Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
3 Scientific and Technical Aspects of AHPS This chapter discusses the major scientific and technical aspects of the Advanced HydrologicPrediction Service (AHPS) as they relate to the following four scientific and technical goals ofAHPS: to produce more accurate products by incorporating advanced hydrologic science into the National WeatherService (NWS) model; to provide products with forecast horizons two weeks or further into the future; to create moreinformation that is useful to assess risk to flooding; and to provide more specific and timely information on fast-risingfloods with increased lead time. The chapter concentrates on the current models and techniques used byAHPS, with an emphasis on the NWS River Forecast System (NWSRFS), and it discusseslimitations, research needs, and options to update those techniques so that AHPS can provide the"basic," "enhanced," and "partnered" hydrologic products that it promises. Major elements of thestatement of task are addressed in this chapter, including discussions about operational floodforecasting and the overall strategy of AHPS to meet these needs, applying modern hydrology andmodeling techniques and technologies to enhance hydrologic predictions, and an assessment of theresearch needs, priorities, and application of research into AHPS operations. The chapter openswith a description of precipitation inputs to hydrologic models. The next section discussesNWSRFS, its limitations, and its areas that need updating to achieve AHPS goals. A section onflash-flood guidance closes the chapter. PRECIPITATION INPUTS TO HYDROLOGIC MODELS Precipitation inputs are used in hydrologic runoff and snow-melt models to generateestimates of rates and stages of streamflow. AHPS is predicated on these hydrologic models;therefore, hydrometeorological inputs, generally, and precipitation inputs, specifically, stronglyinfluence AHPS hydrologic forecasts. AHPS hydrometeorological inputs consist of quantitativeprecipitation estimations (QPEs), satellite-based precipitation measurements, and quantitativeprecipitation forecasts (QPFs). The skill ("skill" is used here as it is used by meteorologicalcommunity to mean the accuracy of a forecast) of NWSRFS hydrologic products is largelydependent on the accuracy of QPE and skill of QPF. Quantitative Precipitation Estimations QPEs come from rain gages or a combination of gages and radar estimates. Historically,precipitation analysis operations have been based on interpolation of gage observations to meanareal precipitation within individual hydrologic basins. Certain functions of the radar-basedprecipitation processing system are controlled from Weather Forecast Offices (WFOs). Finalcontrol of NWS radar operation and the choice of adaptable parameters for most processingalgorithms reside with WFO staff. These operations include decisions on the Z-R relationships(continental convective vs. tropical) used in processing reflectivity data from individual radars, andliaison with other agencies that maintain gages. During major tropical storm rain events, WFOs andRiver Forecast Centers (RFCs) consult on the choice of a Z-R relationship for radars within theirareas of responsibility. 31
32 Toward a New Advanced Hydrologic Prediction Service (AHPS) Recognizing the importance of accurate QPEs as input to hydrologic models, the NWS isdeveloping AHPS techniques for multisensor precipitation estimation that include the use of thenational network of Doppler weather radars (Seo and Breidenbach, 2002) and satellite and lightningdata (Kondragunta, 2002). The NWS is also working towards quantifying the uncertainty of QPEs(McEnery et al., 2005). These activities are positive; however, better documentation is needed toevaluate the effectiveness of these efforts. Publication and dissemination of AHPS activity allow theacademic community and others to learn about and participate in algorithm development,verification and uncertainty estimation. Publication, peer review, and information dissemination willhelp the continual advancement of hydrologic science in AHPS models. AHPS researchersshould periodically publish progress of the development of and improvements toprecipitation products. Satellite-Based Precipitation Estimations Researchers at the National Oceanic and Atmospheric Administration (NOAA)1, theNational Aeronautics and Space Administration (NASA)2, and some universities3 have maderemarkable progress in the development of satellite-based precipitation estimation algorithms. Thisnew generation of algorithms is capable of merging and blending multiple types of observedinformation from both geostationary and low polar-orbiting satellites, and they generate estimatesfor precipitation at various spatial and temporal scales (Hong et al., 2004; Joyce et al., 2004). Theseresearch efforts progressively improve precipitation estimation over regions with limited ground-based observations. They also show much promise to improve coverage over mountainous terrains,especially in the western U.S., where gage and radar coverage are very sparse. Recent research(Yilmaz et al., 2005) shows encouraging results with respect to the use of high-resolution satellite-based precipitation as input to a hydrologic model. Like precipitation estimation algorithms, plans for a new generation of satellite systems areunderway. NASA, with a group of international partners, is developing a constellation of satellitesystems called the Global Precipitation Measurement4 for launch around 2010 that is capable ofproducing global coverage of precipitation every three hours. AHPS developers are stronglyencouraged to work closely with satellite precipitation groups (NASA, NOAA, and those inthe academic community) to ensure that AHPS hydrologic requirements for precipitationare included in the Global Precipitation Measurement mission. Quantitative Precipitation Forecasts A QPF is a prediction of the amount of precipitation that will fall at a given location in agiven time interval. QPFs are issued routinely by the NWS as a part of meteorological forecasts.Intuitively, QPFs would be useful in producing hydrologic forecasts, but there is no strong evidencethat QPFs are being used that way to extend flood and streamflow predictions. There may beseveral reasons why hydrologists do not use QPFs extensively. One reason could be that typicalQPFs provide values averaged in 6-hour aggregations or blocks. Finer temporal scales would be1http://www.cpc.ncep.noaa.gov/products/janowiak/cmorph.html.2http://trmm.gsfc.nasa.gov/publications_dir/precipitation_msg.html.3http://hydis8.eng.uci.edu/CCS.4http://gpm.gsfc.nasa.gov.
