8
Potential Improvements in Flash Flood Warnings

Advances in flash flood forecasting can arise from a range of improvements that are related to observing capabilities based on radar technology and modeling, data assimilation, and decision support systems. Moreover, the processing power of the Advanced Weather Interactive Processing System (AWIPS; Seguin, 2002) makes it feasible to implement a new Interactive Forecast Preparation System (IFPS; Ruth, 2002), which provides not only standard textual products but also extensive data in digital form (e.g., matrix, graphical, and model interpretation) that is accessible through the National Digital Forecast Database (NDFD; Glahn and Ruth, 2003).

Short-term forecasting of extreme weather relies heavily on the knowledge and experience of individual forecasters and their ability to select relevant information from the real-time flux of observations, followed by interpretation and diagnosis of existing conditions, subsequently leading to inference and decision making (e.g., issuing public advisories and warnings). With the complex system of observations currently in place, it is increasingly difficult for even highly experienced forecasters to process the volume of observations and model output that arrives at their workstations, and to analyze and extract the information necessary to provide effective guidance in real time. This opens a wide range of opportunities for intelligence acquisition technology and the engineering of regional expert systems to capitalize on the human intelligence at local Weather Forecast Offices (WFOs) and to make the most of artificial intelligence applications to process, classify, interpret, and synthesize inhomogeneous data.

Intelligence acquisition technology methods, such as neural networks and support vector machines—for example, working off the NDFD—can be used to extend the lead time of expert decision support system prognostics by tailoring them to the local environment and specific customer needs to increase forecast skills regionally and locally (Kuligowksi and Barros, 1998; Hall et al., 1999; Kim and Barros, 2001; Sivapragasam et al., 2001). More-



The National Academies | 500 Fifth St. N.W. | Washington, D.C. 20001
Copyright © National Academy of Sciences. All rights reserved.
Terms of Use and Privacy Statement



Below are the first 10 and last 10 pages of uncorrected machine-read text (when available) of this chapter, followed by the top 30 algorithmically extracted key phrases from the chapter as a whole.
Intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text on the opening pages of each chapter. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Do not use for reproduction, copying, pasting, or reading; exclusively for search engines.

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California 8 Potential Improvements in Flash Flood Warnings Advances in flash flood forecasting can arise from a range of improvements that are related to observing capabilities based on radar technology and modeling, data assimilation, and decision support systems. Moreover, the processing power of the Advanced Weather Interactive Processing System (AWIPS; Seguin, 2002) makes it feasible to implement a new Interactive Forecast Preparation System (IFPS; Ruth, 2002), which provides not only standard textual products but also extensive data in digital form (e.g., matrix, graphical, and model interpretation) that is accessible through the National Digital Forecast Database (NDFD; Glahn and Ruth, 2003). Short-term forecasting of extreme weather relies heavily on the knowledge and experience of individual forecasters and their ability to select relevant information from the real-time flux of observations, followed by interpretation and diagnosis of existing conditions, subsequently leading to inference and decision making (e.g., issuing public advisories and warnings). With the complex system of observations currently in place, it is increasingly difficult for even highly experienced forecasters to process the volume of observations and model output that arrives at their workstations, and to analyze and extract the information necessary to provide effective guidance in real time. This opens a wide range of opportunities for intelligence acquisition technology and the engineering of regional expert systems to capitalize on the human intelligence at local Weather Forecast Offices (WFOs) and to make the most of artificial intelligence applications to process, classify, interpret, and synthesize inhomogeneous data. Intelligence acquisition technology methods, such as neural networks and support vector machines—for example, working off the NDFD—can be used to extend the lead time of expert decision support system prognostics by tailoring them to the local environment and specific customer needs to increase forecast skills regionally and locally (Kuligowksi and Barros, 1998; Hall et al., 1999; Kim and Barros, 2001; Sivapragasam et al., 2001). More-

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California over, the implementation of such systems in a probabilistic framework allows direct incorporation of uncertainty in forecast decisions (Murphy 1977; Krzysztofowicz et al., 1993; Applequist et al., 2002). NEXRAD AND OTHER RADAR SOURCES Potential improvements to the radar hardware and operation are building on a flexible radar scan strategy and upgrades to polarimetric capability, which will assist in expanded volume coverage, data quality control, and precipitation estimation. The networking of existing information collected by a variety of operational radar sources in addition to NEXRAD will provide redundancy and offers the potential for improved horizontal and vertical coverage, including better coverage at low levels in complex terrain and offshore. Augmentation of the existing operational radar network may be required to extend coverage to important areas not properly covered or temporarily exhibiting an enhanced threat level. NEXRAD Scan Strategy Current NEXRAD scan strategies, and some of the associated limitations, were discussed in previous chapters. Coverage diagrams in Chapter 7 illustrate the low-level coverage provided by the Sulphur Mountain NEXRAD using the current scan strategies; Paris (1997a, 1997b, 1998, 2001) treated the same topic in some detail. The Paris reports considered only the altitude of the beam axis, and because the NEXRAD beam is nearly 1.0° wide, there is some detection capability below that axis (as the diagrams in Chapter 7 demonstrate). Nevertheless, the extent of coverage at and below 6000 feet above mean sea level (MSL) is limited as long as the minimum elevation angle in the scan pattern is restricted to 0.5°. It is obvious that use of a lower antenna elevation angle from an elevated radar site would provide greater low-level coverage in directions not obscured by intervening terrain. Brown et al. (2002), Wood et al. (2003a, 2003b), and others have noted advantages that could be obtained by using elevations below 0.5°, perhaps even slightly negative elevations, with NEXRADs at mountaintop sites. The discussion in Chapter 7 demonstrates the improvements in low-level coverage that could be achieved by operating the Sulphur Mountain NEXRAD at elevations down to 0°, or at slightly negative elevations.

