Comprehensive comparison of LSTM and VIC model in river ecohydrological regimes alteration attribution: A case study in Laohahe basin, China

Abstract

Study region: The Laohahe basin, north of China Study focus: In a changing environment, some river’s hydrological regimes have significantly changed and may threatened river ecosystem. Therefore, it is urgent to improve our understanding for the driving mechanism of the evolution of the hydrological regimes. In this study, a comprehensive comparison was carried out to assess the application of the Long Short-Term Memory (LSTM) and the Variable Infiltration Capacity (VIC) model in river ecohydrological regimes alteration attribution. Firstly, the ecohydrological indicators changes were evaluated, including streamflow, ecosurplus, ecodeficit, and hydrological health. Then, the VIC and LSTM models were applied to quantify the climatic and anthropic effects on river ecohydrological indicators, respectively. Finally, the attribution results were compared. New hydrological insights: The research results revealed that (1) the streamflow decreased since 1980 and dramatically decreased after 2000. The ecohydrological indicators also changed significantly (α < 0.05). (2) The alterations were mainly driven by human activities, contributed to more than 80 %. (3) The LSTM and VIC showed little performance difference in quantitative attribution at the monthly and yearly scale, while the LSTM performed a little bad in streamflow simulation at the daily scale. The convenient feature and better simulation capabilities of LSTM provides a new choice for the attribution of ecohydrological regimes evolution, providing a better service in managing local water resources and safeguarding river ecosystems.

Keywords

• Ecohydrological indicators
• LSTM
• Human activities
• Impacts quantification