Scientific and Technical Aspects of AHPS 33more useful to AHPS forecasts for fast-rising flood waters. The standard, 6-hour QPFs show greatvariation in accuracy in basins with complex topography or shorter hydrologic response times. Another reason that hydrologists may not use QPF extensively is that QPFs were developedfor meteorological, not hydrologic, purposes, so verification aspects of accuracy and performanceare neither consistent nor calibrated with other hydrologic models. The skill of QPF for hydrologicforecasting is relatively unknown and results from tests and applications of QPF to date have fallenshort of convincing many hydrologists of their operational value. For these and maybe otherreasons, QPE and QPF remain underutilized in generating hydrologic forecasts. AHPS could serve as a vehicle to connect the meteorologists who generate QPF to thehydrologists who have yet to embrace it (Droegemeier et al., 2000). In order to be useful inhydrologic products, QPE and QPF need systematic evaluation and verification for hydrologicapplications. Systematic evaluation and verification would guide further development andrefinement of the hydrometeorological QPE and QPF at the National Centers for EnvironmentalPrediction, across the NWS, and potentially for use in AHPS. To ensure that QPE and QPF meetthe nation's needs and needs of hydrologic forecasters, the NWS should strengthen QPE andQPF for hydrologic prediction through an end- to-end evaluation that assesses QPE/QPFquality and impacts on flood and streamflow products for basins of diverse size andtopography. THE NWS RIVER FORECAST SYSTEM "Basic" services will upgrade static, diagnostic river gage hydrographs5 to show AHPS riverforecasts in a range of predicted information, including forecasted river levels6, weekly flowprobabilities7, and monthly flow probabilities8. Other hydrologic products and services are alsoenvisioned, such as "enhanced" services of flash-flood guidance information presented graphically9,and "partnered" services, like an internet-based flood map service that uses geographic informationsystems (GIS; Figure 3-1). These, and all AHPS products and services, rely on the current NWSprimary modeling engine, the NWSRFS. NWSRFS was developed in the 1970s and 1980s (NWS, 1972) in a modular framework, andit is based on FORTRAN code. Over the intervening years, it has been incrementally upgraded withnew models, functions, and displays (McEnery et al., 2005). However, more updates are needed to:incorporate advanced hydrologic science into NWS models; provide forecasts for two weeks or further into the future;create information that is useful to assess risk to flooding; and provide other products and services promisedby AHPS. NWSRFS Modules The basic structure of the NWSRFS includes a calibration system (CS), an operationalforecast system (OFS), an interactive forecast program (IFP), and an ensemble streamflow5http://www.crh.noaa.gov/ahps2/hydrograph.php?wfo=fgf&gage=lkbm5&view=1,1,1,1,1,1.6http://newweb.erh.noaa.gov/ahps2/hydrograph.php?wfo=aly&gage=wtfn6&view=1,1,1,1,1,1.7http://newweb.erh.noaa.gov/ahps2/weekly.php?wfo=aly&gage=wtfn6&view=1,1,1,1,1,1.8http://newweb.erh.noaa.gov/ahps2/period.php?wfo=aly&gage=wtfn6&view=1,1,1,1,1,1.9http://www.cnrfc.noaa.gov/flashFloodGuidance.php?cwa=RSA&hour=1.
34 Toward a New Advanced Hydrologic Prediction Service (AHPS)FIGURE 3-1 Flood Internet Map Service prototype for Lewiston, PA.SOURCE: http://www.nws.noaa.gov/oh/ahps/floodims_news.html.prediction (ESP) system (Figure 3-2). The hydrologic model(s) is calibrated for basin-specificconditions in the CS; the calibrated model output is used to develop forecasts in the OFS, IFP, andESP. Generating a hydrologic forecast using the NWSRFS involves a multi-step process that beginsin the OFS. First, precipitation information is collected and analyzed. Next, the appropriatehydrologic model(s) are applied depending on whether the forecast area is due for rain or snow, andthe precipitation data are input into the model. The IFP is then used to analyze the short-termstreamflow forecasts generated by the OFS and to make system adjustments to improve the modelsimulations and forecast. Model states from the OFS are used by the ESP component, inconjunction with calibrated models and historical data, for the generation of longer-termprobabilistic predictions. The functions, limitations, and research needs of each of the NWSRFScomponents are described in the following sections.Functions of NWSRFS Hydrologic Models NWSRFS allows hydrologists to combine the appropriate models in a manner that isdescriptive of the basin, the available data, and the forecast products desired. NWSRFS hydrologicapplications include conceptual rainfall-runoff models, snow, and the Antecedent PrecipitationIndex (API) model. The Sacramento Soil Moisture Accounting model (SAC-SMA; Burnash et al., 1973) is theprimary conceptual rainfall-runoff model used at the RFCs. The main snow model is the Snow-17model originally developed by Anderson (1968; 1973; 1976). Snow-17 is a temperature-indexversion of a full energy budget model. SAC-SMA and Snow-17 are lumped conceptual models thatconvert frozen and liquid precipitation into runoff. Conceptual models do not explicitly represent
Scientific and Technical Aspects of AHPS 35 NWSRFS Components Calibration System(CS) EnsembleStreamflow Prediction (ESP) System Calibration Analysis (Hydrologic Hydrologic window Historical and Analysis and Data Hydraulic Hydraulic Models) Models time Operational Forecast System (OFS) Hydrologic and Statistical Hydraulic Models Analyses Real-Time Analysis and Data Observed short term Assimilation Probabilistic and forecasts Short term to Forecast Extended Data current states Interactive Forecast Interactive Adjustments Program (IFP)FIGURE 3-2 Main components of the NWSRFS.SOURCE: Adapted from McEnery et al., 2005.the measurable physical characteristics or processes of a basin, and are therefore limited to forecastlocations equipped with river stage observations for calibration purposes. For this reason,conceptual models are typically run in lumped mode, compiling information over coarse spatial areasor hom*ogenous conditions (i.e., similar elevation or topography). SAC-SMA and Snow-17 aresometimes used in a semi-distributed capacity, where a diverse basin is subdivided into smaller, morehom*ogenous areas, so that the lumped models are better able to describe the dominant hydrologicalprocesses within each sub-basin area (McEnery et al., 2005). In some RFCs, the API model is used.The API empirical method estimates the amount of surface runoff that will occur in a basin from agiven rainstorm based on an index of moisture stored within a drainage basin before a storm,physical characteristics of the basin, time of year, storm duration, rainfall amount, and rainfallintensity.Limitations of and Research Needs for NWSRFS Hydrologic Models The current lumped conceptual hydrologic models used by the NWS and in the NWSRFSare functional and relatively accurate (Reed et al., 2004), but have some limitations associated withtheir use. One limitation is that these models are based on the original empirical, lumped waterbalance accounting procedure and the FORTRAN computer coding standards of 20-30 years ago.The Office of Hydrologic Development's Hydrology Laboratory (HL) recognizes that thehydrologic modeling approaches used for AHPS products need to be updated from the currentNWSRFS. HL has started making some modifications, such as addressing issues of modeling at
