The National Water Model (NWM) is a key tool for flood forecasting and planning and water management. Key challenges facing NWM include calibration and parameter regionalization when confronted with big data. We present two novel versions of high-resolution (~37 km2) differentiable models (a type of physics-informed machine learning): one with implicit, unit-hydrograph-style routing and another with explicit Muskingum-Cunge routing in the river network. The former predicts streamflow at basin outlets whereas the latter presents a discretized product that seamlessly covers rivers in the conterminous United States (CONUS). Both versions used neural networks to provide multiscale parameterization and process-based equations to provide structural backbone, trained them together (“end-to-end”) on 2,807 basins across CONUS, and evaluated them on 4,997 basins. Both versions show the great potential to elevate future NWMs for extensively calibrated as well as ungauged sites: the median daily Nash-Sutcliffe efficiency (NSE) of all 4,997 basins is improved to around 0.68 from 0.49 of NWM3.0. As they resolve heterogeneity, both greatly improved simulations in the western CONUS and also in the Prairie Pothole Region, a long-standing modeling challenge. The Muskingum-Cunge version further improved performance for basins >10000 km2. Overall, our results show how neural-network-based parameterizations can improve NWM performance for providing operational flood predictions while maintaining interpretability and multivariate outputs. We provide a CONUS-scale hydrologic dataset for further evaluation and use. The modeling system supports the Basic Model Interface (BMI), which allows seamless integration with the next-generation NWM.

Irrigation expansion is often posed as a promising option to enhance food security. Here, we assess the influence of expansion of irrigation, primarily in rural areas of the contiguous United States (CONUS), on the intensification and spatial proliferation of surface freshwater scarcity. Our study shows that the rainfed to irrigation-fed (RFtoIF) transition of water-scarce croplands can impact scarcity in both transitioned and non-transitioned regions, with the magnitude of impact being dependent on multiple factors including local water demand, abstractions in the river upstream, and the buffering capacity of ancillary water sources to cities. Overall, RFtoIF transition will result in an additional 169.6 million hectares or 22% of the total CONUS land area facing moderate or severe water scarcity. Analysis of just the 53 large urban clusters with 146 million residents shows that the transition will result in 97 million urban population facing water scarcity for at least one month per year on average versus 82 million before the irrigation expansion. While these reported figures are subject to simulation uncertainties despite efforts to exercise due diligence, the study unambiguously underscores the need for strategies aimed at boosting crop productivity to incorporate the effects on water availability throughout the entire extent of the flow networks, instead of solely focusing on the local level. The results further highlight that if irrigation expansion is poorly managed, it may increase urban water scarcity, thus also possibly increasing the likelihood of water conflict between urban and rural areas.

Partitioning evapotranspiration (ET) into its primary components, i.e., evaporation (E) and plant transpiration (T), is needed in a range of hydrometeorological applications. Using vegetation index (VI) to obtain spatially resolved T:ET ratio over large areas has emerged as a promising approach in this regard. Here, we assess the effectiveness of this approach in differently managed wheat systems. Results show a weak relation between T:ET and VI in disturbed (i.e., grazed) systems. Flux partitions based on a canonical T:ET vs. VI relation or one derived in a neighboring undisturbed wheat system introduce large errors in disturbed systems, thus underscoring the limits on the transferability of the VI-based ET partitioning approach. The effectiveness of the VI-based approach is found to be related to the strength of correlation between VI and vapor pressure deficit and/or radiation. This correlation metric can help identify settings where the approach is likely to be effective.