Assessment of the impact of climate change on streamflow of Ganjiang River catchment via LSTM-based models

Abstract

Study region: Ganjiang River catchment, China. Study focus: Quantifying the effects of the impacts climate change on streamflow is of great importance for regional water resources management. In this study, four LSTM-based models, i.e., LSTM, Stack-LSTM, Bi-LSTM and CNN-LSTM, were constructed to assess hydrological changes under future climate change based on the bias-corrected Coupled Model Intercomparison Project Phase 6 meteorological data from four shared socioeconomic pathways (SSPs). New hydrological insights for the study region: (1) The Bi-LSTM achieves the best streamflow simulation performance during the validation period of 2006–2016, followed by the LSTM, Satck-LSTM and CNN-LSTM. The climate will get warmer and wetter in future scenarios with mean annual precipitation and temperature increasing by 3.0–6.2% and 8.3–13.4% (compared to baseline period), respectively. (2) The predicted streamflow will have a decrease of 1.5–16.5% during 2026–2075 under all SSP scenarios. The results of LSTM, Stack-LSTM, and Bi-LSTM models show a decrease of 0.27–50% in November to March, indicating that the dry season will become even drier in the future. These results reveal that the hydrological regime of the Ganjiang River is likely to change and will be characterized by greater seasonal uncertainty and more potential for extreme events due to significant warming and wetting over the future periods.

Keywords

• Climate change
• Monthly runoff prediction
• LSTM-based models
• CMIP6
• Ganjiang River catchment