36 Toward a New Advanced Hydrologic Prediction Service (AHPS)finer spatial and temporal scales and incorporating more physically based process equations in futureversions of the models (Koren et al., 2004; Smith et al., 2004a, b). Model resolution has been at the center of a debate in the hydrologic community about theadvantages/disadvantages of lumped versus distributed models (Figure 3-3). One side of the debatenotes that lumped models create coarse but accurate results, even though they do not effectivelyrepresent spatial variability of hydrologic processes, or intra-basin differences in elevation or terrain.Distributed models are designed to work at spatial and temporal scales finer than lumped models.The other side of the debate, the argument in favor of distributed models, posits that becausedistributed models can account for differences in site specific characteristics, including basin size,topography, land cover, they are more appropriate for AHPS products and services (McEnery et al.,2005). Fueling both of these arguments are recent research efforts that focus on downscaling andimproving spatial and temporal resolution of hydrologic models (Reed et al., 2004; Zhang et al.,2001). The selection of lumped or distributed models for AHPS products is non-trivial because thetype of hydrologic model(s) used in AHPS development will strongly impact AHPS products andservices. AHPS currently uses lumped models, but the NWS is keenly aware of some of the benefitsthat distributed models may bring to AHPS. Therefore, the NWS is considering (1) making aswitch from lumped models to distributed modeling for AHPS products; and (2) whether a singledistributed model or a suite of distributed models will best achieve AHPS goals and purposes. TheNWS must also determine how it will reconcile the incompatibility of the existing NWSRFSsoftware structure with distributed modeling applications. To help address whether to switch to distributed models from the current lumped models,HL launched the Distributed Model Intercomparison Project (DMIP; Smith et al., 2004c) to guideAHPS' future distributed modeling research and applications. HL invited researchers from theacademic and non-academic communities to participate in the DMIP project. DMIP Phase I hasbeen completed and its results are summarized in Box 3-1; DMIP Phase II is being planned toaddress complex basin issues of snow and orography. The experimental design of DMIP Phase I was based on the comparison of distributedmodels applied to a common set of test data. Model simulations were compared to observedstreamflow data as well as simulations generated from a lumped application of SAC-SMA. Resultsof DMIP Phase I have been published as a series in a special issue of Journal of Hydrology in 2004(Box 3-1). Perhaps the first finding is the most critical and challenging for AHPS (Finding 1, Box 3-1): overall, the lumped hydrologic models performed better than, or slightly inferior to, a wellcalibrated distributed model. The NWS initiated DMIP to assist with the distributed modelingchoice for AHPS, but DMIP Phase I research results do not clearly delineate whether AHPSproducts should use lumped or distributed models. The NWS has expended resources for research on the issue of distributed models andcomparison studies and now needs to make clear how the final model choice(s) will be made. Adecision-making framework that will be used to select the next hydrologic model(s) is as importantas the DMIP research efforts. A decision-making process should establish a template for the HL toselect its model(s) and methods that standardize the mechanisms that will be used across RFCs andWFOs as they adjust hydrologic models to local conditions. An advisory group comprised ofexperts from outside of an internal to the NWS could help NWS develop this framework and guideits implementation across the NWS. From NWS personnel and written materials, it is difficult to
Scientific and Technical Aspects of AHPS 37 Hydrologic Modeling Approaches 1. Rainfall, properties 1. Rainfall, properties in averaged over basin each grid 2. One rainfall/runoff model 2. Rainfall/runoff model in each grid 3. Prediction at only one 3. Prediction at many points pointFIGURE 3-3 Differences between lumped and distributed model approaches.SOURCE: Adapted from Smith et al., 2004a.ascertain the mechanism used to guide AHPS model selection and implementation. DMIP has beena valuable effort to compare various models and has identified additional research questions thatneed to be addressed to provide a robust suite of AHPS models, but it is not clear how or whenDMIP will converge to an ultimate decision about AHPS model choice(s). DMIP Phase II, likeDMIP Phase I, has no stated strategy that outlines steps from DMIP results to the selection of thenext generation of model(s) for AHPS. Therefore, the NWS should strengthen connectionsbetween DMIP Phase I/DMIP Phase II and AHPS goals. The NWS should clarify thecriteria and decision-making process for selecting the next generation of hydrologicmodel(s) for AHPS, using an advisory group that involves modeling experts from inside andoutside of the NWS to ensure that the state-of-the-art modeling advances are incorporatedobjectively into NWSRFS. Another limitation for DMIP and similar exercises is the lag time between research andimplementation into AHPS operations that may be too long to be effective. Protracted interveningtime may inhibit AHPS developers from fully exploiting new modeling capabilities and achieve theAHPS goal of producing advanced hydrologic products. Finally, there is a general observation about the mixed level of documentation of hydrologicmodels used in NWSRFS. While advancements and modifications to the SAC-SMA and NWSRFShave taken place over several decades and have been reported from time to time in conferences,proceedings, and in peer-reviewed journal papers, there needs to be more publication anddocumentation of the internal activities related to model development and decision making. TheNWS has been proactive in publishing its work on some modeling research and development efforts(distributed modeling, cold seasons, and model calibration) but less in other areas (ESP, verification,etc.). A good example of documenting updates to hydrologic models is the U.S. Army Corps of
38 Toward a New Advanced Hydrologic Prediction Service (AHPS) BOX 3-1 Summary of DMIP Phase I Findings 1. Although the lumped model outperformed distributed models in more cases than distributed models outperformed the lumped model, some calibrated distributed models can perform at a level comparable to or better than a calibrated lumped model (the current operational standard). The wide range of accuracies among model results suggest that factors such as model formulation, parameterization, and the skill of the modeler can have a bigger impact on simulation accuracy than simply whether or not the model is lumped or distributed. 2. Clear gains in distributed model performance can be achieved through some type of model calibration. On average, calibrated models outperformed uncalibrated models during both the calibration and validation periods. 3. Gains from applying a distributed simulation model at NWS forecast basin scales (on the order of 1,000 km2) will depend on the basin characteristics. 4. The Christie basin is a small basin nested in the Eldon Basin, and is distinguishable in the DMIP study because of its small size. Christie, compared with larger basins, showed improved calibrated, peak flow results likely because the lumped "calibrated" model parameters (from the parent basin calibration, Eldon) are scale dependent and distributed model parameters that account for spatial variability within Eldon are less scale dependent. The Christie results indicate that more studies on small, nested basins are needed to confirm and better understand these results. 5. Among calibrated results, models that combine techniques of conceptual rainfall-runoff and physically-based distributed routing consistently showed the best performance in all but the smallest basin. Gains from calibration indicate that determining reasonable a priori parameters directly from physical characteristics of a watershed is generally a more difficult problem than defining reasonable parameters for a conceptual lumped model through calibration. SOURCE: Reed et al., 2004.Engineers Hydrologic Engineering Center's series (see Box 4-1 in Chapter 4). The NWS needs toprovide stronger documentation to allow the research community to learn about andcontribute to AHPS research and development.Function of the NWSRFS Calibration System The NWSRFS hydrologic models use hydrometeorological inputs (precipitation andtemperature) to generate hydrologic outputs (streamflow and evapotranspiration). These modelscontain empirical coefficients and parameters that require site-specific calibration and properestimation of model parameters for the hydrologic model to work successfully. Calibration andparameterization occur in the CS phase of NWSRFS. Extensive research related to hydrologicmodel calibration has been reported in the literature (see Duan et al., 2003). In the NWSRFS,simulated streamflow is calibrated statistically and visually against the observed streamflow todetermine which model parameters need adjustment to improve alignment. After the models have
