western Australia.
Hirsch, R. M., and L. A. De Cicco. 2015. User Guide to Exploration and Graphics for RivEr Trends (EGRET) and dataRetrieval: R Packages for Hydrologic Data. USGS, Techniques and Methods, No. 4-A10. https://doi.org/10.3133/tm4A10.
Hirsch, R. M., D. L. Moyer, and S. A Archfield. 2010. Weighted Regressions on Time, Discharge, and Season (WRTDS), with an application to Chesapeake Bay River Inputs1. Journal of the American Water Resources Association 46(5):857–880.
Hirsch, R. M., J. R. Slack, and R. A. Smith. 1982. Techniques of trend analysis for monthly water quality data. Water Resources Research 18(1):107-121.
Hoang, L., E. M. Schneiderman, R. Mukundan, K. E. Moore, E. M. Owens, and T. S. Steenhuis. 2017. Predicting saturation-excess runoff distribution with a lumped hillslope model: SWAT-HS. Hydrological Processes 31(12):2226-2243.
Hoang, L., R. Mukundan, K. E. Moore, E. M. Owens, and T. S. Steenhuis. 2019. Phosphorus reduction in the New York City water supply system: A water quality success story confirmed with data and modeling. Ecological Engineering 135:75-88.https://doi.org/10.1016/j.ecoleng.2019.04.029.
Hodges, B., and C. Dallimore. 2014. Estuary, Lake and Coastal Ocean Model: ELCOM. v.3.0 User Manual. Centre for Water Research, Univ. of Western Australia.
Hodges, B., and C. Dallimore. 2016. Aquatic Ecosystem Model: AEM3D. v.1.0 User Manual. HydroNumerics, Docklands, Victoria, Australia.
Hong, B., Swaney, D., and Howarth, R. 2011. A toolbox for calculating net anthropogenic nitrogen inputs (NANI). Environmental Modelling and Software. 26(5):551-688.
Hong, B., and D. Swaney. 2013. GWLFXL 2.1.1 User’s Manual. Cornell University. https://www.researchgate.net/publication/263357315_GWLFXL_211_Users%27_Manual_Generalized_Watershed_Loading_ Functions_for_Excel_v_211.
Hong, B., D. P. Swaney, M. McCrackin, A. Svanback, C. Humborg, B. Gustafsson, A. Yershova, and A. Pakhomau. 2017. Advances in NANI and NAPI accounting for the Baltic drainage basin: spatial and temporal trends and relationships to watershed TN and TP fluxes. Biogeochemistry 133(3):245-261.
Hoos, A. B., R. B. Moore, A. M. Garcia, G. B. Noe, S. E. Terziotti, C. M. Johnston, and R. L. Dennis. 2013. Simulating stream transport of nutrients in the eastern United States, 2002, using a spatially-referenced regression model and 1: 100,000-scale hydrography. U.S. Geological Survey Scientific Investigations Report 2013-5102. https://pubs.usgs.gov/sir/2013/5102.
Howarth, R. W., G. Billen, D. Swaney, A. Townsend, N. Jaworski, K. Lajtha, J.A. Downing, R. Elmgren, N. Caraco, T. Jordan, F. Berendse, J. Freney, V. Kudeyarov, P. Murdoch, and Z. Zhao-Liang. 1996. Regional nitrogen budgets and riverine N and P fluxes for the drainages to the North Atlantic Ocean: Natural and human influences. Biogeochemistry 35:75–139.
Howarth, R. W., D. P. Swaney, G. Billen, J. Garnier, B. Hong, C. Humborg, P. Johnes, C.-M. Mörth, and R. M. Marino. 2012. Nitrogen fluxes from large watersheds to coastal ecosystems controlled by net anthropogenic nitrogen inputs and climate. Frontiers in Ecology and the Environment 10(1):37–43.
