Distributed Model Intercomparison Project Research Articles

Improving WetSpa model to predict streamflows for gaged and ungaged catchments

In the second phase of the Distributed Model Intercomparison Project (DMIP2), the WetSpa model is applied to simulate flows at basin and subbasin scales. Parent basins and their nested subbasins are modeled as gaged and ungaged basins, respectively. Available observations in the subbasins were only used to validate the model predictions. Gaged basins simulation results show that the predictions and observations are in good agreement. However, major peaks are underestimated, as is often the case in runoff modeling. Underestimation of high flows and in particular peak flows indicates that the prediction of actual runoff coefficients in the current WetSpa model needs to be improved. Also, sensitivity analysis of the model parameters reveals that the baseflow recession coefficient is the most sensitive parameter and care should be taken when modeling ungaged basins. Hence, by estimating this parameter for each subbasin separately, the model performance for the subbasins can be improved. To do this, a Boussinesq groundwater flow equation is used to improve the prediction of baseflow recession coefficients in the subbasins. Comparison between the original and the modified WetSpa models shows that the modified model yields relatively higher performances for the subbasins, creating a more accurate model for predicting ungaged subbasins.

The distributed model intercomparison project – Phase 2: Experiment design and summary results of the western basin experiments

Summary The Office of Hydrologic Development (OHD) of the U.S. National Oceanic and Atmospheric Administration’s (NOAA) National Weather Service (NWS) conducted the two phases of the Distributed Model Intercomparison Project (DMIP) as cost-effective studies to guide the transition to spatially distributed hydrologic modeling for operational forecasting at NWS River Forecast Centers (RFCs). Phase 2 of the Distributed Model Intercomparison Project (DMIP 2) was formulated primarily as a mechanism to help guide the U.S. NWS as it expands its use of spatially distributed watershed models for operational river, flash flood, and water resources forecasting. The overall purpose of DMIP 2 was to test many distributed models forced by high quality operational data with a view towards meeting NWS operational forecasting needs. At the same time, DMIP 2 was formulated as an experiment that could be leveraged by the broader scientific community as a platform for the testing, evaluation, and improvement of distributed models. DMIP 2 contained experiments in two regions: in the DMIP 1 Oklahoma basins, and second, in two basins in the Sierra Nevada Mountains in the western USA. This paper presents the overview and results of the DMIP 2 experiments conducted for the two Sierra Nevada basins. Simulations from five independent groups from France, Italy, Spain and the USA were analyzed. Experiments included comparison of lumped and distributed model streamflow simulations generated with uncalibrated and calibrated parameters, and simulations of snow water equivalent (SWE) at interior locations. As in other phases of DMIP, the participant simulations were evaluated against observed hourly streamflow and SWE data and compared with simulations provided by the NWS operational lumped model. A wide range of statistical measures are used to evaluate model performance on a run-period and event basis. Differences between uncalibrated and calibrated model simulations are assessed. Results indicate that in the two study basins, no single model performed best in all cases. In addition, no distributed model was able to consistently outperform the lumped model benchmark. However, one or more distributed models were able to outperform the lumped model benchmark in many of the analyses. Several calibrated distributed models achieved higher correlation and lower bias than the calibrated lumped benchmark in the calibration, validation, and combined periods. Evaluating a number of specific precipitation-runoff events, one calibrated distributed model was able to perform at a level equal to or better than the calibrated lumped model benchmark in terms of event-averaged peak and runoff volume error. However, three distributed models were able to provide improved peak timing compared to the lumped benchmark. Taken together, calibrated distributed models provided specific improvements over the lumped benchmark in 24% of the model-basin pairs for peak flow, 12% of the model-basin pairs for event runoff volume, and 41% of the model-basin pairs for peak timing. Model calibration improved the performance statistics of nearly all models (lumped and distributed). Analysis of several precipitation/runoff events indicates that distributed models may more accurately model the dynamics of the rain/snow line (and resulting hydrologic conditions) compared to the lumped benchmark model. Analysis of SWE simulations shows that better results were achieved at higher elevation observation sites. Although the performance of distributed models was mixed compared to the lumped benchmark, all calibrated models performed well compared to results in the DMIP 2 Oklahoma basins in terms of run period correlation and %Bias, and event-averaged peak and runoff error. This finding is noteworthy considering that these Sierra Nevada basins have complications such as orographically-enhanced precipitation, snow accumulation and melt, rain on snow events, and highly variable topography. Looking at these findings and those from the previous DMIP experiments, it is clear that at this point in their evolution, distributed models have the potential to provide valuable information on specific flood events that could complement lumped model simulations.