Scientific and Technical Aspects of AHPS 39been calibrated for a specific basin, the optimal set of parameters can be combined with real-timehydrometeorological data in the OFS to predict streamflow (Koren et al., 2003; Smith et al., 2003).Limitations of and Research Needs for the Calibration System The primary limitation of the NWSRFS CS is a gap between the state-of-the-art calibrationcapabilities and what is used in operations in the NWS RFCs. The NWS is aware of the need forclosing this gap and recently has made needed improvements along these lines. One suchimprovement is the interactive Calibration Assistance Program, which incorporates GIS andinteractive user interfaces into the NWSRFS (Smith et al., 2003). Another improvement has beenthe development of a regional parameter estimation scheme that relates soil information to theparameters of the SAC-SMA (Koren et al., 2003). These efforts are commended. The benefits ofthese advancements to AHPS will be realized as they are translated into NWSRFS operations. RFCand WFO staff training on the purpose, protocol, and function of calibration and parameterizationimprovements will help ensure appropriate and consistent use of new techniques. The NWSshould continue efforts to improve and expand AHPS calibration capabilities, accelerate therate of transfer of the latest calibration techniques into its operational AHPS-NWSRFSversion, and conduct adequate training of modeling personnel to ensure appropriate andconsistent use of the new techniques. Like calibration advancements, model parameterization improvements in coupled climate/hydrologic models need to be transferred into AHPS operation. With the increasing demand forlonger-term hydrologic predictions, and the AHPS goal to provide longer range forecasts of two weeks orfurther into the future, it is necessary to improve the interface between climate/land-surface models andhydrologic rainfall runoff models. The HL has recently spearheaded the international ModelParameter Estimation Experiment (MOPEX) to develop enhanced a priori estimates of hydrologicmodel parameters for both gaged and ungaged basins (Duan et al., 2006). MOPEX would providevaluable support to the incorporation of more physically based modeling capabilities into AHPS.MOPEX, and efforts like it, are expected to have a strong connection to the DMIP effort, as modelcalibration was identified (see Box 3-1) as the possible and pivotal component of modelperformance in DMIP Phase I. The development of MOPEX is commended, and the goals ofDMIP and MOPEX should be compatible with each other and with AHPS.Functions of the NWSRSF Operational Forecast System and the Interactive Forecast Program The elements of OFS and IFP are similar, but they perform slightly different functions. TheOFS is larger than IFP and includes pre-processing data (computing areal and temporal averages),model setup (storing parameters in the data base), and model computations. OFS reads raw stationdata in near real-time, estimates missing data as required, and then it uses these data to calculatemean areal time series of precipitation, temperature and potential evapotranspiration. Calibratedmodels in the OFS are forced with these processed time series to generate river forecasts with leadtimes that typically range from one day to two weeks. The IFP is the graphical interface to the forecast component of OFS. Through the IFPinterface, forecasters manually adjust the model simulations to match the current observations asclosely as possible. The forecasters can adjust the model inputs, model states, model parameters (ina few cases) and model outputs. The forecasting component of the OFS maintains an account ofthe current model states that describe the hydrologic condition of the basin, including snow cover,
40 Toward a New Advanced Hydrologic Prediction Service (AHPS)soil moisture and channel storage, by storing these values in the operational database. The samemodels used in the forecast component of the OFS are used in the IFP. The model states stored bythe OFS reflect the modifications made by the forecasters in IFP. The updated model states areneeded as starting points for the subsequent forecasts made with the ESP system.Limitations of and Research Needs for OFS and IFP From the information available about the OFS/IFP component of the AHPS NWSRFS, afew observations and recommendations are noted. First, site visit interviews and other interviewswith NWS forecasters indicate that the current design of the OFS/IFP is difficult to use. Thisdifficulty is attributed to missing or hard-to-use- graphical interfaces. The NWS should review thecurrent suite of operational software and develop a comprehensive plan for refreshing thatsoftware. Two other, related concerns with the current configuration of OFS/IFP include the lack of a"model only" forecast run and automatic data assimilation. The OFS/IFP model state updatingprocess is done manually, and at least two problems are associated with the manual approach. First,a strong possibility exists for an individual forecaster to introduce error or bias when manuallyadjusting models. Manual adjustments are based primarily upon forecaster expertise, which will varyamong individuals. The resulting ad-hoc, inconsistent methods do not constitute a robust scientificapproach to assimilating observations into a model simulation in real-time. Second, forecasters'manual control of model output may obscure or thwart scientific advances that improve forecastskill and certainty. An NWS staff member stated that, "you might improve the [hydrologic] models100 fold, and never see any improvement [in forecasts] because the forecasters are always stickingtheir fingers into the mix." To avoid these problems, AHPS developers could adopt currentpractices from the meteorological forecast side of the NWS. NWS meteorologists run their models,"hands off" or "model only," and then transfer the forecast outputs "hands off" to the forecastoffices. NWS meteorologists use post-processing techniques to make forecast adjustments prior topublic issuance and they document these adjustments for future verification purposes. Like theirmeteorological counterparts in the NWS, hydrologic forecasters should run hydrologicmodels primarily in a "model only" mode, make forecast adjustments with post-processingtechniques, and document these adjustments for future verification purposes. Elements of current real-time hydrologic data assimilation are recognized as problematic.There are many sources of error with routing, snow, runoff, and precipitation, and current dataassimilation uses only a single data point (typically river stage) for updating. The current lack of afully automated, robust data assimilation component precludes "model only" forecast runs. Withouta "model only" forecast system, it is not possible to assess the impact of a new calibration or otherscientific advancements. These limitations, in addition to the problems associated with manuallyproduced forecasts, suggest that the NWS should automate real-time hydrologic data assimilation.AHPS developers should consider automating the OFS/IFP component of the AHPS-NWSRFS and develop a systematic mechanism to include new research results and erroranalysis techniques into the operational OFS/IFP component. A switch from a manual to "model only" and an automated data assimilation process willimpact current forecasters' responsibilities. Although the responsibilities of the individualforecasters would change with an automated process, the role fulfilled by forecasters in the forecastprocess is essential and will continue to be important with automated data assimilation. In no wayshould forecasters be removed from the forecast process, and the NWS is urged to redefinethe role of the hydrologic forecaster in a fully automated data assimilation process.
