The WRF‐Hydro hydrological model has been used in many applications in the past with some level of history matching in the majority of these studies. In this study, we use the iterative Ensemble Smoother (iES), a powerful parameter estimation methodology implemented in the open‐source PEST++ software. The iES provides an ensemble solution with an uncertainty bound instead of a single best estimate which has been the common approach in the previous WRF‐Hydro studies. We discuss the importance of accounting for observation noise which results in a wider spread in the model solution. We investigate the impact of constructing objective functions by differentially weighting the observations to tune the model response toward model outputs appropriate for a specific application. Results confirm the necessity of differentially weighting the observations before calculation of the objective function as the optimization algorithm struggles with calculating parameter updates with uniform weighting. We also show that we achieve better model performance in terms of verification metrics with higher emphasis on the high flow events, when the objective function is tuned toward an application where the extreme events are of importance. We then investigate the impact of estimating more parameters, in particular we estimate a larger number of snow parameters. Results show a large improvement in the model performance. In summary, our study demonstrates the efficacy of employing iES alongside differential weighting of observations, highlighting its potential to enhance hydrological model parameter estimation.
A soil erosion and sediment transport model (WRF-Hydro-Sed) is introduced to WRF-Hydro. As a process-based, fully distributed soil erosion model, WRF-Hydro-Sed accounts for both overland and channel processes. Model performance is evaluated using observed rain gauge, streamflow, and sediment concentration data during rainfall events in the Goodwin Creek Experimental Watershed in Mississippi, USA. Both streamflow and sediment yield can be calibrated and validated successfully at a watershed scale during rainfall events. Further discussion reveals the model’s uncertainty and the applicability of calibrated hydro- and sediment parameters to different events. While an intensive calibration over multiple events can improve the model’s performance to a certain degree compared with single event-based calibration, it might not be an optimal strategy to carry out considering the tremendous computational resources needed.
Snow influences land–atmosphere interactions in snow-dominated areas, and snow melt contributes to basin streamflows. However, estimating snowpack properties such as the snow depth (SD) and snow water equivalent (SWE) from land surface model simulations remains a challenge. This is, in part, due to uncertainties in the atmospheric forcing variables, which propagate into hydrological model predictions. This study implements the Weather Research and Forecasting (WRF)-Hydro framework with the Noah-Multiparameterization (Noah-MP) land surface model in the NOAA’s National Water Model (NWM) version 2.0 configuration to estimate snow in a single column and subsequently the streamflow across the Aroostook River’s sub-basins in Maine for water years (WY) 2014–2016. This study evaluates how differences between two atmospheric forcing datasets, the North American Land Data Assimilation version 2 (NLDAS-2) and in situ (Station), translate into differences in the simulation of snow. NLDAS-2 was used as the meteorological forcing in the retrospective NWM 2.0 simulations. The results from the single-column study showed that differences in the simulated SWE and SD were linked to differences in the 2 m air temperature (T2m), which influenced the precipitation partitioning of rain and snow, as parameterized in Noah-MP. The negative mean bias of −0.7 K (during the accumulation period) in T2m for NLDAS-2, compared to the Station forcing, was a major factor that contributed to the positive mean bias of +52 mm on average in the peak SWE in the NLDAS-2-forced Noah-MP simulation during the study period. The higher T2m values at the Station led to higher sensible heat fluxes towards the snowpack, which led to a higher amount of net energy at the snow’s surface and melt events during the accumulation season in Station-forced Noah-MP simulations. The results from the retrospective NWM version 2.0′s simulation in the basin showed that the streamflow estimates were closer to the United States Geological Survey gage observations at the two larger sub-basins (NSE = 0.9), which were mostly forested, compared to the two smaller sub-basins (NSE ≥ 0.4), which had more agricultural land-use. This study also showed that the spring snowmelt timing was captured quite well by the timing of the decline in the simulated SWE and SD, providing an early indication of melt in most sub-basins. The simulated fractional snow cover area (fSCA) however provided less information about the changes in snow or onset of snowmelt as it was mostly binary (full snow cover in winter), which differed from the more realistic fSCA values shown by the Moderate Resolution Imaging Spectroradiometer.