Distributed assimilation of satellite‐based snow extent for improving simulated streamflow in mountainous, dense forests: An example over the DMIP2 western basins

Snow cover area affects snowmelt, soil moisture, evapotranspiration, and ultimately streamflow. For the Distributed Model Intercomparison Project – Phase 2 Western basins, we assimilate satellite‐based fractional snow cover area (fSCA) from the Moderate Resolution Imaging Spectroradiometer, or MODIS, into the National Weather Service (NWS) SNOW‐17 model. This model is coupled with the NWS Sacramento Heat Transfer (SAC‐HT) model inside the National Aeronautics and Space Administration's (NASA) Land Information System. SNOW‐17 computes fSCA from snow water equivalent (SWE) values using an areal depletion curve. Using a direct insertion, we assimilate fSCAs in two fully distributed ways: (1) we update the curve by attempting SWE preservation, and (2) we reconstruct SWEs using the curve. The preceding are refinements of an existing simple, conceptually guided NWS algorithm. Satellite fSCA over dense forests inadequately accounts for below‐canopy snow, degrading simulated streamflow upon assimilation during snowmelt. Accordingly, we implement a below‐canopy allowance during assimilation. This simplistic allowance and direct insertion are found to be inadequate for improving calibrated results, still degrading them as mentioned above. However, for streamflow volume for the uncalibrated runs, we obtain: (1) substantial to major improvements (64–81%) as a percentage of the control run residuals (or distance from observations), and (2) minor improvements (16–22%) as a percentage of observed values. We highlight the need for detailed representations of canopy‐snow optical radiative transfer processes in mountainous, dense forest regions if assimilation‐based improvements are to be seen in calibrated runs over these areas.

Analysis of inconsistencies in multi-year gridded quantitative precipitation estimate over complex terrain and its impact on hydrologic modeling

Summary It is common that bias trends of long term precipitation data change over time due to various factors such as gauge relocation and changes in data processing methods. Temporal consistency of this error characteristic of precipitation data is as important as accuracy itself for reliable streamflow prediction with a hydrologic model. The main goal of this paper is to illustrate the detection and adverse effect of inconsistent precipitation data on distributed hydrologic model simulations over a mountainous basin in the Sierra Nevada Mountains of California. We used 1-h 4 km gridded precipitation time series intended for the second phase of the Distributed Model Intercomparison Project (DMIP-2), sponsored by the US National Weather Service (NWS). We present various analyses to investigate the consistency of an hourly gridded precipitation time series from October 1988 through September 2006. First, hourly gridded precipitation data were aggregated into monthly mean areal precipitation totals over the basin and compared with basin average totals derived from the Parameter-Elevation Regressions on Independent Slopes Model (PRISM) monthly values. Second, double mass analysis was preformed on several discrete locations within the basin using both the DMIP-2 gridded precipitation and PRISM data. In addition, we performed a statistical consistency test on the DMIP-2 time series. Both analyses lead to the conclusion that over the entire analysis time period a clear change in bias in the DMIP-2 data occurred in the beginning of 2003. Moreover, it was found that the PRISM data also exhibit some inconsistency. The inconsistency of two elevation zone mean area precipitation (MAP) time series computed from the DMIP-2 gridded precipitation fields was corrected by adjusting hourly values based on the result from double mass analysis. Model simulations using the adjusted MAP data are improved compared to simulations with the inconsistent MAP input data.