Scientific and Technical Aspects of AHPS 41Function of the NWSRFS Ensemble Streamflow Prediction System The final component of the NWSRFS is the ESP system. The current version of the ESPcomponent of NWSRFS has a modular design (Figure 3-4) and allows future streamflow traces to beanalyzed for peak flows, minimum flows, and flow volumes. ESP assumes that historicalmeteorological data are representative of possible future conditions and uses past traces for thesame-season and location as input data to produce probabilistic hydrologic forecasts. Knowledge ofthe current climatology is often used to weight the years of simulated streamflow based on thesimilarity between the climatological conditions of each historical year and the current year. Morespecifically, the ESP component blends together historical temperature and precipitation datasequences and deterministic meteorological forecasts to form ensemble inputs to hydrologic modelsthat produce forecasts out to several months (Werner et al., 2005). A well designed andimplemented ESP would progress AHPS towards fulfilling its goal of providing products with forecasthorizons of two weeks or further into the future.Limitations of and Research Needs for ESP In the past decade, there have been significant advances in the development of ensembletools in the fields of atmospheric and hydrologic sciences. Based on a recent spate of publishedworks by NWS personnel or about ESP improvements for the NWSRFS, AHPS researchers appearto be developing and presenting these advancements for incorporation into AHPS products andforecasts. Published or written documentation exists for some, but not all, ESP sub-systems in theNWSRFS. ESP sub-systems with published documentation include: the ESP pre-processor(Schaake et al., 2005); ESP verification (Bradley et al., 2004); medium-range forecasts (Werner et al.,2005); climate index weighting (Werner et al., 2004); and the ESP post-processor (Seo et al., 2006).These recent publications reflect strong advancements of ESP tools and their potentialincorporations into the NWSRFS and by extension to AHPS products and forecasts. Through these and other research efforts, AHPS aims to improve the ESP system andproduce seamless and consistent probabilistic forecasts. Probabilistic forecasts explicitly quantifylevels of uncertainty associated with each ensemble forecast. Quantified uncertainty for hydrologicforecasts offers several advantages, including the ability to archive forecasts and assess the overallskill of hydrologic forecasts over time based on comparisons against observed conditions. NWSmeteorological forecasts consistently have associated, quantified uncertainties, but hydrologicforecasts historically have not. Fortunately, recent developments in hydrologic modeling and toolssteadily increase the number and percentage of probabilistic hydrologic forecasts. The NWS andAHPS researchers are commended for advancing the ESP tools for possible incorporation into theNWSRFS, and for publishing and documenting their results. Still, the NWS needs to more stronglyconnect these advancements to the overall AHPS Development and Implementation plan andspecify how they fit into the envisioned sequence of implementation of the ESP (Figure 3-5). Figure 3-5 presents the proposed sequence of the enhancements for short- to long-termforecasting services with the approximate delivery to RFCs. From this documentation (NWS, 2004),
42 Toward a New Advanced Hydrologic Prediction Service (AHPS) Architecture Management Meteorological Forecasts Ensemble Pre- Processor Hydrologic Ensemble Processed Reservoir, River Ensemble Forcing Input Ensemble Streamflow Post- Streamflow Regulation, & Product Ensembles Processor Ensembles Processor Ensembles Hydraulic Models Dissemination Probabilistic Forecasts Probabilistic Verification Verification InformationFIGURE 3-4 System components for the ESP.SOURCE: NWS, 2004. Pre-processor 1 Pre-processor 1 Short-term Smoothing Pre-processor 2 Short-term Smoothing Verification 1 (Short- and Med-term) Pre-processor Verification 1 (Med-term) Verification 2 Verification 3 CY 2004 CY 2005 CY 2006 CY 2007 Initial Condition Post-Processor 1 Hydraulic Model Model Parameters (Ensemble) Model Structure Post-processor 3 Post-processor 1 River Regulation Initial Conditions River Regulation Reservoirs Post-Processor 2 Reservoirs Evaluation On-going Activities: Implementation - Enhancements for Fielded SystemsFIGURE 3-5 Envisioned sequence of implementation of the ensemble system.SOURCE: NWS, 2004.it is unclear what experimental design and methods are being used to develop the ESP sub-systemsand whether a prototype of this framework is being tested before implementation in this succession(Figure 3-5). Furthermore, this schedule seems incomplete because it omits important elementssuch as research, analysis, and operational development, and the supporting text (NWS, 2004) doesnot fully describe all development activities associated with implementing this sequence. Therefore,AHPS should document its overall strategy about ESP, including priorities for the ESPsystem and sub-system development, testing, and implementation. The AHPS approach to
Scientific and Technical Aspects of AHPS 43quantifying uncertainties in operational forecasts must be articulated. In addition, AHPSshould clarify connections between current and future research activities and the AHPSoverall development and implementation and ESP sequencing plans.Function of NWSRFS Verification Verification includes documenting the uncertainty expected for each forecast andmonitoring over time the accuracy of the forecasts against observed conditions. AHPS developersare commended for including verification in the NWSRFS, which could provide long-term statisticson the skill of AHPS and all NWSRFS forecasts. Quantification of uncertainty in forecasts shouldinclude measures of bias or accuracy and measures of variability of the ensemble forecasts. Biasmeasures can derive from comparisons of forecasts and observed field conditions; variancemeasures can be calculated from the statistical variability of the forecast. Inclusion of verificationsub-systems in the ESP system design (Bradley et al., 2004), as well as in the OFS, is needed andlong overdue. Unlike meteorological forecasts, little is known about hydrologic forecasts and actual riverforecast skill. The assumption that forecasts have been improving over time may not be truebecause it is not documented whether the forecasts have skill over simple persistence forecasts. Theimportance of verification was highlighted in a recent Ph.D. dissertation (Welles, 2005) when 10years of NWS river stage forecasts for 5 locations and 20 years of NWS forecasts for 11 locationswere evaluated using standard verification metrics. The improvement in the forecast skill was not asgreat as had been anticipated (Welles, 2005), although these results are not definitive due to thelimited sample size of the study. This research underscores the need to implement a long-termverification strategy and maintenance of a forecast archive for future forecast verification andNWSRFS evaluation.Limitations of and Research Needs for Verification in the NWSRFS Hydrometeorologists need to understand the skill characteristics of their forecasts, and thiscan only be accomplished through rigorous verification of the forecasts, including quantification ofuncertainty of the forecasts, and quantifying the accuracy and variability of ensemble forecastscompared to observed field conditions. It is possible to make available verification information,such as river forecast skill, as an AHPS product for each forecast point. While the inclusion of averification sub-component in AHPS NWSRFS is commended, there is a pressing need fora long-term strategy and maintenance of a forecast archive for future verification andNWSRFS evaluation. Overarching Limitations of NWSRFS The software architecture of NWSRFS is a major limiting factor in the development andimplementation of AHPS. The NWSRFS software is based on antiquated FORTRAN-basedalgorithms of the 1970s and 1980s, and current distributed, statistical, and probabilistic models thatare in various stages of research for use in AHPS products are not aligned with it.