Jarvie, H. P., L. T. Johnson, A. N. Sharpley, D. R. Smith, D. B. Baker, T. W. Bruulsema, and R. Confesor. 2017. Increased soluble phosphorus loads to Lake Erie: Unintended consequences of conservation practices. Journal of Environmental Quality 45(1):123-132.
Kleinman, P. J. A. 2017. The persistent environmental relevance of soil phosphorus sorption saturation. Current Pollution Reports 3:141–150.
Kleinman, P. J., A. N. Sharpley, R. W. McDowell, D. N. Flaten, A. R. Buda, L. Tao, L. Bergstrom, and Q. Zhu. 2011. Managing agricultural phosphorus for water quality protection: Principles for progress. Plant Soil 349(1−2):169−182.
Kleinman, P. J., R. M. Fanelli, R. M. Hirsch, A. R. Buda, Z. M. Easton, L. A. Wainger, C. Brosch, M. Lowenfish, A. S. Collick, A. Shirmohammadi, K. Boomer, J. A. Hubbart, R. B. Bryant, and G. W. Shenk. 2019. Phosphorus and the Chesapeake Bay: Lingering issues and emerging concerns for agriculture. Journal of Environmental Quality 48(5):1191-1203.
Kruger, T., H. Kusamaatmaja, A. Kuzmin, O. Shardt, G. Silva and E. M. Viggen. 2017. The Lattice Boltzmann Method: Principles and Practices. Switzerland: Springer International Publishing.
Landers, M. N., T. D. Straub, M. S. Wood, and M. M. Domanski. 2016. Sediment acoustic index method for computing continuous suspended-sediment concentrations. U.S. Geological Survey Techniques and Methods, Book 3, Chapter C5. http://dx.doi.org/10.3133/tm3C5.
Lee, C. J., R. M. Hirsch, and C. G. Crawford. 2019. An evaluation of methods for computing annual water-quality loads. U.S. Geological Survey Scientific Investigations Report 2019-5084. https://doi.org/10.3133/sir2019508.
Lee, C. J., R. M. Hirsch, G. E. Schwarz, D. J. Holtschlag, S. D. Preston, C. G. Crawford, and A. V. Vecchia. 2016. An evaluation of methods for estimating decadal stream loads. Journal of Hydrology 542:185–203.
Meals, D. W., S. A. Dressing, and T. E. Davenport. 2010. Lag time in water quality response to best management practices: A review. Journal of Environmental Quality 39(1):85.
Miller, M. P., P. D. Capel, A. M. García, and S. W. Ator. 2020. Response of nitrogen loading to the Chesapeake Bay to source reduction and land use change scenarios: A SPARROW‐informed Analysis. Journal of the American Water Resources Association 56(1):100-112.
Moorman, M. C., A. B. Hoos, S. B. Bricker, R. B. Moore, A. M. García, and S. W. Ator. 2014. Nutrient Load Summaries for Major Lakes and Estuaries of the Eastern United States, 2002. U.S. Geological Survey Data Series 820. https://pubs.usgs.gov/ds/0820.
Mukundan, R., D. C. Pierson, E. M. Schneiderman, and M. S. Zion. 2015. Using detailed monitoring data to simulate spatial sediment loading in a watershed. Environmental Monitoring and Assessment 187:532. doi: 10.1007/s10661-015-4751-8.
Murphy, R. R., E. Perry, J. Harcum, and J. Keisman. 2019. A Generalized Additive Model approach to evaluating water quality: Chesapeake Bay case study. Environmental Modelling and Software 118 (2019):1-13.
NASEM (National Academy of Sciences, Engineering, and Medicine). 2018. Review of the New York City Department of Environmental Protection Operations Support Tool for Water Supply. Washington, DC: The National Academies Press.
National Research Council (NRC). 2000. Watershed Management for Potable Water Supply: Assessing the New York City Strategy. Washington, DC: The National Academies Press.