The distributed model intercomparison project – Phase 2: Motivation and design of the Oklahoma experiments

► We introduce the 2nd phase of the Distributed Model Intercomparison Project. ► Experiments include simulations of soil moisture and streamflow. ► Results for DMIP 2 Oklahoma experiments presented in companion paper. The Office of Hydrologic Development (OHD) of the National Oceanic and Atmospheric Administration’s (NOAA) National Weather Service (NWS) conducted the second phase of the Distributed Model Intercomparison Project (DMIP 2). After DMIP 1, the NWS recognized the need for additional science experiments to guide its research-to-operations path towards advanced hydrologic models for river and water resources forecasting. This was accentuated by the need to develop a broader spectrum of water resources forecasting products (such as soil moisture) in addition to the more traditional river, flash flood, and water supply forecasts. As it did for DMIP 1, the NWS sought the input and contributions from the hydrologic research community. DMIP 1 showed that using operational precipitation data, some distributed models could indeed perform as well as lumped models in several basins and better than lumped models for one basin. However, in general, the improvements were more limited than anticipated by the scientific community. Models combining so-called conceptual rainfall-runoff mechanisms with physically-based routing schemes achieved the best overall performance. Clear gains were achieved through calibration of model parameters, with the average performance of calibrated models being better than uncalibrated models. DMIP 1 experiments were hampered by temporally-inconsistent precipitation data and few runoff events in the verification period for some basins. Greater uncertainty in modeling small basins was noted, pointing to the need for additional tests of nested basins of various sizes. DMIP 2 experiments in the Oklahoma (OK) region were more comprehensive than in DMIP 1, and were designed to improve our understanding beyond what was learned in DMIP 1. Many more stream gauges were located, allowing for more rigorous testing of simulations at interior points. These included two new gauged interior basins that had drainage areas smaller than the smallest in DMIP 1. Soil moisture and routing experiments were added to further assess if distributed models could accurately model basin-interior processes. A longer period of higher quality precipitation data was available, and facilitated a test to note the impacts of data quality on model calibration. Moreover, the DMIP 2 calibration and verification periods contained more runoff events for analysis. Two lumped models were used to define a robust benchmark for evaluating the improvement of distributed models compared to lumped models. Fourteen groups participated in DMIP 2 using a total of sixteen models. Ten of these models were not in DMIP 1. This paper presents the motivation for DMIP 2 Oklahoma experiments, discusses the major project elements, and describes the data and models used. In addition, the paper introduces the findings, which are covered in a companion results paper ( Smith et al., this issue ). Lastly, the paper summarizes the DMIP 1 and 2 experiments with commentary from the NWS perspective. Future papers will cover the DMIP 2 experiments in the western USA mountainous basins.

Results of the DMIP 2 Oklahoma experiments

Phase 2 of the Distributed Model Intercomparison Project (DMIP 2) was formulated primarily as a mechanism to help guide the US National Weather Service (NWS) as it expands its use of spatially distributed watershed models for operational river, flash flood, and water resources forecasting. The overall purpose of DMIP 2 was to test many distributed models with operational quality data with a view towards meeting NWS operational forecasting needs. At the same time, DMIP 2 was formulated as an experiment that could be leveraged by the broader scientific community as a platform for testing, evaluating, and improving the science of spatially distributed models.This paper presents the key results of the DMIP 2 experiments conducted for the Oklahoma region, which included comparison of lumped and distributed model simulations generated with uncalibrated and calibrated parameters, water balance tests, routing and soil moisture tests, and simulations at interior locations. Simulations from 14 independent groups and 16 models are analyzed. As in DMIP 1, the participant simulations were evaluated against observed hourly streamflow data and compared with simulations generated by the NWS operational lumped model. A wide range of statistical measures are used to evaluate model performance on both run-period and event basis. A noteworthy improvement in DMIP 2 was the combined use of two lumped models to form the benchmark for event improvement statistics, where improvement was measured in terms of runoff volume, peak flow, and peak timing for between 20 and 40 events in each basin.Results indicate that in general, those spatially distributed models that are calibrated to perform well for basin outlet simulations also, in general, perform well at interior points whose drainage areas cover a wide range of scales. Two of the models were able to provide reasonable estimates of soil moisture versus depth over a wide geographic domain and through a period containing two severe droughts. In several parent and interior basins, a few uncalibrated spatially distributed models were able to achieve better goodness-of-fit statistics than other calibrated distributed models, highlighting the strength of those model structures combined with their a priori parameters. In general, calibration solely at basin outlets alone was not able to greatly improve relative model performance beyond that established by using uncalibrated a priori parameters. Further, results from the experiment for returning DMIP 1 participants reinforce the need for stationary data for model calibration: in some cases, the improvements gained by distributed models compared to lumped were not realized when the models were calibrated using inconsistent precipitation data from DMIP 1. Event-average improvement of distributed models over the combined lumped benchmark was measured in terms of runoff volume, peak flow, and peak timing for between 20 and 40 events. The percentage of model-basin pairs having positive distributed model improvement at basin outlets and interior points was 18%, 24%, and 28% respectively, for these quantities. These values correspond to 14%, 33%, and 22% respectively, in DMIP 1. While there may not seem to be much gain compared to DMIP 1 results, the DMIP 2 values were based on more precipitation–runoff events, more model-basin combinations (148 versus 51), more interior ungauged points (9 versus 3), and a benchmark comprised of two lumped model simulations.In addition, we propose a set of statistical measures that can be used to guide the calibration of distributed and lumped models for operational forecasting.