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California The NEXRAD antennas are capable of operating at elevation angles down to about –1.0°, and scan strategies using such lower elevations have been proposed. In the original NEXRAD configuration, any effort to implement a change in scan strategy faced a major software undertaking that also would have taxed the data-handling hardware beyond the available capacity. The advent of the open-systems architecture—the open radar product generator (ORPG) already in place and the forthcoming open radar data acquisition (ORDA)—will greatly simplify the process of implementing such changes. The hardware capabilities are enhanced and the software changes needed to implement the new scan strategy become more practical. The capability to implement NEXRAD scans at a greater variety of elevation angles will soon be in place. However, any change in the NEXRAD scan strategy impacts not only the base data but also the various precipitation products and the many algorithms that operate on those data. The impact of a changed scan strategy on those products and algorithms imposes additional technical hurdles and software requirements; thorough planning and testing will be necessary before implementation of any new scan strategy. Recommendation: The National Weather Service should improve nationwide NEXRAD coverage of low-level precipitation and wind, especially for elevated radar sites in complex terrain, through the adoption of a modified scan strategy that will allow scanning at lower elevation angles. The use of lower, and perhaps even negative, elevation angles would allow monitoring of precipitation and wind at lower altitudes and, hence, would provide a more representative assessment of near-surface rainfall rates. Flexible selection of elevation angle steps would allow greater ability to avoid terrain blockage and to capture low-level meteorological phenomena. The NWS should make necessary hardware and software changes to the NEXRAD system to allow this type of modified scan strategy at the Sulphur Mountain site and other NEXRAD installations nationwide.

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California Precipitation Estimation and Data Quality Control Dual-Polarization Measurement of Rainfall Several decades of extensive research have matured polarimetric weather radar technology to a level where dual-polarization radar data can be used operationally to (1) distinguish meteorological from nonmeteorological radar echoes and thereby improve data quality compared to standard Doppler radar; (2) provide a basis for hydrometeor type classification; and (3) improve the measurement accuracy of rainfall over standard Z-R methods (e.g., Zrnic and Ryzhkov, 1999; Doviak et al., 2000; Straka et al., 2000; Bringi and Chandrasekar, 2001; Zeng et al., 2001). The planned upgrade of the NEXRAD network to dual-polarization capability will give great impetus for development of new operational algorithms, in particular, for data quality control and rainfall estimation. There is also increasing momentum for application of dual-polarized weather radars operated at shorter wavelengths (C- and X-bands), and a number of manufacturers are gearing up for, or already offering, dual-polarized Doppler weather radars in their standard product line, as well as dual-polarization upgrade kits for existing radars. Polarization-diversity radar measurements of precipitation have advanced our understanding of microphysical processes and, through this the estimation of rainfall. The most commonly used polarimetric radar parameters for rainfall estimation are the radar reflectivity factor obtained at horizontal and vertical polarization, the differential reflectivity, and the specific differential propagation phase (Bringi and Chandrasekar, 2001). In addition, the copolar correlation parameter is used extensively for data quality assessment. The differential reflectivity is a good indicator of the average raindrop size within the radar sampling volume and, thus, is used in conjunction with the radar reflectivity factor for improved rainfall rate estimation (Seliga and Bringi, 1976). The specific differential phase exhibits advantages over reflectivity-based parameters because it is a phase measurement that is immune to the absolute calibration of the radar system (e.g., Chandrasekar et al., 1990). Moreover, phase-based parameters are less sensitive to other typical radar problems such as partial beam filling and contamination by hail or terrain (Vivekanandan et al., 1999). At NEXRAD frequencies, use of the specific differential phase for rainfall estimation is particularly valuable for moderate to heavy rain rates. Because of increased level of random errors for phase measurements at lower rain rates, hybrid methods that combine rainfall estimation algorithms based on radar reflectivity, differential reflectivity, and specific differential phase are developed that blend these techniques for the

