Abstract
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 long short-term memory network, a deep learning approach for learning both spatial and temporal patterns, to predict streamflow at 226 stream gauges across southwestern Canada. The model is forced by gridded climate reanalysis data and trained to predict observed daily streamflow between 1980 and 2015. To interpret the model's learning of both spatial and temporal patterns, we introduce a set of experiments with evaluation metrics to track the model's response to perturbations in the input data. The model performs well in simulating 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 time series to obtain relatively warmer and colder input data, the modelled indicator of freshet timing and peak flow changes in accordance with the transition timing from below- to above-freezing temperatures. We also demonstrate that modelled August streamflow in partially glacierized basins is sensitive to perturbations in August temperature, and that this sensitivity increases with glacier cover. The results demonstrate the suitability of a convolutional long short-term memory network architecture for spatiotemporal hydrological modelling, making progress towards interpretable deep learning hydrological models.
Highlights
The use of deep learning (DL) has gained traction in geophysical disciplines as an active field of exploration in efforts to maximize the use of growing in situ and remote sensing datasets (Bergen et al, 2019; Reichstein et al, 2019; Shen, 2018)
We find that the convolutional neural networks (CNNs)–Long short-term memory (LSTM) model outperforms this simple linear benchmark, achieving a greater Nash–Sutcliffe efficiency (NSE) at 222 out of 226 stations
This study investigated the applicability of a sequential CNN–LSTM model for regional-scale hydrological modelling, where the model was forced by gridded climate data to predict streamflow at multiple stream gauge stations simultaneously
Summary
The use of deep learning (DL) has gained traction in geophysical disciplines as an active field of exploration in efforts to maximize the use of growing in situ and remote sensing datasets (Bergen et al, 2019; Reichstein et al, 2019; Shen, 2018). Notable are DL models which have been found to outperform traditional hydrological models applied at regional scale, including those for streamflow prediction at daily temporal scale (Kratzert et al, 2018, 2019b), at hourly temporal scale (Gauch et al, 2021), and at ungauged basins (Kratzert et al, 2019a). These recent DL-based studies have emphasized the development of lumped hydrological models with inputs that are aggregated to the basin level. A key opportunity exists, to develop a DL hydrological model which can utilize spatially discretized forcing data at regional scale
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
