RESUMO
Cross-border impact assessment of cascade reservoir operation on hydrological regimes is a vital prerequisite for the sustainable development and management of transboundary waters. However, assessment based on traditional hydrological modeling for transboundary rivers is limited by the availability of meteorological and hydrological data. In this study, a combined data-driven model (CV-LSTM) was built to simulate natural runoff without dam construction in the Upper Mekong River. Then, the simulated natural runoff was compared against the observation runoff influenced by dam operation to assess the impacts of reservoirs on flood and drought events. The research results are as follows: (1) CV-LSTM improved simulation performance by effectively utilizing spatial information from more extensive and diversified data, and it could overcome the limitation of classic data-driven models in that the spatial heterogeneity of input variables cannot be sufficiently considered. (2) Reservoir operation decreased the annual streamflow of the Upper Mekong River by 1.07% during the 2001-2016. In particular, with the operation of two mega reservoirs (Xiaowan and Nuozadu) after 2008, the annual streamflow decreased by 3.95%. (3) The upstream reservoirs exerted significant runoff regulative effects on the Lower Mekong during 2001-2016. Drought duration and severity significantly decreased at the Chiang Sean hydrological station, flood frequency decreased by 11%, and the mean day of flood occurrence decreased by 30%. This study developed an innovative approach, CV-LSTM, based on open-source spatial information, which could effectively analyze the cross-border impact of cascade reservoir operation. The results provide new insights into the quantitative assessment of the transboundary influence of upstream-downstream runoff change induced by cascade dams in international rivers.