On the use of spatial regularization strategies to improve calibration of distributed watershed models

Hydrologic models require the specification of unknown model parameters via calibration to historical input‐output data. For spatially distributed models, the large number of unknowns makes the calibration problem poorly conditioned. Spatial regularization can help to stabilize the problem by facilitating inclusion of additional information. While a common regularization approach is to apply a scalar multiplier to the prior estimate of each parameter field, this can cause problems by simultaneously changing both the mean and the variance of the distribution. This paper explores a multiple‐criteria regularization approach that facilitates adjustment of the mean, variance, and shape of the parameter distribution, using prior information to constrain the problem while providing sufficient degrees of freedom to enable model performance improvements. We also test simple squashing functions to help in maintaining conceptually reasonable parameter values throughout the spatial domain. We apply the method to three basins in the context of the Distributed Model Intercomparison Project (DMIP2), obtaining considerable performance improvements at the basin outlet. However, the prior parameter estimates are found to give much better performance at the interior points (treated as ungauged), suggesting that the spatial information has not been properly exploited. The results also suggest that basin outlet hydrographs may not be particularly sensitive to spatial parameter variability and that an overall basin mean value may be sufficient for flow forecasting at the outlet, although not at the interior points. We discuss weaknesses in our study approach and suggest diagnostically more powerful strategies to be pursued.

Assessing the impacts of precipitation bias on distributed hydrologic model calibration and prediction accuracy

Physics-based distributed (PBD) hydrologic models predict runoff throughout a basin using the laws of conservation of mass and momentum, and benefit from more accurate and representative precipitation input. Vflo™ is a gridded distributed hydrologic model that predicts runoff and continuously updates soil moisture. As a participating model in the second Distributed Model Intercomparison Project (DMIP2), Vflo™ is applied to the Illinois and Blue River basins in Oklahoma. Model parameters are derived from geospatial data for initial setup, and then adjusted to reproduce the observed flow under continuous time-series simulations and on an event basis. Simulation results demonstrate that certain runoff events are governed by saturation excess processes, while in others, infiltration-rate excess processes dominate. Streamflow prediction accuracy is enhanced when multi-sensor precipitation estimates (MPE) are bias corrected through re-analysis of the MPE provided in the DMIP2 experiment, resulting in gauge-corrected precipitation estimates (GCPE). Model calibration identified a set of parameters that minimized objective functions for errors in runoff volume and instantaneous discharge. Simulated streamflow for the Blue and Illinois River basins, have Nash–Sutcliffe efficiency coefficients between 0.61 and 0.68, respectively, for the 1996–2002 period using GCPE. The streamflow prediction accuracy improves by 74% in terms of Nash–Sutcliffe efficiency when GCPE is used during the calibration period. Without model calibration, excellent agreement between hourly simulated and observed discharge is obtained for the Illinois, whereas in the Blue River, adjustment of parameters affecting both saturation and infiltration-rate excess processes were necessary. During the 1996–2002 period, GCPE input was more important than model calibration for the Blue River, while model calibration proved more important for the Illinois River. During the verification period (2002–2006), calibration was more important than GCPE input for the Illinois River, but both calibration and GCPE input improved prediction accuracy within the Blue River.

