Deep learning for monthly rainfall–runoff modelling: a large-sample comparison with conceptual models across Australia

• DOI: 10.5194/hess-28-1191-2024
• Publication Date: 2024-03-13T12:19:30Z
• Authors: Stephanie R. Clark; Julien Lerat; Jean-Michel Perraud; Peter Fitch
• Keywords: Sample (material)

Abstract. A deep learning model designed for time series predictions, the long short-term memory (LSTM) architecture, is regularly producing reliable results in local and regional rainfall–runoff applications around world. Recent large-sample hydrology studies North America Europe have shown LSTM to successfully match conceptual performance at a daily step over hundreds of catchments. Here we investigate how these models perform monthly runoff predictions relatively dry variable conditions Australian continent. The matches historic data availability also important future water resources planning; however, it provides significantly smaller training datasets than series. In this study, continental-scale comparison (WAPABA model) performed on almost 500 catchments across Australia with aggregated variety catchment sizes, flow conditions, hydrological record lengths. study period covers wet phase followed by prolonged drought, introducing challenges making outside known – that will intensify as climate change progresses. show matched or exceeded WAPABA prediction more two-thirds catchments, largest gains versus occurred large LSTMs struggled less generalise (e.g. under new conditions), few observations due did not demonstrate clear benefit either LSTM.