Evaluation and interpretation of convolutional-recurrent networksfor regional hydrological modelling

Abstract. Deep learning has emerged as a useful tool across geoscience disciplines; however, there remain outstanding questions regarding the suitability of unexplored model architectures and how to interpret model learning for regional scale hydrological modelling. Here we use a convolutional-recurrent network, a deep learning approach for learning both spatial and temporal patterns, to predict streamflow at 226 stream gauges across the region of southwestern Canada. The model is forced by gridded climate reanalysis data and trained to predict observed daily streamflow between 1979 and 2015. To interpret the model learning of both spatial and temporal patterns, we introduce two experiments with evaluation metrics to track the model's response to perturbations in the input data. The model performs well in simulating the daily streamflow over the testing period, with a median Nash-Sutcliffe Efficiency (NSE) of 0.68 and 35 % of stations having NSE > 0.8. When predicting streamflow, the model is most sensitive to perturbations in the input data prescribed near and within the basins being predicted, demonstrating that the model is automatically learning to focus on physically realistic areas. When uniformly perturbing input temperature timeseries to obtain relatively warmer and colder input data, the modelled freshet timing and intensity changes in accordance with the transition timing from below- to above-freezing temperatures. The results demonstrate the suitability of a convolutional-recurrent network architecture for spatiotemporal hydrological modelling, making progress towards interpretable deep learning hydrological models.

Перевод пока недоступен