Towards robust methods to couple lumped rainfall–runoff models and hydraulic models: A sensitivity analysis on the Illinois River

Summary Accurate prediction of a flood inundation area constitutes an essential part of a flood forecasting system. When a river reach receives significant lateral inflows, flood inundation modelling requires the joint application of a hydrological model to calculate lateral inflows and a hydraulic model to calculate water levels along the river reach. In this study, we compared different strategies to couple the GR4J lumped rainfall–runoff model and the linearised diffusive wave propagation model. These strategies introduced variations in the nature of the connections between the two models using combinations of point and uniformly distributed lateral inflows. The coupled model was then applied to the Illinois River case study for which 10 years of hourly data were made available within the Distributed Model Intercomparison Project, now in phase 2. The simulations were assessed at the downstream end of the reach and at two interior points considered to be ungauged during the calibration process. The results first show that including uniformly distributed inflows made the coupled model more robust and stable compared to only using point flow input. A similar level of performance was reached with models using point inflows only, but at the cost of more uncertain parameters and less stable model performance when changing test periods. Second, identifying the optimal number of tributaries to be modelled individually by the hydrological model was easier when a combination of uniformly distributed and point inflows was used. In this case, model performance was less sensitive to the number of tributaries used and the inclusion of two or three tributaries appeared sufficient to obtain satisfactory performance for the simulations on the main channel. Last, the results on the main channel differed from those obtained on tributaries: overall performance was better on the main channel and required a lower degree of lateral inflow resolution, which suggests that upstream flow measurements provide valuable information for flow simulation on ungauged points.

WetSpa model application in the Distributed Model Intercomparison Project (DMIP2)

This paper describes the application of a spatially distributed hydrologic model (WetSpa) Water and Energy Transfer between Soil, Plants and Atmosphere, for the second phase of the Distributed Model Intercomparison Project (DMIP2) study. The model implementation is based on 30-m spatial resolution and 1 h time-step for all basins and interior watersheds involved in the DMIP study. Rainfall inputs are derived from Next Generation Radar (NEXRAD). The distributed model parameters are based on an extensive database of watershed characteristics available for the region, including digital maps of topography, soil type, and landuse. The model is calibrated and validated on part of the river flow records for each basin and applied to the smaller interior watersheds not used in calibration to assess the model performance in ungaged basins. The statistics improve significantly with calibration of the global model parameters but even for uncalibrated simulations, the WetSpa model reproduces flow rates of acceptable accuracy for most cases. To evaluate the model performance during calibration and validation periods, an Aggregated Measure ( AM ) is introduced that measures different aspects of the simulated hydrograph such as shape, size and volume. The statistics for the five calibration basins show that the model produces very good to excellent results for the calibration period. With the exception of Blue River basin, the overall model performance for the validation period remains good to very good , indicating that the model is able to simulate the relevant hydrologic processes in the basins accurately. The performance of the uncalibrated model for the subcatchments is more variable, but the hourly flow rates generally reproduced with reasonable accuracy indicating an encouraging performance of the model.

GOES Solar Radiation for Evapotranspiration Estimation and Streamflow Prediction

This study explores the value of incoming solar radiation estimates derived from geostationary operational environmental satellites (GOES) as an alternative solar radiation data source to historical solar radiation estimates for use in rainfall – runoff simulations. At 26 sites in the United States, the GOES-based solar radiation (\iR²=0.79) and ET\dp (R²=0.94) estimates agreed well with the ground-based estimates. There was a slight positive bias for the GOES-based solar radiation (5.4%) and potential evapotranspiration (8.7%) estimates compared to the ground-based estimates, with the strong and significant biases noted during the warm seasons. However, individual stations biases ranged from – 12 to 27%. Daily streamflow predictions made using National Weather Service River Forecast System’s Sacramento Soil Moisture Accounting rainfall.runoff model for four NOAA Distributed Model Intercomparison Project watersheds show that although notable peak storm flow differences occurred occasionally during extended drying periods, on average, the GOES-based potential evapotranspiration estimates slightly outperformed the simulations using ground-based data on an annual basis and were relatively insensitive to biases. However, on seasonal time scales, differences in soil water storage are evident and can result in numerous moderate differences in streamflow predictions.