44 Toward a New Advanced Hydrologic Prediction Service (AHPS) The NWS recently started to address the NWSRFS system software problems through theCommunity Hydrologic Prediction System (CHPS)10. CHPS is an effort to redesign forecast systemarchitecture based on a web-service architecture. CHPS would enable sharing of modelingapplications across the hydrologic "community," which is composed of people from research,government, and academic organizations. As described, CHPS would provide a modular modelingframework for the development, enhancement, integration, and application of a wide variety ofmodels and associated analysis and forecasting tools. CHPS is made up primarily of NWSemployees and it has a strong NWS focus. Similar frameworks to CHPS are in development atother federal agencies and research centers. The U.S. Geological Survey's Modular ModelingSystem11, the U.S. Department of Agriculture's Object Modeling System12, the U.S. Department ofEnergy's (DOE) Framework for Risk Analysis in Multimedia Environmental Systems (Whelan, etal., 1997), and the DOE's Dynamic Information Architecture System13 are examples of majorsoftware development efforts. The experiences gained from other federal modelingcollaborations should be considered in the development of CHPS. Even with CHPS and other NWS efforts to address the NWSRFS software problems, thecurrent NWSRFS system must operate until new AHPS functionality is developed and implemented.The NWS will either fit new AHPS capabilities into the existing framework or abandon NWSRFSfor a new, redesigned approach. The addition of new hydrologic methods into NWSRFS in somecases, such as in distributed modeling, may be impossible given the current structure. Furthermore,site specific hydrologic conditions may require alternative or even multiple models and techniques tobe applied at a particular location in order to optimize forecast skill. The NWSRFS is a barrier to the AHPS goal of producing more accurate products by incorporatingadvanced hydrologic science into the NWS model. The existing forecast system severely limits the ability to:(1) test research advances within the NWSRFS framework; (2) add new and diverse hydrologicfeatures to the system; and (3) accelerate the transfer of new technology to operations. Toincorporate the state-of-the-art hydrologic modeling capabilities, the NWS should invest inthe next generation of NWSRFS that includes a flexible framework that allows alternativemodels, methods, or features that can be tested, verified and implemented expediently. Atotal redesign of the NWSRFS is needed for AHPS to fulfill its scientific and technical goals.A redesign would involve updating NWSRFS to current state-of-the-art software and hardwarestandards and using software that is more modular in design to support future modifications andenhancements of AHPS. FLASH-FLOOD GUIDANCE Flash-floods occur within a few short hours from the onset of heavy precipitation, and rankamong the top natural hazards in the U.S because they cause major losses of life and property. Likethe NWSRFS, the scientific foundations of current NWS flash-flood guidance and flash-floodwarnings generation are derived from 1970s and 1980s techniques. Primarily, flash-flood forecastinghas remained in the meteorological domain, and few, if any, hydrologic tools and models have beendeveloped to forecast flash-floods (Droegemeier et al., 2000). AHPS has a goal to provide more specificand timely information on fast-rising floods with increased lead time. In order to achieve this goal, AHPS will10http://www.nws.noaa.gov/om/water/ahps/BAMS_Article.pdf.11http://wwwbrr.cr.usgs.gov/projects/SW_precip_runoff/mms/.12http://oms.ars.usda.gov/.13http://www.dis.anl.gov/DIAS/.
Scientific and Technical Aspects of AHPS 45need to update the hydrologic scientific basis for flash-flood guidance. The needs of updating thescientific basis for flash-flood guidance have been identified before (NRC, 1996) and the followingrecommendation is repeated here to help guide the NWS to fulfill the AHPS flash-flood goal: The NWS should improve the scientific basis that underpins the forecasting of floods that occur in the zero to six-hour time frame. WFO and RFC staff should be enabled to contribute to this effort by facilitating their access to adequate training, continuing education, and university cooperative programs. Furthermore, they should be able to access state-of-the-art geographic information systems, digital elevation models, and drainage and land-use data.There is no evidence of any significant progress in development of hydrologically-based flash-floodmodeling systems; hence, the issues identified in 1996 (NRC, 1996) still apply today. According to NWS presentations, there are plans to develop site-specific SAC-SMA todetermine local hydrologic preconditions for flash-flooding for AHPS. The NWS is consideringusing this "semi-distributed" model at RFCs (see earlier section, Hydrologic Models of NWSRFS),and perhaps eventually employing a statistical distributed model to replace the current flash-floodguidance. The statistical model would be based on developing the frequency distribution offlooding at ungaged locations based on retrospective QPE data. The technical basis for thesevarious approaches to flash-flood guidance and the flash-flood problem are not well documented.The utility and importance of the choice of hydrologic model for AHPS products is noted (seeearlier section, Hydrologic Models of NWSRFS), and the same issues discussed with respect toNWSRFS models apply to flash-flood forecasting because model selection will be central to flash-flood forecasting envisioned for AHPS. Like with hydrologic models in NWSRFS, the step-by-steptesting and evaluation plans are not defined for transitioning to distributed modeling for flash-floodforecasting. The NWS plans to implement a national verification program for its flash-flood monitoringand prediction (FFMP) effort (NWS, 2004). Verification is welcome, but more detail aboutdeliverables and milestones needs to accompany it. The plans for improvements in the productionand use of QPE and QPF in FFMP are mentioned as well, and again, a schedule for deliverables andmilestones along with an evaluation plan is required. Therefore, the NWS should provideadequate documentation within AHPS of the scientific details and the implementationstrategy for its end-to-end flash-flood hazards forecast generation and dissemination. Coordination between RFCs and WFOs in hydrological and hydrometeorological analysesand modeling will be required for producing and delivering forecasts, warnings and watches at localscales where the information is useful and actionable for the public. AHPS should include theforecast of flash-flood hazards and generation of warnings (dissemination) at the local WFO-level inits suite of activities. As a core capability, AHPS should include support for the forecast offlash-flood hazards and generation of warnings at the local WFO-level. CHAPTER SUMMARY This chapter describes and evaluates the scientific, technical, and modeling aspects of AHPS.The NWSRFS is the primary modeling engine for hydrologic forecasts, and each of the modularelements of NWSRFS is discussed, and recommendations are made to improve NWSRFS and helpthe NWS to fulfill the scientific and technical goals of AHPS. The evaluation of these scientific and