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California respective rainfall intensity range in which they are best suited (Chandrasekar and Bringi, 2004). Besides contributing to improved rainfall estimation, dual-polarization measurements are very effective for data quality control, for example, detecting anomalous propagation and ground clutter contamination. Moreover, hydrometeor classification based on dual-polarization radar observations greatly assists in detecting high-density ice particles, such as hail, and the melting layer (bright band), which is useful for vertical profile corrections. Vertical Profile Correction Radar precipitation observations are obtained some distance above the ground. The height of the observations above ground increases with increasing distance from the radar due to Earth’s curvature. This effect is substantially amplified for radars located on mountaintops. For quantitative rainfall estimation, problems arise because the vertical structure of precipitation intensity varies significantly with altitude. The varying vertical profile of reflectivity introduces uncertainty that can be the dominant source of error in radar precipitation estimates. For example, the radar beam may intersect the melting layer, resulting in a bright band contamination (yielding rainfall overestimation) or illuminate the ice phase of a precipitating cloud system aloft (resulting in a significant underestimation of rainfall). Even within the rainy portion of a storm (i.e., below the 0° level) the intensity may vary with altitude due to evaporation (intensity reduction) or further growth of precipitation particles (intensification). Taking into account the varying structure of radar reflectivity with height is very important for quantitative radar-based surface rainfall estimation. The precipitation processing system currently in place for NEXRAD (Fulton et al., 1998) does not yet incorporate vertical profile corrections. Several approaches have been developed to deal with this problem, including climatological or short-range profile corrections, range-dependent probability matching, and neural network approaches (e.g., Joss and Pittini, 1991; Joss and Lee, 1995; Amitai, 1999; Vignal et al., 2000; Seo et al., 2000; Liu et al., 2001). To some degree, all involve the vertical extrapolation of observations made aloft to the ground. The NEXRAD ORPG and ORDA developments will ease implementation of such extrapolations, and the resulting improvements in radar precipitation estimates should ultimately enhance NWS flash flood warning capabilities. Such approaches, however, may be less suitable for radars located on a mountain top, where the vertical profile cannot be observed all the way to the ground even at close ranges. Moreover, the

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California complex terrain around such sites may cause significant spatial variations in the vertical profile of reflectivity (e.g., Germann and Joss, 2002) that will be difficult to account for. Coverage of low-level precipitation may thus have to be achieved either by changing the radar scan strategy or by deployment of additional radars at lower altitudes. Estimation Uncertainty Characterization of the uncertainty of radar-based rainfall estimates should be as important as the estimates themselves. This realization, however, has not infiltrated the operational radar rainfall estimation, and it is also rarely implemented in research-mode efforts (e.g., Anagnostou et al., 1999; Steiner et al., 1999; Brandes et al., 2001, 2002). Uncertainties in radar rainfall estimates may arise from a number of sources that are related to the measurement accuracy, data quality, retrieval assumptions, and spatial and temporal representativeness of the observations (e.g., Austin, 1987). Measurement uncertainties (e.g., precision, sensor resolution and sensitivity) affect both reflectivity and phase (i.e., velocity) observations, which has implications for the achievable accuracy of radar-estimated parameters (e.g., Metcalf and Ussailis, 1984; Kostinski, 1994). Ground or anomalous propagation clutter, graupel and hail, or melting ice particles, if not properly identified and treated, may significantly increase rainfall estimates (e.g., Balakrishnan and Zrnic, 1990; Steiner and Smith, 2002). Assumptions about the particle size distribution, particle shape, and canting angle1 determine the retrieved rainfall (Chandrasekar et al., 1988; Jameson, 1989, 1991; Jameson and Kostinski, 1999; Zrnic et al., 2000; Illingworth and Blackman, 2002; Steiner et al., 2004). Moreover, the observations made at some spatial and temporal resolutions may not be representative for surface rainfall without considering the vertical and horizontal structure of precipitation and the associated wind field (e.g., Klazura et al., 1999; Jordan et al., 2003; Morin et al., 2003). Other Radar Sources Networking Existing Radars NEXRAD, particularly those located in regions of complex terrain, are hampered in providing complete radar coverage for the area due to block- 1   The canting angle is the inclination of the particle minor axis from the vertical.

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California age of the radar beam by terrain that extends above the radar horizon, as well as by the problem of the radar beam overshooting precipitation echoes below the radar horizon or within blocked sectors. For this reason, it is important that weather forecasters be able to make use of all available operational radar information in order to obtain more complete coverage of the weather within the WFO domain and of weather affecting specific areas of concern. This can be achieved by facilitating access to real-time or near-real-time data from other existing radars within the area, which may include government (e.g., Federal Aviation Administration), military, and commercial (e.g., local TV station) radars. FAA Radars The Federal Aviation Administration (FAA) has several networks of surveillance radars installed across the United States to collect radar information in real time and provide this data to the air traffic management units (TMUs) and other aviation end users. Radars deployed by the FAA include the Terminal Doppler Weather Radar (TDWR), the Airport Surveillance Radar (ASR), and the Air Route Surveillance Radar (ARSR). Four of the national FAA surveillance radar networks are shown in Figure 8.1. A brief description of the capabilities of these radars follows, and Weber (2000) provides additional details. (a) Terminal Doppler Weather Radar TDWRs are sited near major cities to provide coverage over airports of severe weather hazardous to aviation operations. In particular, TDWRs scan the airspace for detection of low-altitude wind shear, which is known to cause aircraft accidents and human fatalities (Fujita, 1981; Wilson et al., 1984). This 5-cm-wavelength radar measures both reflectivity (weather echoes) and Doppler velocity. It has a 0.5° beamwidth and thus collects higher-resolution data than the NEXRADs. Upgrades to the TDWR Radar Product Generator computer system in 2001 now enable these radars to execute complete 360° volume scans every 3 minutes (2–3 minutes faster than the NEXRADs) while still providing a surface elevation scan every 60 seconds for detection of low-level wind shear. The two lowest radar scans are collected at 0.2° and 0.5° elevation and the typical range of radar coverage is 150 km.

