Benchmarking data-driven rainfall–runoff models in Great Britain: a comparison of long short-term memory (LSTM)-based models with four lumped conceptual models

Abstract. Long short-term memory (LSTM) models are recurrent neural networks from the field of deep learning (DL) which have shown promise for time series modelling, especially in conditions when data abundant. Previous studies demonstrated applicability LSTM-based rainfall–runoff modelling; however, LSTMs not been tested on catchments Great Britain (GB). Moreover, opportunities exist to use spatial and seasonal patterns model performances improve our understanding hydrological processes examine advantages disadvantages simulation. By training two LSTM architectures across a large sample 669 GB, we demonstrate that Entity Aware (EA LSTM) simulate discharge with median Nash–Sutcliffe efficiency (NSE) scores 0.88 0.86 respectively. We find outperform suite benchmark conceptual models, suggesting an opportunity additional refine models. In summary, show largest performance improvements north-east Scotland south-east England. The England remained difficult model, part due inability configured this study learn groundwater processes, human abstractions complex percolation properties hydro-meteorological variables typically employed modelling.