46 Toward a New Advanced Hydrologic Prediction Service (AHPS)technical aspects of AHPS resulted in numerous, specific recommendations throughout the chapter,and the major recommendations are noted here as three broad observations. First, improvements are needed to the precipitation inputs to the hydrologic models that areused to generate AHPS hydrologic forecasts. The quality of the hydrologic forecast depends on thequality of its precipitation inputs, and improvements to precipitation inputs will work towardsfulfilling the AHPS goal of creating information that is useful to assess risk to flooding through better, moreaccurate forecasts. Therefore, AHPS should strengthen the QPE and QPF through an end-to-endevaluation that assesses QPE/QPF quality and impacts on flood and streamflow products for basinsof diverse size and topography. In addition to improving QPE and QPF, AHPS developers areencouraged to work with satellite precipitation groups to ensure the AHPS hydrologic requirementsfor precipitation are considered in other federal activities, such as NASA's Global PrecipitationMeasurement mission. Second, the modeling capability needs improvements for AHPS to produce more accurateproducts and incorporate advanced hydrologic science in the NWS hydrologic models. Also noted in the modelingevaluation was a gap between the state-of-the-art hydrologic modeling capabilities and those used inAHPS product development. The current AHPS model, the NWSRFS, is described as needingupdates, better verification, and better alignment with models that have finer spatial and temporalresolution. Therefore, AHPS should invest in the next generation of NWSRFS that includes aflexible framework that allows alternative models, methods, or features that can be tested, verified,and implemented expediently. CHPS, DMIP, and other collaborative activities to address themodeling capability of AHPS are commended, and the committee recommends that the NWSstrengthen connections between DMIP Phase I/DMIP Phase II and AHPS goals. The committeealso recommends that the NWS clarify the criteria and decisionmaking process for selecting thenext generation hydrologic model(s) for AHPS, using an advisory board that involves modelingexperts from inside and outside of the NWS to ensure that the state-of-the-art modeling advancesare incorporated objectively into NWSRFS. Finally, a recurrent finding in this evaluation was that very few scientific and technicalaspects of AHPS are documented. The program will benefit from greater publication, peer review,and dissemination of its current and recent activities to improve the hydrologic science andtechnology used in AHPS product development and operation. The full list of this chapter'srecommendations is presented in Box 3-2. BOX 3-2 Recommendations · AHPS researchers should periodically publish progress of the development of and improvements to precipitation products. · AHPS developers are strongly encouraged to work closely with satellite precipitation groups (NASA, NOAA, and those in the academic community) to ensure that AHPS hydrologic requirements for precipitation are included in the Global Precipitation Measurement mission. · The NWS should strengthen QPE and QPF for hydrologic prediction through an end- to-end evaluation that assesses QPE/QPF quality and impacts on flood and streamflow products for basins of diverse size and topography. continues
Scientific and Technical Aspects of AHPS 47 BOX 3-2 Continued · The NWS should strengthen connections between DMIP Phase I/DMIP Phase II and AHPS goals and clarify the criteria and a decision-making process for selecting the modeling engine for AHPS. To do so, the NWS should form an advisory structure that involves modeling experts from inside and outside of the NWS to ensure that the state-of-the-art modeling advances are incorporated into NWSRFS operations. · The NWS needs to provide stronger documentation to allow the research community to learn about and contribute to AHPS research and development. · The NWS should continue efforts to improve and expand AHPS calibration capabilities, accelerate the rate of transfer of the latest calibration techniques into its operational AHPS- NWSRFS version, and conduct adequate training of modeling personnel to ensure appropriate and consistent use of the new techniques. · The goals of DMIP and MOPEX should be compatible with each other and with AHPS. · The NWS should review the current suite of operational software and develop a comprehensive plan for refreshing that software. · Like their meteorological counterparts in the NWS, hydrologic forecasters should run hydrologic models primarily in a "model only" mode, make forecast adjustments with post-processing techniques, and document these adjustments for future verification purposes. · AHPS developers should consider automating the OFS/IFP component of the AHPS-NWSRFS and develop a systematic mechanism to include new research results and error analysis techniques into the operational OFS/IFP component. · In no way should forecasters be removed from the forecast process, and the NWS is urged to redefine the role of the hydrologic forecaster in a fully automated data assimilation process. · AHPS should document its overall strategy about ESP, including priorities for the ESP system and sub-system development, testing, and implementation. The AHPS approach to quantifying uncertainties in operational forecasts must be articulated. In addition, AHPS should better specify connections between current and future research activities and the AHPS overall development and implementation and ESP sequencing plans. · While the inclusion of a verification sub-component in AHPS NWSRFS is commended, there is a pressing need for a long-term strategy and maintenance of a forecast archive for future verification and NWSRFS evaluation. · The experiences gained from other federal modeling collaborations should be considered in the development of CHPS. · To incorporate the state-of-the-art hydrologic modeling capabilities, the NWS should invest in the next generation of NWSRFS that includes a flexible framework that allows alternative models, methods, or features that can be tested, verified and implemented expediently. A total redesign of the NWSRFS is needed for AHPS to fulfill its scientific and technical goals. continues
48 Toward a New Advanced Hydrologic Prediction Service (AHPS) BOX 3-2 Continued · The NWS should improve the scientific basis that underpins the forecasting of floods that occur in the zero to six-hour time frame. WFO and RFC staff should be enabled to contribute to this effort by facilitating their access to adequate training, continuing education, and university cooperative programs. Furthermore, they should be able to access state-of-the-art geographic information systems, digital elevation models, and drainage and land-use data. · The NWS should provide adequate documentation within AHPS of the scientific details and the implementation strategy for its end-to-end flash-flood hazards forecast generation and dissemination. · As a core capability, AHPS should include support for the forecast of flash-flood hazards and generation of warnings at the local WFO-level. REFERENCESAnderson, E. 1968. Development and testing of snow pack energy balance equations. Water Resources Research 4(1):19-37.Anderson, E. 1973. National Weather Service River Forecast System: Snow Accumulation and Ablation Model. NOAA Technical Memorandum NWS Hydro-17. Silver Spring, MD: NWS.Anderson, E. 1976. A Point Energy and Mass Balance Model of a Snow Cover. NOAA Technical Report: NWS 19. Silver Spring, MD: NWS.Bradley, A., S. Schwartz, and T. Hashino. 2004. Distributions-oriented verifications of ensemble streamflow predictions. Journal of Hydrometeorology 5: 532-45.Burnash, R., R. Ferral, and R. McGuire. 1973. A generalized streamflow simulation system: Conceptual modeling for digital computers. Technical Report, Joint Federal and State River Forecast Center. Sacramento, CA.: NWS and California Department of Water Resources.Droegemeier, K., J. Smith, S, Businger, C. Doswell, J. Doyle, C. Duffy, E. Foufoula-Georgiou, T. Graziano, L. James, V. Krajewski, M. LeMone, D. Lettenmaier, C. Mass, R. Pielke, P. Ray, S. Rutlegde, J. Schaake, and E. Zipser. 2000. Hydrological aspects of weather prediction and flood warnings: Report of the ninth prospectus development team of the U.S. Weather Research Program. Bulletin of the American Meteorological Society 81(11): 2665-80.Duan, Q., H. Gupta, S. Sorooshian, A. Rousseau, and R. Turcotte, eds. 2003. Calibration of Watershed Models: Water Science and Application Series Volume 6. Washington, DC: American Geophysical UnionDuan, Q., J. Schaake, V. Andréassian, S. Franks, G. Goteti, H. Gupta, Y. Gusev, F. Habets, A. Hall, L. Hay, T. Hogue, M. Huang, G. Leavesley, X. Liang, O. Nasonova, J. Noilhan, L. Oudin, S. Sorooshian, T. Wagener, E. Wood. 2006. Model Parameter Estimation Experiment (MOPEX): An overview of science strategy and major results from the second and third workshops. Journal of Hydrology 320(1-2): 3-17.Hong, Y., K. Hsu, X. Gao, and S. Sorooshian. 2004. Precipitation estimation from remotely sensed information using an artificial neural network--cloud classification system. Journal of Applied Meteorology 43:1834-52.