NYC DEP (New York City Department of Environmental Protection). 2006. Calibrate and Verify GWLF Models for Pepacton, Ashokan and West Branch Watersheds.
NYC DEP. 2007. Calibrate and Verify GWLF Models for Calibrate and Verify VSLF Models for Schoharie, Neversink, and Rondout Watersheds.
NYC DEP. 2011. Watershed Protection Program Summary and Assessment. March. NYC Bureau of Water Supply.
NYC DEP. 2016a. Watershed Water Quality Monitoring Plan. May.
NYC DEP. 2016b. Multi-Tiered Water Quality Modeling Program Annual Status Report. March.
NYC DEP. 2018. The 2017 Watershed Water Quality Annual Report.
NYC DEP. 2019. The 2018 Watershed Water Quality Annual Report.
NYCDEP. 2017. Multi-Tiered Water Quality Modeling Program Annual Status Report. Bureau of Water Supply.
Oelsner, G. P., and E. G. Stets. 2019. Recent trends in nutrient and sediment loading to coastal areas of the conterminous U.S.: Insights and global context. Science of the Total Environment 654:1225-1240.
Porter, J. H., A. H. Matonse and A. Frei. 2015. The New York City Operations Support Tool (OST): Managing water for millions of people in an era of changing climate and extreme hydrological events. Journal of Extreme Events 2(2):1550008.
Pradhanang, S., R. Mukundan, E. M. Schneiderman, M. Zion, A. Anandhi, D. C. Pierson, A. Frei, Z. M. Easton, D. R. Fuka, and T. S. Steenhuis. 2013. Streamflow responses to projected climate change in New York City water supply watershed. Journal of the American Water Resources Association 49(6):1308-1326.
Rasmussen, P. P., J. R. Gray, G. D. Glysson, and A. C. Ziegler. 2009. Guidelines and procedures for computing time-series suspended-sediment concentrations and loads from in-stream turbidity-sensor and streamflow data. U.S. Geological Survey Techniques and Methods Book 3, Chapter C4. Reston, VA: USGS. https://pubs.usgs.gov/tm/tm3c4/pdf/TM3C4.pdf.
Reckhow, K.H., S.C. Coffey, M.H. Henning, K. Smith, and R. Banting. 1992. Eutromod: Technical Guidance and Spreadsheet Models for Nutrient Loading and Lake Eutrophication. North American Lake Management Society. 95 pp.
Reckhow, K. H. 2014. Assessment of the Value of New Information to Inform Water Quality Management Decisions. Ken Reckhow’s Water Quality Wire. http://kreckhow.blogspot.com/2014/05/assessment-of-value-of-new-information.html.
Reeder, R., and D. Westermann. 2006. Soil Management Practices. Pp. 1-87 In: Environmental Benefits of Conservation on Cropland. The Status of Our Knowledge. M Schnepf and C. Cox (eds.). Ankeny, IA: Soil and Water Conservation Society.
Runkel, R., C. Crawford, and T. A. Cohn. 2004. Load Estimator (LOADEST): A FORTRAN Program for Estimating Constituent Loads in Streams and Rivers. Chapter A5 in U.S. Geological Survey Techniques and Methods Book 4, Chapter A5. Reston, VA: USGS. https://pubs.usgs.gov/tm/2005/tm4A5/pdf/508final.pdf.
Schneiderman, E. M., D. C. Pierson, D. G. Lounsbury, M. S. Zion. 2002. Modeling the hydrochemistry of the Cannonsville watershed with generalized watershed loading functions (GWLF). Journal of the American Water Resources Association 38(5):1323-1347.
Schneiderman, E. M., T. S. Steenhuis, D. J. Thongs, Z. M. Easton, M. S. Zion, G. F. Mendoza, M. T. Walter, and A. L. Neal. 2007. Incorporating variable source area hydrology into the curve number based Generalized Watershed Loading Function model. Hydrologic Processes 21:3420-3430.