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California FIGURE 8.1 Locations of FAA surveillance radars. SOURCE: Mark Weber (2000). (b) Airport Surveillance Radar The FAA uses ASRs to monitor and track en route aircraft as well as the traffic near major airports. There exist a variety of ASRs (ASR-4 through ASR-11) located at airport facilities throughout the United States. Table 8.1 lists a subset of the ASR facilities located in the southwestern United States. There are twenty-nine ASR facilities operating in California alone, including two at Los Angeles International Airport and one located in Santa Barbara. The radars with the most advanced state-of-the-art capabilities are the ASR-9 and ASR-11 10-cm-wavelength radars. Although these radars operate with a

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California 5.0°-wide, vertically integrated, fan-shaped, elevation beam, they are capable of measuring and displaying radar reflectivity data. ASR-9s scan out to 111 km in range with an update frequency of 1 minute. In 2001 and 2002, 35 of the ASR-9s across the nation (see red stars in Figure 8.1) were upgraded with new Weather Systems Processors to measure Doppler wind velocity. This includes the two ASR-9s located at the Los Angeles airport. These new radar processors improve detection of weather echoes by eliminating increased ground clutter contamination that occurs under anomalous propagation conditions. The FAA plans to upgrade all of the remaining ASR-9s in the near future with the new radar processor and increase the radar range of all the radars from 111 km to 222 km. Additional ASR-11s are to be installed at smaller airports and Air Force bases across the country. (c) Air Route Surveillance Radars The ARSR-4 is a long-range surveillance radar that the FAA has sited along the perimeters of the United States. Two of these radars are located in southern California (Figure 8.1). It is a phased-array system that operates at L-band, produces 10 simultaneous 2° elevation beamwidths, and is able to scan rapidly in azimuth. The ARSR-4 collects reflectivity data, but is limited by a minimum reflectivity level of 30 dBZ and has a problem with properly representing maximum reflectivity within a storm. The FAA and the NWS are currently considering upgrading these radars with a new Doppler weather processor to provide wind velocity information and improved reflectivity data. Military Radars The Department of Defense (DoD) requirement for the NEXRAD network was that the radars provide coverage within 65 km for their priority facilities. In certain situations the selected NEXRAD sites were not able to meet the needs of both the military and other funding agencies (i.e., NWS and FAA). As a consequence, several of the military facilities across the United States have their own NEXRADs, such as Vandenberg Air Force Base (AFB) in Southern California. The DoD also operates ASRs at some of its other nonpriority bases. Table 8.1 includes the ASRs located at military facilities in California, including radars at Point Mugu and Edwards AFB.

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California TABLE 8.1 California (western Pacific region) Airport Surveillance Radar (ASR) facilities, effective 0901Z 24 February 2000 to 0901Z 20 April 2000 City/Airport HT ASR VAR Bakersfield 37 ASR-4E 14E Burbank 57 ASR-9 14E Camp Pendleton   ASR-9 14E Edwards AFB 17 ASR-5 15E Fresno   ASR-8 15E Fresno 17 ASR-4 16E Garden Grove 37 ASR-9 14E Long Beach   ASR-8 15E Long Beach   ASR-9 14E Los Angeles Intl 17 ASR-9 14E Los Angeles Intl 37 ASR-9 14E Marysville AFB 17 ASR-9 17E Merced   ASR-20 15E Monterey (Penin) 47 ASR-8 15E Mountain View   ASR-T   North Island NAS   ASR-8 13E Oakland Intl 17 ASR-9 17E Ontario Intl   ASR-8 14E Palm Springs   ASR-8 13E Palm Springs 47 ASR-5 14E Point Mugu   ASR-7 14E Riverside   ASR-5   Sacramento AFB 17 ASR-9 17E San Diego 67 ASR-9 14E San Jose (Moffett) 17 ASR-9 17E Santa Ana 37 ASR-5 15E Santa Barbara 17 ASR-8 15E Stockton 17 ASR-7 16E Stockton   ASR-11 16E SOURCE: Rita Roberts, National Center for Atmospheric Research. Commercial Radars Many television stations across the nation have purchased their own Doppler radar systems and use these data daily in their telecasts. The web site http://www.weatherexpress.com/wmradar.htm contains a list and links for local TV station weather radars in the United States. In addition to using radar data in their daily broadcasts, several television networks also provide reflectivity-based, precipitation imagery on their web sites. The KABC television network in Los Angeles operates a 5-cm-wavelength radar that is located north of Northridge, California, on Oden Mountain at an elevation

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California ARTS Latitude Longitude Ground Elevation II 35 26 27.9 119 03 35.4 510 III 34 12 14.8 118 21 44.0 740   33 17 14.1 117 19 47.4   II 34 52 21.9 117 54 41.2 2330   36 32 14.4 119 43 04.9   II 36 46 51.3 119 43 09.7 332 III 33 47 31.1 118 00 08.2 31   33 49 08.9 118 08 19.8 31   33 47 31.1 118 00 08.2   III 33 57 13.4 118 24 28.7 115 III 33 55 56.6 118 24 25.0 115 III 39 07 47.7 121 27 39.7 113   37 22 34.8 120 33 14.5   II 36 41 40.9 121 45 28.7 140   37 25 28.3 122 00 53.4     32 42 13.8 117 12 58.8   III 37 42 22.1 122 13 31.2 3   34 03 09.0 117 35 39.6     33 51 43.0 116 25 49.0   III 33 50 05.3 116 30 22.7 440   34 07 06.6 119 07 34.4     34 03 15.2 117 35 43.9   III 38 40 26.7 121 21 55.8 110 III 32 52 59.8 117 08 37.8 450 III 37 25 28.3 122 00 54.4 10 III 33 39 46.1 117 42 47.6 395 II 34 25 26.3 119 50 32.4 65 II 37 53 56.6 121 14 00.6 25   37 53 15.0 121 14 37.0   of 3802 feet. This network makes use of its radar and a mosaic of NEXRAD data (including the Sulphur Mountain NEXRAD) in their broadcasts, along with providing routinely updated, full 360° radar information in near real time on its web site. Data Access and Mosaic Radar Capabilities Following the installation of NEXRAD across the nation, commercial vendors such as WSI and Kavouras (now Meteorlogix, LLC) and private