Scientific and Technical Aspects of AHPS 49Joyce, R., J. Janowiak, P. Arkin, and P. Xie. 2004. CMORPH: A method that produces global precipitation estimates from passive microwave and infrared data at high spatial and temporal resolution. Journal of Hydrometeorology 5: 487-503.Kondragunta, C. 2002. Abstract H21B-08: An experimental multi-sensor rainfall estimation technique. EOS Transactions of the American Geophysical Union 83 (Spring Meeting Supplement).Koren, V., M. Smith, and Q. Duan. 2003. Use of a priori parameter estimates in the derivation of spatially consistent parameter sets of rainfall-runoff models. In Calibration of Watershed Models: Water Science and Application Series Volume 6, Q. Duan, H. Gupta, S. Sorooshian, A. Rousseau, and R. Turcotte, eds. Washington, DC: American Geophysical Union.Koren, V., E. Anderson, and M. Smith. 2004. NWS-HL Cold Season Processes Research and Development (Hydrology Lab Internal Publication). Silver Spring, MD: NWS.McEnery, J., J. Ingram, Q. Duan, T. Adams, and L. Anderson. 2005. NOAA's Advanced Hydrologic Prediction Service: Building Pathways for Better Science in Water Forecasting. Available on-line at http://www.nws.noaa.gov/om/water/ahps/BAMS_Article.pdf. Accessed December 20, 2005.NRC (National Research Council). 1996. Toward a New National Weather Service: Assessment of Hydrologic and Hydrometeorological Operations and Services. Washington, DC: National Academy Press.NWS (National Weather Service). 1972. National Weather Service River Forecast System River Forecast Procedures. NOAA Technical Memo. NWS HYDRO-14. Silver Spring, MD: NWS.NWS. 2004. Draft: Advanced Hydrologic Prediction Service (AHPS) Development and Implementation Plan. Available on-line at http://www.nws.noaa.gov/oh/rfcdev/docs/AHPS% 20%20Plan%208_2_04-1.pdf. Accessed May 26, 2005.Reed, S., V. Koren, M. Smith, Z. Zhang, F. Moreda, D. Seo, and DMIP Participants. 2004. Overall Distributed Model Intercomparison Project results. Journal of Hydrology 298: 27-60.Schaake, J., J. Demargne, M. Mullusky, E. Welles, L. Wu, H. Herr, X. Fan, and D. Seo. 2005. Precipitation and temperature short-term ensemble forecasts from existing operational single-value forecasts. Silver Spring, MD: NWS.Seo, D., and J. Breidenbach. 2002. Real-time correction of spatially nonuniform bias in radar rainfall data using rain gauge measurements. Journal of Hydrometeorology 3: 93-111.Seo, D., H. Herr, and J. Schaake. 2006. A statistical post-processor for accounting of hydrologic uncertainty in short-range ensemble streamflow prediction. Silver Spring, MD: NWS.Smith, M., D. Laurine, V. Koren, S. Reed, and Z. Zhang. 2003. Hydrologic Model Calibration. In Calibration of Watershed Models: Water Science and Application Series Volume 6, Q. Duan, H. Gupta, S. Sorooshian, A. Rousseau, and R. Turcotte, eds. Washington, DC: American Geophysical Union.Smith, M., V. Koren, Z. Zhang, S. Reed, D. Seo, F. Moreda, V. Kuzmin, Z. Cui, and R. Anderson. 2004a. NOAA NWS Distributed Hydrologic Modeling Research and Development. NOAA Technical Report NWS 45. Available on-line at http://www.nws.noaa.gov/oh/hrl/distmodel/NOAA_TR45.pdf. Accessed December 19, 2005.Smith, M., V. Koren, S. Reed, Z. Zhang, F. Moreda, R. Anderson, V. Kuzmin, Z. Cui, and E. Anderson. 2004b. NWS Hydrologic Model Calibration Research and Development. Silver Spring, MD: NWS.Smith, M., D. Seo, V. Koren, S. Reed, Z. Zhang, Q. Duan, F. Moreda, and S. Cong. 2004c. The Distributed Model Intercomparison Project (DMIP): Motivation and experiment design. Journal of Hydrology 298(1-4): 4-26.
50 Toward a New Advanced Hydrologic Prediction Service (AHPS)Welles, E. 2005. Verification of River Stage Forecasts, Ph.D. Thesis. May, 2005. Tucson, AZ: University of Arizona.Werner, K., D. Brandon., M. Clark, and S. Gangopadhyay. 2004. Climate index weighting schemes for NWS ESP-based seasonal volume forecasts. Journal of Hydrometeorology 5: 1076-90.Werner, K., D. Brandon., M. Clark, and S. Gangopadhyay. 2005. Incorporating medium-range numerical weather model output into the Ensemble Streamflow Prediction system of the National Weather Service. Journal of Hydrometeorology 6: 101-14.Whelan G., K. Castleton, J. Buck, G. Gelston, B. Hoopes, M. Pelton, D. Strenge, and R. Kickert. 1997. Concepts of a Framework for Risk Analysis in Multimedia Environmental Systems (FRAMES). PNNL-11748. Richland, WA: Pacific Northwest National Laboratory.Zhang, Z., V. Koren, M. Smith, and S. Reed. 2001. Application of a distributed modeling system using gridded NEXRAD data. Fifth International Symposium on Hydrological Applications of Weather Radar, Heian-kaikan, Kyoto, Japan.Yilmaz, K., T. Hogue, K. Hsu, S. Sorooshian, H. Gupta, and T. Wagener. 2005. Intercomparison of rain gauge, radar and satellite-based precipitation estimates with emphasis on hydrologic forecasting. Journal of Hydrometeorology 6(4):497-517.
