An official website of the United States government
Abstract With an increasing number of continental‐scale hydrologic models, the ability to evaluate performance is key to understanding uncertainty and making improvements to the model(s). We hypothesize that any model, running a single set of physics, cannot be “properly” calibrated for the range of hydroclimatic diversity as seen in the contenintal United States. Here, we evaluate the NOAA National Water Model (NWM) version 2.0 historical streamflow record in over 4,200 natural and controlled basins using the Nash‐Sutcliffe Efficiency metric decomposed into relative performance, and conditional, and unconditional bias. Each of these is evaluated in the contexts of meteorologic, landscape, and anthropogenic characteristics to better understand where the model does poorly, what potentially causes the poor performance, and what similarities systemically poor performing areas share. The primary objective is to pinpoint traits in places with good/bad performance and low/high bias. NWM relative performance is higher when there is high precipitation, snow coverage (depth and fraction), and barren area. Low relative skill is associated with high potential evapotranspiration, aridity, moisture‐and‐energy phase correlation, and forest, shrubland, grassland, and imperviousness area. We see less bias in locations with high precipitation, moisture‐and‐energy phase correlation, barren, and grassland areas and more bias in areas with high aridity, snow coverage/fraction, and urbanization. The insights gained can help identify key hydrological factors underpinning NWM predictive skill; enforce the need for regionalized parameterization and modeling; and help inform heterogenous modeling systems, like the NOAA Next Generation Water Resource Modeling Framework, to enhance ongoing development and evaluation.
more »
« less
This content will become publicly available on September 17, 2026
Interrogating process deficiencies in large-scale hydrologic models with interpretable machine learning
Large-scale hydrologic models are increasingly being developed for operational use in the forecasting and planning of water resources. However, the predictive strength of such models depends on how well they resolve various functions of catchment hydrology, which are influenced by gradients in climate, topography, soils, and land use. Most assessments of hydrologic model uncertainty have been limited to traditional statistical methods. Here, we present a proof-of-concept approach that uses interpretable machine learning techniques to provide post hoc assessment of model sensitivity and process deficiency in hydrologic models. We train a random forest model to predict the Kling–Gupta efficiency (KGE) of National Water Model (NWM) and National Hydrologic Model (NHM) streamflow predictions for 4383 stream gauges in the conterminous United States. Thereafter, we explain the local and global controls that 48 catchment attributes exert on KGE prediction using interpretable Shapley values. Overall, we find that soil water content is the most impactful feature controlling successful model performance, suggesting that soil water storage is difficult for hydrologic models to resolve, particularly for arid locations. We identify nonlinear thresholds beyond which predictive performance decreases for NWM and NHM. For example, soil water content less than 210 mm, precipitation less than 900 mm yr−1, road density greater than 5 km km−2, and lake area percent greater than 10 % contributed to lower KGE values. These results suggest that improvements in how these influential processes are represented could result in the largest increases in NWM and NHM predictive performance. This study demonstrates the utility of interrogating process-based models using data-driven techniques, which has broad applicability and potential for improving the next generation of large-scale hydrologic models.
more »
« less
- PAR ID:
- 10637330
- Publisher / Repository:
- European Geosciences Union
- Date Published:
- Journal Name:
- Hydrology and Earth System Sciences
- Volume:
- 29
- Issue:
- 18
- ISSN:
- 1607-7938
- Page Range / eLocation ID:
- 4457 to 4472
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Climate warming in alpine regions is changing patterns of water storage, a primary control on alpine plant ecology, biogeochemistry, and water supplies to lower elevations. There is an outstanding need to determine how the interacting drivers of precipitation and the critical zone (CZ) dictate the spatial pattern and time evolution of soil water storage. In this study, we developed an analytical framework that combines intensive hydrologic measurements and extensive remotely-sensed observations with statistical modeling to identify areas with similar temporal trends in soil water storage within, and predict their relationships across, a 0.26 km 2 alpine catchment in the Colorado Rocky Mountains, U.S.A. Repeat measurements of soil moisture were used to drive an unsupervised clustering algorithm, which identified six unique groups of locations ranging from predominantly dry to persistently very wet within the catchment. We then explored relationships between these hydrologic groups and multiple CZ-related indices, including snow depth, plant productivity, macro- (10 2 ->10 3 m) and microtopography (<10 0 -10 2 m), and hydrological flow paths. Finally, we used a supervised machine learning random forest algorithm to map each of the six hydrologic groups across the catchment based on distributed CZ properties and evaluated their aggregate relationships at the catchment scale. Our analysis indicated that ~40–50% of the catchment is hydrologically connected to the stream channel, lending insight into the portions of the catchment that likely dominate stream water and solute fluxes. This research expands our understanding of patch-to-catchment-scale physical controls on hydrologic and biogeochemical processes, as well as their relationships across space and time, which will inform predictive models aimed at determining future changes to alpine ecosystems.more » « less