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California Recommendation: The NWS should consider augmenting the NEXRAD network with additional short-range radars to improve observation of low-level meteorological phenomena. The NEXRAD is designed for long-range coverage with a single-beam antenna that provides coverage of a large volume of the atmosphere. Although this is adequate for many situations, NEXRAD coverage at low altitudes far away from the radar can be insufficient due to Earth’s curvature and terrain-induced blockage. Additional temporary local problems may arise, for example, as in areas affected by recent wildfires. Some of these problems can be resolved by deploying additional radars, which could be smaller, cheaper, and more easily (and even adaptively) deployable than NEXRAD. To maximize their use, these systems should be networked together, data formats standardized, and metadata established. MODELING, DATA ASSIMILATION, AND DECISION SUPPORT SYSTEMS National numerical weather prediction (NWP) models, such as the National Centers for Environmental Prediction (NCEP) ETA model, provide forecasts of precipitation rate and accumulation for the contiguous United States, and forecasters have access to these models daily. Regional-scale nonhydrostatic mesoscale NWP models may be initialized and nested within global- and national-scale models to provide higher-resolution predictions at the regional to local scale (Doyle, 1997; Mass et al., 2002; Grimit and Mass, 2002). Given the vast improvements in computing resources, the use of these models is no longer restricted to a research mode; they now can be run on workstations at WFOs in support of the forecasters (e.g., Manobianco et al., 1996). Quantitative characterization of specific states of the atmosphere and their uncertainty as a result of both model and observation errors is key to data assimilation in NWP models. In contrast, detection, classification, and space-time tracking of weather features places different requirements on data assimilation in the context of operational forecasting, which has led to the development of expert decision systems.

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California Modeling and Data Assimilation Mesoscale Modeling and Data Assimilation A relatively new weather prediction model, called the Weather Research and Forecasting (WRF) model (Michalakes et al., 2001), has been developed collaboratively by several agencies: the National Center for Atmospheric Research (NCAR), NCEP, Forecast Systems Laboratory (FSL), Center for Analysis and Prediction of Storms (CAPS), Air Force Weather Agency (AFWA), and Naval Research Laboratory (NRL). The WRF infrastructure has been designed to provide more flexibility and operational efficiency in the ability to nest mesoscale models within the national grid, to select the domain size and grid spacing used, or to run ensemble forecasts, and flexibility in the use of cumulus parameterization schemes or the explicit treatment of convection on high-resolution (e.g., 4-km) grids. Additional improvements provided by this model can be found at: http://www.wrf-model.org. National-scale precipitation rate and accumulation forecast products are available for viewing on-line at this web site daily, and forecasters at several WFOs across the United States have taken advantage of the opportunity to access and use these products. One of the most promising attributes of the WRF model is the improved prediction of mesoscale convective systems (MCSs) using explicit treatment of convection on a 4-km grid, compared to prior forecasts of MCSs that employed a cumulus parameterization scheme on a 10-km grid (Done et al., 2004). The WRF forecasts with explicit convection were found to be far superior to the 10-km-grid forecasts in their ability to forecast the MCSs that actually occurred, the number of MCSs that would occur each day, and the organization of the precipitating systems. There also is evidence that the explicit version of the model can predict finer distinctions in the precipitation structure, with features such as bow-echoes, lines of supercell storms, and the convective and stratiform regions of squall lines, than possible with some of the existing operational models. Plans are under way to run and assess the performance of the WRF model during the 2004–2005 winter season (Weisman, personal communication). It is anticipated that improved prediction of the structure, organization, and intensity of precipitating storms in the forecast models will lead to improved 12- to 24-hour outlooks for severe weather, including the potential for flash floods. Assimilation of Doppler radar data with high spatial and temporal resolution has been shown to provide improved accuracy in the representation of three-dimensional wind, thermodynamic, and microphysical fields in