Application of HSPF to the Distributed Model Intercomparison Project: Case Study

The Hydrologic Simulation Program–Fortran (HSPF) model was used to simulate streamflow for the Distributed Model Intercomparison Project (DMIP). This research describes how to integrate a next generation radar, NEXRAD data set into HSPF model and how to generate hydrologic runoff associated with the inputs of the spatial rainfalls, and discusses the challenges of HSPF during automatic calibration processes. The model performance of simulated streamflows with calibration and without calibration was also evaluated to assess the sensitivity of final estimated parameters through calibration procedures to initial parameters derived from physical watershed configuration. An automatic calibration scheme was adopted to optimize objectives that the writer specified in terms of statistical measures. Parameter estimation, a model-independent parameter estimator, was used as an automatic calibration tool in the hydrologic calibration of HSPF. Overall, the calibrated simulations outperformed the uncalibrated simulatio...

Multimodel Combination Techniques for Analysis of Hydrological Simulations: Application to Distributed Model Intercomparison Project Results

Abstract This paper examines several multimodel combination techniques that are used for streamflow forecasting: the simple model average (SMA), the multimodel superensemble (MMSE), modified multimodel superensemble (M3SE), and the weighted average method (WAM). These model combination techniques were evaluated using the results from the Distributed Model Intercomparison Project (DMIP), an international project sponsored by the National Weather Service (NWS) Office of Hydrologic Development (OHD). All of the multimodel combination results were obtained using uncalibrated DMIP model simulations and were compared against the best-uncalibrated as well as the best-calibrated individual model results. The purpose of this study is to understand how different combination techniques affect the accuracy levels of the multimodel simulations. This study revealed that the multimodel simulations obtained from uncalibrated single-model simulations are generally better than any single-member model simulations, even the best-calibrated single-model simulations. Furthermore, more sophisticated multimodel combination techniques that incorporated bias correction step work better than simple multimodel average simulations or multimodel simulations without bias correction.

Discretization scale dependencies of the ensemble flow range versus catchment area relationship in distributed hydrologic modeling

Summary Earlier research results have shown that the range of the ensemble flow simulations from a well-validated operational hydrologic model possesses a well-defined log-linear relationship to catchment area for which the ensembles are generated. The ensemble range is a function of both parametric and rainfall input uncertainty. In this paper the sensitivity of this scaling relationship for simulation uncertainty to the numerical discretization scale of the distributed model is quantified when the model is driven by WSR-88D (NEXRAD) radar data and for two catchments in the south-central United States. The distributed model used is subcatchment (rather than pixel) based, has components that are based on operational models employed in the US National Weather Service, and has been validated for the two application catchments of this work as part of the Distributed Model Intercomparison Project (DMIP). The approach taken develops a new parsimonious model for spatially correlated radar-rainfall errors with radar-pixel error variance that depends on the magnitude of the observed rainfall. Monte Carlo methods are used to translate parametric and radar-rainfall input uncertainty to ensemble flow simulations at a number of subcatchments of varying size. Selected events in the period from May 1993 to July 1999 were analyzed. The results confirm that the ensemble flow range is dependent on subcatchment scale with a well-defined log-linear relationship for both application catchments. This finding holds for all uncertainty scenarios examined and across all the scales of distributed model discretization. It is further found that a higher model discretization leads to a shorter range of simulation uncertainty for subcatchment spatial scales that are substantially larger than the average subcatchment size. Flow simulation uncertainty that is due to either parametric or input uncertainty alone scale in a way that is similar to the combined uncertainty scaling.