-
Discharge values from the National Water Model (NWM) were compared to USGS stream gage discharge observations for the suburban Red Clay Creek watershed (drainage area ~140 km2 and mixed land-use), in Pennsylvania and Delaware, from 2016 to 2018. 18-hour retrospective simulations from the NWM were used with concurrent hourly USGS discharge observations from three locations along the Red Clay Creek. Results indicate that the mean of discharge estimates from the NWM and from USGS observations significantly differed and that the NWM generally underestimates low-flow conditions and overestimates high-flow conditions. Watershed size also impacted NWM performance (with performance degrading in smaller watersheds). A meteorological analysis determined that convective rainfall events were associated with 66% of the largest differences between NWM discharge estimates and USGS observations while mid-latitude cyclone stratiform precipitation events accounted for the other 34%. Lastly, of the largest 15 differences between the NWM and observations, 13 occurred with pre-cursor soil moisture that was below the mean (dry soil conditions), in conjunction with heavy rainfall. Given the NWM’s recent operational implementation, and its status as Prototype guidance, the results of this study present specific geographical and climatological findings that can aid in the NWM’s continued validation and improvement for similar regions.more » « less
-
Abstract Quantitative estimations of ecohydrological water partitioning into evaporation and transpiration remains mostly based on plot‐scale investigations that use well‐instrumented, small‐scale experimental catchments in temperate regions. Here, we attempted to upscale and adapt the conceptual tracer‐aided ecohydrology model STARRtropics to simulate water partitioning, tracer, and storage dynamics over daily time steps and a 1‐km grid larger‐scale (2565 km2) in a sparsely instrumented tropical catchment in Costa Rica. The model was driven by bias‐corrected regional climate model outputs and was simultaneously calibrated against daily discharge observations from 2 to 30 years at four discharge gauging stations and a 1‐year, monthly streamwater isotope record of 46 streams. The overall model performance for the best discharge simulations ranged in KGE values from 0.4 to 0.6 and correlation coefficients for streamflow isotopes from 0.3 to 0.45. More importantly, independent model‐derived transpiration estimates, point‐scale residence time estimates, and measured groundwater isotopes showed reasonable model performance and simulated spatial and temporal patterns pointing towards an overall model realism at the catchment scale over reduced performance in the headwaters. The simulated catchment system was dominated by low‐seasonality and high precipitation inputs and a marked topographical gradient. Climatic drivers overrode smaller, landcover‐dependent transpiration fluxes giving a seemingly homogeneous rainfall‐runoff dominance likely related to model input bias of rainfall isotopes, oversimplistic Potential Evapotranspiration (PET) estimates and averaged Leaf Area Index (LAI). Topographic influences resulted in more dynamic water and tracer fluxes in the headwaters that averaged further downstream at aggregated catchment scales. Modelled headwaters showed greater storage capacity by nearly an order of magnitude compared to the lowlands, which also favoured slightly longer residence times (>250 days) compared to superficially well‐connected groundwater contributing to shorter streamflow residence times (<150 days) in the lowlands. Our findings confirm that tracer‐aided ecohydrological modelling, even in the data‐scarce Tropics, can help gain a first, but crucial approximation of spatio‐temporal dynamics of how water is partitioned, stored and transported beyond the experimental catchment scale of only a few km2.more » « less
-
Abstract Understanding how land cover change will impact water resources in snow‐dominated regions is of critical importance as these locations produce disproportionate runoff relative to their land area. We coupled a land cover evolution model with a spatially explicit, physics‐based, watershed process model to simulate land cover change and its impact on the water balance in a 5.0 km2headwater catchment spanning the alpine–subalpine transition on the Colorado Front Range. We simulated two potential futures both with greater air temperature (+4°C/century) and more precipitation (+15%/century, MP) or less precipitation (−15%/century, LP) from 2000 to 2100. Forest cover in the catchment increased from 72% in 2000 to 84% and 83% in 2050 and to 95% and 92% in 2100 for MP and LP, respectively. Surprisingly, increases in forest cover led to mean increases in annual streamflow production of 12 mm (6%) and 2 mm (1%) for MP and LP in 2050 with an annual control streamflow of 208 mm. In 2100, mean streamflow production increased by 91 mm (44%) and 61 mm (29%) for MP and LP. This result counters previous work as runoff production increased with forested area due to decreases in snow wind‐scour and increases in drifting leeward of vegetation, highlighting the need to better understand the impacts of forest expansion on the spatial pattern of snow scour, deposition and catchment effective precipitation. Identifying the hydrologic response of mountainous areas to climate warming induced land cover change is critically important due to the potential water resources impacts on downstream regions.more » « less