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California mesoscale and cloud-scale numerical models (Verlinde and Cotton, 1993; Sun and Crook, 1997, 1998, Xu et al., 2001; Weygandt et al., 2002a, 2002b; Snyder and Zhang, 2003; Sun, 2004; Xiao et al., 2004; Zhang et al., 2004) and to improve short-term forecast skill over existing numerical models. One such system that has been run operationally at the Washington, D.C.-Baltimore, Maryland, WFO is the four-dimensional Variational Doppler Radar Analysis System (VDRAS; Sun and Crook, 2001) which assimilates single-Doppler radar data into a numerical cloud model. VDRAS provides three-dimensional wind fields used by forecasters to analyze wind shifts and surface convergence in the atmospheric boundary layer (0–3 km altitude) that are precursors to the development of convective precipitating storms (Wilson and Schreiber, 1986). A different data assimilation scheme that uses Newtonian nudging has been employed by Xu et al. (2003) to forecast winds and precipitation accumulation associated with winter storms. This system ingests NEXRAD data from several radars in the northeastern United States into a mesoscale numerical model (MM5) and is run operationally for the FAA. Techniques such as those mentioned above, which produce high-resolution wind fields that are updated every 6–12 minutes, could be a significant asset for NWS forecasters in regions such as the California coast, where wind speed and direction play a critical role in the distribution and intensity of rainfall. White et al. (2003) cited the importance of monitoring the orientation of wind speed and direction in the lowest few kilometers of the atmosphere for the onset of low-level jets that trigger upslope flow and enhanced rainfall (exceeding flash flood guidance thresholds) along the coastal mountains of Southern California. Coupled Modeling Coupled hydrological-meteorological operational systems have been developed and run for the purpose of integrating local precipitation prediction models with hydrological models on the catchment scale and for generating short-term flash-flood warnings with hourly forecasts (Georgakakos and Hudlow 1984; Georgakakos, 1986b, 1986c, 1987; Bae et al., 1995; Miller and Kim, 1996; Westrick and Mass, 2001; and Westrick et al., 2002). The greatest benefit achieved in using a coupled approach was found to be when the forecast lead time was comparable to the response time of a flash flood-prone catchment of interest (Georgakakos and Foufoula-Georgiou, 1991). Current approaches in the United States and Europe, particularly where water management and flash floods in complex terrain are of upper-

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California most concern, now incorporate the use of weather radar data to provide quantitative precipitation estimation (QPE) and numerical model-based quantitative precipitation forecasts (QPF) for hydrometeorological forecasting systems (French and Krajewski, 1994; Lee and Georgakakos, 1996; Dolcine et al., 1998; Yates et al., 2000; Johnson et al., 2001; Mecklenburg et al., 2002; Cress et al., 2002). One such system currently being tested operationally is a prototype software system that provides operational estimation and prediction of rainfall and streamflow in the mountainous subcatchments areas of the Panama Canal watershed in Panama, Central America (Georgakakos, 2002). This fully automated system incorporates information from radar reflectivity and hourly rainfall data, discharge and surface weather reports from Automated Local Evaluation in Real Time (ALERT) rain gauges, upper-air radiosonde observations, and ETA model forecasts. Use of the system fosters a close collaboration between meteorologists and hydrologists on a day-to-day basis. Verification of the short-term predictions of rainfall and streamflow has been very good. Another notable example of an operational coupled hydrometeorological-hydrologic forecast system is employed in the Pacific Northwest (Westrick and Mass, 2001; Westrick et al., 2002). The accuracy of the observation-tuned MM5-DHSVM (fifth generation Pennsylvania State University-NCAR Mesoscale Model-University of Washington Distributed-Hydrology-Soil-Vegetation Model) simulated peak streamflows exceeded that achieved by the NWS Northwest River Forecast Center (RFC) results, although the results were highly sensitive to data density and quality.2 Prediction Uncertainty Numerically predicted meteorological parameter fields, such as temperature, wind, and precipitation, exhibit uncertainty that depends on the model physics, initial conditions, and forecast lead time (e.g., Cortinas and Stensrud, 1995; Harris et al., 2001; Germann and Zawadzki, 2004; Walser et al., 2004). Estimation of the prediction uncertainty can be achieved through probabilistic ensemble modeling based on perturbed initial conditions or simulations based on varied physical parameterizations (e.g., Du et 2   The National Research Council recently released a report assessing the USGS National Streamflow Information Program (NRC, 2004). The conclusions and recommendations from this report may improve the quality of streamflow data, which in turn could lead to improvement in the development and calibration of hydrologic models.

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California al., 1997; Stensrud et al., 1999; Richardson, 2000; Krishnamurti et al., 2000; Tracton and Du, 2001; Mullen and Buizza, 2001; Ebert, 2001; Toth et al., 2003; Grimit and Mass, 2002). The range of model results provides a measure of the anticipated agreement or disagreement of the various forecasts, with smaller ranges, suggesting higher confidence in a particular forecast. Similarly, the uncertainty of hydrologic predictions based on standalone hydrologic models that use radar-estimated rainfall as input (e.g., Ogden et al., 2000; Sharif et al., 2004) or coupled meteorological-hydrologic models that may also use numerically predicted rainfall can be estimated by probabilistic means (e.g., Ganguly and Bras, 2003; Nijssen and Lettenmaier, 2004). Seo et al. (2003) present the latest NWS developments for real-time assimilation of hydrometeorological and hydrologic data into operational hydrologic forecasting. A special issue of the Journal of Hydrology (2004) is devoted to the Distributed Model Intercomparison Project, a NWS-spearheaded effort to evaluate coupled hydrometeorological-hydrologic models and assess associated uncertainties in streamflow prediction. Decision Support Systems Success in short-term forecasting of extreme weather has heretofore relied on the knowledge and experience of individual forecasters. They are able to select relevant information from the real-time flux of observations and to interpret and diagnose existing conditions, all of which lead to their inference and decision making when issuing public advisories and warnings. Inevitably, capacity-building and ensuring the continuity of forecasting expertise constitute a great challenge for the NWS. This opens a wide range of opportunities for intelligence acquisition technology and the engineering of regional expert decision support systems to capitalize on the human intelligence at local WFOs and to make the most of artificial intelligence applications to process, classify, interpret, and synthesize inhomogeneous data. Data Integration A new data integration technology became available recently to WFOs as a component of AWIPS. The Flash Flood Monitoring and Prediction (FFMP; J. A. Smith et al., 2000; Filiaggi et al., 2002) application is a radar-based Advanced Hydrologic Prediction Services (AHPS; see Box 8.1) decision support tool that evolved from the Areal Mean Basin Estimated Rainfall