Towards the characterization of streamflow simulation uncertainty through multimodel ensembles

Distributed hydrologic modeling holds significant promise for improved estimates of streamflow with high spatial resolution. However, uncertainty in model structure and parameters, which are distributed in space, and in operational weather radar rainfall estimates, which comprise the main input to the models, contributes to significant uncertainty in distributed model streamflow simulations over a wide range of space and time scales. Using the simulations produced for the Distributed Model Intercomparison Project (DMIP), this paper develops and applies sample-path methods to characterize streamflow simulation uncertainty by diverse distributed hydrologic models. The emphasis in this paper is on the model parameter and structure uncertainty given radar rainfall forcing. Multimodel ensembles are analyzed for six application catchments in the Central US to characterize model structure uncertainty within the sample of models (both calibrated and uncalibrated) participating in DMIP. Ensembles from single distributed and lumped models are also used for one of the catchments to provide a basis to characterize the impact of parametric uncertainty versus model structure uncertainty in flow simulation statistics. Two main science questions are addressed: (a) what is the value of multimodel streamflow ensembles in terms of the probabilistic characterization of simulation uncertainty? And (b) how do probabilistic skill measures of multimodel versus single-model ensembles compare? Discussed also are implications for the operational use of streamflow ensembles generated by distributed hydrologic models. The results support the serious consideration of ensemble simulations and predictions created by diverse models in real time flow prediction.

Evaluation of a physics-based distributed hydrologic model for flood forecasting

A fully distributed, physics-based rainfall–runoff model called r.water.fea is applied within the Distributed Model Inter-comparison Project (DMIP) organized by the US National Weather Service. Simulations are performed for two basins, the Illinois River and Blue River in Oklahoma. The r.water.fea model is an event-based model that derives parameters and is calibrated using geospatial data. Longstanding research on the Blue and Illinois River basins resulted in a calibrated model using eight events. In order to draw statistical comparisons, the number of events was augmented for the purposes of DMIP. Model performance is evaluated for the Blue and Illinois for the initial and augmented set of storm events. An important finding related to the stability of calibrated parameters from the original 8 to 18-event storm series was observed. As more events were added to expand the number of storms, parameter values changed only slightly. Beyond the calibration phase, a verification period was also used to test the validity of the calibrated parameters. Consistent results were found between the calibration and verification period. In fact prediction accuracy was better in some cases during the verification period, which adds to the confidence in the calibrated model predictions and the methodology. Interior points are used to identify internal model consistency and achievable accuracy. At the interior points located at Watts and Savoy, predictions were biased at Savoy but had better R 2 values than obtained at Tahlequah in terms of volume and peak. Watts had comparable bias and nearly identical prediction accuracy compared to Tahlequah. During the verification period for the Blue and Illinois, volume predictions had an accuracy of RMSE=17 and 19 mm. Peak discharge in the two basins was predicted with an accuracy of RMSE=105 and 292 m 3/s, respectively. Closer agreement in volume than peak or timing was found in both watersheds, which may indicate the need for improved channel characteristics and routing. The peak discharge predictions achieved by this model in the Illinois during the verification period are biased towards over-prediction by 16% with an R 2 of 0.716. Peak discharge prediction accuracy in the Blue River during the verification period is biased towards under-prediction by 13% with an R 2 of 0.438. The performance demonstrates that geospatial data may be used to parameterize and calibrate a fully distributed physics-based model, and is capable of making reliable predictions at the outlet and at some interior points.

The distributed model intercomparison project (DMIP): motivation and experiment design

The distributed model intercomparison project (DMIP) was formulated as a broad comparison of many distributed models amongst themselves and to a lumped model used for operational river forecasting in the US. DMIP was intended to provide guidance on research and implementation directions for the US National Weather Service as well as to address unresolved questions on the variability of rainfall and its effect on basin response. Twelve groups participated, including groups from Canada, China, Denmark, New Zealand, and the US. Numerous data sets including seven years of concurrent radar-rainfall and streamflow data were provided to participants through web access. Detailed modeling instructions specified calibration and verification periods and modeling points. Participating models were run in ‘simulation’ mode without a forecast component. DMIP proved to be a successful endeavour, providing the hydrologic research and forecasting communities with a wealth of results. This paper presents the background and motivations for DMIP and describes the major project elements.