Schwarz, G. E., A. B. Hoos, R. B. Alexander, and R. A. Smith. 2006. The SPARROW Surface Water-Quality Model—Theory, Applications and User Documentation: Chapter B3 in U.S. Geological Survey, Techniques and Methods Book 6.
Sharpley, A. N., P. J. A. Kleinman, R. W. McDowell, M. Gitau, and R. B. Bryant. 2002. Modeling phosphorus transport in agricultural watersheds: Processes and possibilities. Journal of Soil and Water Conservation 57(6):425−439.
Smith, R. A., G. E. Schwarz, and R. B. Alexander. 1997. Regional interpretation of water-quality monitoring data. Water Resources Research 33(12):2781-2798.
Soil and Water Conservation Society. 2006. Final Report from the Blue Ribbon Panel Conducting an External Review of the U.S. Department of Agriculture Conservation Effects Assessment Project. Ankeny, IA.
Stamm, C., H. P. Jarvie, and T. Scott. 2014. What’s more important for managing phosphorus: Loads, concentrations, or both? Environmental Science and Technology 48:23-24.
Strickling, H. L., and D. R. Obenour. 2018. Leveraging spatial and temporal variability to probabilistically characterize nutrient sources and export rates in a developing watershed. Water Resources Research 54:5143–5162.
Swaney, D. P., and R. W. Howarth. 2019. County, subregional, and regional phosphorus data derived from the net anthropogenic nitrogen/phosphorus (NANI/NAPI) toolbox. Data in Brief 25:104265. https://doi.org/10.1016/j.dib.2019.104265.
Swaney, D. P., B. Hong, and R. W. Howarth. 2018. NANI/NAPI Calculator Toolbox Version 3.1 Documentation. Cornell University, unpublished resource. http://www.eeb.cornell.edu/biogeo/nanc/nani/NANI%20NAPI_Calculator_Toolbox_Version_3.1_Documentation.pdf.
Tolson, T. A., and C. A. Shoemaker. 2008. Efficient prediction uncertainty approximation in the calibration of environmental simulation models. Water Resources Research 44:W04411. doi:10.1029/2007WR005869.
Tomer, M. D., and M. A. Locke. 2011. The challenge of documenting water quality benefits of conservation practices: a review of USDA-ARS’s conservation effects assessment project watershed studies. Water Science and Technology 64(1):300-310.
Tufte, E. R. 2001. The Visual Display of Quantitative Information (Vol. 2). Cheshire, CT: Graphics Press.
Ventana Systems Inc. 1999. Vensim Reference Manual Version 4. Boston, MA: Harvard.
Ward, R. C., Loftis, J. C., and G. B. McBride. 1986. The “data-rich but information-poor” syndrome in water quality monitoring. Environmental Management 10(3):291–297.
Wellen, C., G. B. Arhonditsis, T. Labencki, and D. Boyd. 2012. A Bayesian methodological framework for accommodating interannual variability of nutrient loading with the SPARROW model. Water Resources Research 48:W10505. doi:10.1029/2012WR011821.
Wells, S. A. 2018. W2 Version 4.1 Release Notes. Portland, OR: Portland State University.
White, E. D., Z. M. Easton, D. R. Fuka, A. S. Collick, E. Adgo, M. McCartney, S. B. Awulachew, Y. G. Selassie, and T. S. Steenhuis. 2011. Development and application of a physically based landscape water balance in the SWAT model. Hydrologic Processes 25:915–925.
Yang, J., P. Reichert, and K. C. Abbaspour. 2007. Bayesian uncertainty analysis in distributed hydrologic modeling: A case study in the Thur River basin (Switzerland). Water Resources Research 43:W10401, doi:10.1029/2006WR005497.
Zhang, Q., and R. M. Hirsch. 2019. River water‐quality concentration and flux estimation can be improved by accounting for serial correlation through an autoregressive model. Water Resources Research 55(11):9705-9723.
This page intentionally left blank.