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California BOX 8.1 Advanced Hydrologic Prediction Services The AHPS is a new program that was created as part of the modernization of the NWS hydrology program. AHPS is a web-based suite of forecast products that make use of modernized NWS technologies and improved meteorological and hydrological scientific understanding. The objective of the program is to provide improvements to river and flood forecasting capabilities and to improve communications between the NWS, cooperating agencies, and stakeholders to meet the diverse and evolving needs of all of these groups (NWS, 2001b). AHPS will achieve this objective by delivering several enhancements to the NWS Hydrologic Services Program (NWS, 2002b). Among these are better forecast accuracy—by incorporating advanced science into hydrologic modeling systems and coupling atmospheric and hydrologic models and forecast information on all time scales; more specific and timely information on fast-rising floods—by using tools to more rapidly identify small basins affected by heavy rainfall, identify excessive runoff locations, and predict the extent and timing of inundation; new types of forecast information—by incorporating new techniques for quantifying forecast certainty and conveying this information; and expanded outreach—by engaging partners and customers in all aspects of the hydrologic services improvement effort. The National Research Council currently is reviewing the scope of the AHPS program, and it will produce a report by mid-2005. More information about the AHPS program can be found at http://www.nws.noaa.gov/oh/ahps. (AMBER; Davis and Jendrowski, 1996) algorithm. This system assists local forecasters by monitoring precipitation accumulations at the hydrologic catchment scale to interpret a hydrologic threat within the context of an evolving meteorological situation to provide short-term, high-resolution flash flooding guidance. A prerequisite for utilization of FFMP is the topographic mapping of all the basins and subbasins covered by a WFO, including their hydrologic characteristics such as a basin’s outline and average slope, soil characteristics and conditions, foliage coverage, and channel hydraulic measures. Time-varying parameters, such as soil moisture content or infiltration capacity, may be edited and updated by forecasters based on guidance received from the RFC. The FFMP monitors initiation and movements of storms based on radar, satellite, and lightning data, and keeps track in real time of the rainfall

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California FIGURE 8.7 FFMP display showing precipitation and flooding information at the county level. SOURCE: National Weather Service. accumulation and hydrologic conditions for each of the basins. The radar-estimated rainfall over a given catchment is operationally updated and the accumulation up to that point in time is compared to the respective flash flood guidance stored for that location. FFMP entails an interactive graphical display that enables the forecaster to monitor the responsible domain (Figure 8.7) and zoom in on any catchment by mouse click to retrieve current status and flash flood guidance information (Figure 8.8). The potential for flooding is color-coded to provide a quick assessment of where the situation may reach or exceed flash flood threat levels. The high-resolution information and zoom-in capability and the automated bookkeeping of the FFMP system enable forecasters to stay abreast of potentially hazardous weather situations more easily, issue much more specific warnings, and do so with improved skills. Thus, FFMP bears the potential to greatly enhance flash flood monitoring, prediction, and warning. Although this technology is available, it has been implemented at only a few WFOs nationwide, because it requires a significant effort to digitize

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California FIGURE 8.8 FFMP display showing precipitation and flooding information at the hydrologic basin level. SOURCE: National Weather Service. the topographic and hydrologic information for each of the potentially thousands of basins and subbasins covered by a WFO. However, some agencies (e.g., county public works departments, water resource divisions) already have such topographic and hydrologic information for some of their watersheds; WFO collaboration with such agencies may somewhat lessen this burdensome task and facilitate the implementation of FFMP. Future developments of FFMP may include incorporation of numerically predicted precipitation to increase the lead time in issuing flash flood warnings, which would be most valuable for regions of complex terrain. Expert and Fusion Systems The important concept behind data fusion and coupled hydrological-meteorological systems is to quickly combine relevant features in the observational and model datasets and produce a guidance product for the forecaster to view and edit if necessary (Roberts et al., 2003). The intent is to

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California Recommendation: In addition to the original NEXRAD siting considerations, future siting of radars in complex terrain should include detailed assessments of coverage in areas at risk for flash flooding. The original NEXRAD siting procedures considered primarily meteorological processes, radar coverage, and ground clutter. Increased understanding of hydrologic processes of runoff production and streamflow response, combined with the application of radar to real-time hydrologic prediction for individual catchments (e.g., FFMP), enables incorporation of hydrologic aspects of the land surface into future siting processes. Readily available detailed digital information on topography and other relevant spatial information now make it possible to analyze a radar’s coverage of the near-ground portion of the atmosphere using geographic information systems (GIS) technology. The potential for complete and partial radar beam blockage can be evaluated in the context of hydrologic basins for which coverage is sought. Basin size, average slope, orientation with respect to the movement of dominant weather patterns, characteristics of soil, land cover and land use, and channel hydraulic aspects determine the amount of rain that is likely to cause flash flooding and where it may occur. These characteristics, together with local hazard vulnerabilities, can help determine priorities of site selection. save the forecaster significant time in assimilating all of the observational information by doing this automatically using a series of computer algorithms run every 6 to 10 minutes. This allows the forecaster more time to focus attention on the specific areas of concern for hazardous weather and to provide timely dissemination of warnings to the public. Most forecasters generally have high confidence in their ability to identify any threatening weather in the data. However, a human impacts study conducted by Anderson-Berry et al. (2004) on forecaster performance during the Sydney 2000 Olympics games showed that forecasters felt they could produce even better forecasts with access to more advanced technology and the automation of some products in order to give them more time to study the weather situation and local conditions. Furthermore, the Sydney forecasters felt that their warnings were superior to those they would have issued using routine storm tracking techniques. Thus, to advance the potential for providing improvements in flash flood forecasting, consideration should be given to employ-