Application of TOPNET in the distributed model intercomparison project

This paper describes the application of a networked version of TOPMODEL, TOPNET, as part of the Distributed Model Intercomparison Project (DMIP). The model implementation is based on a topographically derived river network with spatially distributed sub-basins draining to each network reach. The river network is mapped from the US National Elevation Dataset Digital Elevation Model (DEM) using procedures that objectively estimate drainage density from geomorphic principles. Rainfall inputs are derived from NEXRAD (radar) for each sub-basin. For each sub-basin, the wetness index distribution is derived from the DEM. The initial model parameters for each sub-basin are estimated using look up tables based on soils (STATSGO) and vegetation (1-km AVHRR). These initial model parameters provide the spatially distributed pattern of parameters at the scale of each sub-basin. Calibration uses a multiplier for each parameter to adjust the parameters while retaining the relative spatial pattern obtained from the soils and vegetation data. Parameter multipliers were calibrated using the shuffled complex evolution algorithm [J. Optim. Theory Appl. 61 (1993)] with the objective to minimize the mean square error between observed and modeled hourly streamflows. We describe the model and calibrated results submitted for all basins for the time periods involved in the DMIP study. We were encouraged by the relatively good performance of the model, especially in comparison to streamflow from smaller interior watersheds not used in calibration and simulated as ungaged basins. The limited resources used to achieve these results show some of the potential for distributed models to be useful operationally.

Impacts of parametric and radar rainfall uncertainty on the ensemble streamflow simulations of a distributed hydrologic model

We diagnose the manner with which rainfall input and parametric uncertainty influence the character of the flow simulation uncertainty in a validated distributed hydrologic model. An extensive Monte Carlo numerical experiment was undertaken for several study watersheds in the southern Central Plains of the United States. It examined the sensitivity of ensemble flow simulations produced by the distributed model HRCDHM to uncertainty in parametric and radar rainfall input. The watersheds are associated with the Distributed Model Intercomparison Project (DMIP) organized by the US National Weather Service Office of Hydrologic Development. The model validated well in DMIP both for watershed outlets and interior points on various scales with Nash-Sutcliffe efficiencies of 0.6–0.9 for hourly flow simulations [J. Hydrol. (2004) 14504, this issue], and we expect that the qualitative nature of the results of this study are of greater applicability than for this model alone. The uncertainty scenarios included: parametric uncertainty involving multiple soil model parameters simultaneously, routing model parameter uncertainty, rainfall uncertainty under two different error distributions, and combined uncertainty in both parameters and input. The flow sensitivities are summarized in terms of a relative measure of the dispersion in the flow ensembles computed for each event, and for several watershed locations consisting of the watershed outlet and additional interior locations. The results consistently show that the flow simulation uncertainty is strongly dependent on catchment scale for all cases of prescribed parametric and radar-rainfall input uncertainty. Simulation uncertainty is significantly reduced for larger scales of distributed model resolution. The consistency of this result across the selected watershed locations allows for the development of scaling relationships between catchment size and the flow uncertainty measure. The derived scaling relationship may be used to infer pronounced small-scale simulation uncertainties in distributed hydrologic model applications. Several fruitful future research directions are identified including the incorporation of model structure uncertainty in the analysis.

Overall distributed model intercomparison project results

This paper summarizes results from the Distributed Model Intercomparison Project (DMIP) study. DMIP simulations from twelve different models are compared with both observed streamflow and lumped model simulations. The lumped model simulations were produced using the same techniques used at National Weather Service River Forecast Centers (NWS-RFCs) for historical calibrations and serve as a useful benchmark for comparison. The differences between uncalibrated and calibrated model performance are also assessed. Overall statistics are used to compare simulated and observed flows during all time steps, flood event statistics are calculated for selected storm events, and improvement statistics are used to measure the gains from distributed models relative to the lumped models and calibrated models relative to uncalibrated models. Although calibration strategies for distributed models are not as well defined as strategies for lumped models, the DMIP results show that some calibration efforts applied to distributed models significantly improve simulation results. Although for the majority of basin-distributed model combinations, the lumped model showed better overall performance than distributed models, some distributed models showed comparable results to lumped models in many basins and clear improvements in one or more basins. Noteworthy improvements in predicting flood peaks were demonstrated in a basin distinguishable from other basins studied in its shape, orientation, and soil characteristics. Greater uncertainties inherent to modeling small basins in general and distinguishable inter-model performance on the smallest basin (65 km 2) in the study point to the need for more studies with nested basins of various sizes. This will improve our understanding of the applicability and reliability of distributed models at various scales.