Integrated hydrologic models can simulate coupled surface and subsurface processes but are computationally expensive to run at high resolutions over large domains. Here we develop a novel deep learning model to emulate continental-scale subsurface flows simulated by the integrated ParFlow-CLM model. We compare convolutional neural networks like ResNet and UNet run autoregressively against our novel architecture called the Forced SpatioTemporal RNN (FSTR). The FSTR model incorporates separate encoding of initial conditions, static parameters, and meteorological forcings, which are fused in a recurrent loop to produce spatiotemporal predictions of groundwater. We evaluate the model architectures on their ability to reproduce 4D pressure heads, water table depths, and surface soil moisture over the contiguous US at 1km resolution and daily time steps over the course of a full water year. The FSTR model shows superior performance to the baseline models, producing stable simulations that capture both seasonal and event-scale dynamics across a wide array of hydroclimatic regimes. The emulators provide over 1000x speedup compared to the original physical model, which will enable new capabilities like uncertainty quantification and data assimilation for integrated hydrologic modeling that were not previously possible. Our results demonstrate the promise of using specialized deep learning architectures like FSTR for emulating complex process-based models without sacrificing fidelity.

Long Short-Term Memory (LSTM) networks are a deep learning technology to exploit long-term dependencies in the input-output relationship, which has been observed in the response of groundwater dynamics to atmospheric and land surface processes. We introduced an indirect method based on LSTM networks to estimate monthly water table depth anomalies (wtd_a) across Europe from monthly precipitation anomalies (pr_a). The network has further been optimized by including supplementary hydrometeorological variables, which are routinely measured and available at large scales. The data were obtained from daily integrated hydraulic simulation results over Europe from 1996 to 2016, with a spatial resolution of 0.11° (Furusho-Percot et al., 2019), and separated into a training set, a validation set and a test set at individual pixels. We compared test performances of the LSTM networks locally at selected pixels in eight PRUDENCE regions with random combinations of monthly pr_a, evapotranspiration anomaly, and soil moisture anomaly (θ_a) as input variables. The optimal combination of input variables was pr_a and θ_a, and the networks with this combination achieved average test R^2 between 47.88% and 91.62% in areas with simulated wtd ≤ 3 m. Moreover, we found that introducing θ_a improved the ability of the trained networks to handle new data, indicating the substantial contribution of θ_a to explain groundwater state variation. Therefore, including information about θ_a is beneficial, for instance in the estimation of groundwater drought, and the proposed optimized method may be transferred to a real-time monitoring of groundwater drought at the continental scale using remotely sensed soil moisture observations. Furusho-Percot, C., Goergen, K., Hartick, C., Kulkarni, K., Keune, J. and Kollet, S.: Pan-European groundwater to atmosphere terrestrial systems climatology from a physically consistent simulation, Sci. data, 6(1), 320, doi:10.1038/s41597-019-0328-7, 2019.