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California ing the use of data assimilation and fusion systems in WFOs in the near future. Elements of expert decision support systems already are present in the training modules available at WFOs and nationwide through AWIPS and IFPS user interface systems, which facilitate visualization and integration of multiple types of data (Meiggs et al., 1998; Ruth et al., 1998). The training modules consist of a limited number of cases at each WFO (e.g., the committee observed cases of heavy precipitation and flooding in Ventura County while at the Lox Angeles-Oxnard WFO), and the analysis software works on a sophisticated graphical display and analysis tool on a user prompt basis. The ability to retrieve historical case studies and to generate localized probabilistic products with 0- to 12-hour lead times, which will help stakeholders better understand uncertainty, would improve the quality and reliability of the support provided by IFPS to operational forecasters. This will continue to be important as FFMP and other expert decision support systems become widely available (Mass, 2003). In addition to data assimilation methodologies, data fusion systems are being run operationally that combine different observational datasets together to produce short-term, 0- to 2-hour forecasts (or nowcasts) of precipitation rate and precipitation accumulation. One component of data fusion systems is the sophisticated feature detection and tracking algorithms that run on the radar (Dixon and Wiener, 1993; Wolfson et al. 1998; Germann and Zawadzki, 2002; Seed, 2003) and satellite (Evans and Shemo, 1996; Nesbitt et al., 2000; Roberts and Rutledge, 2003) data to monitor storm growth, evolution, and motion. NCAR runs an automated convective-storm nowcast system for the FAA, called the Auto-Nowcast System (ANC; Mueller et al., 2003), which produces 0- to 1-hour precipitation rate nowcasts every 6 minutes. In addition to nowcasting storm growth, evolution and decay, the ANC produces time- and location-specific nowcasts of storm initiation through the use of (1) Doppler radar to detect wind shifts or convergence boundaries at low altitudes; (2) satellite information to monitor the growth of cumulus clouds; (3) the VDRAS cloud model to obtain vertical motions in the boundary layer; and (4) national numerical model output to delineate regions of atmospheric instability (Saxen et al. 2004). Through the use of fuzzy logic (McNeill and Freiberger, 1993), the ANC is able to apply confidence values and add together the most important features detected in the observations to produce short-term convective storm nowcasts. Additional systems that are operational include the National Convective Weather Forecast (NCWF; Megenhardt et al., 2000, 2004) system, which incorporates radar, lightning, and numerical model data, to produce

OCR for page 117
Flash Flood Forecasting Over Complex Terrain: With an Assessment of the Sulphur Mountain NEXRAD in Southern California 0- to 2-hour convective storm nowcasts and is currently in use at the Aviation Weather Center in Kansas City. Forecasters use this regularly updated (every 10 minutes), automated weather product as guidance in anticipating enroute aviation weather hazards. The United Kingdom Meteorology Office has run a system called Nimrod (Golding, 1998) operationally for several years. This system, which combines radar and satellite information with numerical model output, produces 0- to 6-hour precipitation rate and accumulation forecasts every 10 minutes for both meteorological and hydrological purposes; accurate warning for flash floods are a primary goal in the use of this tool. Recommendation: To increase the accuracy and lead time of flash flood forecasts and warnings, the NWS should continue to implement new technologies and techniques including (a) the Flash Flood Monitoring and Prediction program at all Weather Forecast Offices, (b) polarimetric modifications to NEXRAD, (c) data assimilation systems that integrate radar and other operational datasets into coupled hydrometeorological and hydrological models, and (d) data fusion systems. Extensive opportunities exist for forecasters to take advantage of the rapid advancement of technology to improve forecasts, watches, and warnings. The FFMP system, which requires adaptation to the specific watersheds served by each WFO, would facilitate more specific flash flood warnings. In addition, as part of its new Advanced Hydrologic Prediction Services, the NWS is encouraged to continue its effort to develop and evaluate hydrologic and coupled meteorological-hydrologic models to advance technologies useful for improved flash flood guidance and warnings. The polarimetric modification would improve the data quality and quantitative precipitation measurement capabilities of NEXRAD. Real-time data assimilation systems that incorporate observations into high-resolution mesoscale numerical models provide rapidly updated wind and precipitation forecasts. Data fusion systems, such as the Auto-Nowcast system, incorporate all available observation datasets together with numerical model output to produce very short range (0- to 2-hour), site-specific forecasts. These advances can produce improved forecasts, including ensemble and probabilistic forecasts, of precipitation rate and accumulation that are essential for flash flood forecasting. To enhance the usefulness of the forecast, quantification of uncertainty (e.g., probability forecasts) should be an integral component of these products (NRC, 2003).