Cumulative and offsetting effects of Streamflow Response to Climate change and Large Reservoir Group in the Jinsha River Basin, China

Study Region

Jinsha River Basin (JRB), situated in the upper sections of the Yangtze River basin, China

Study Focus

The Jinsha River Basin (JRB), renowned as the world's largest hydropower base. In this study, we proposed a novel approach for separating the reservoir's influence on streamflow and the response mechanisms in diverse watersheds. We integrated XGBoost (the eXtreme Gradient Boosting), CNN-LSTM (Convolutional Neural Networks, Long Short-Term Memory) and Informer with 6 hydrological stations data for streamflow reconstruction. Traditional hydrological model SWAT (Soil and Water Assessment Tool) served as a benchmark.

New Hydrological Insights for the Region

Reservoir groups were exerting average 40.7 % influence on streamflow after 2010, altering the streamflow with maximum 20 % in the upstream, 42 % midstream and 60 % downstream in August in JRB. Also, reservoir regulation significantly shifted the seasonal streamflow pattern (especially low water season) in the JRB. Under extreme wet (SDI≥2) in flood season (June-November), the maximum reservoir regulated streamflow offsetting varied between 14.3 % and 28.6 % (2000–4000 m³/s out of Q_max=14000 m³/s). Conversely, under extreme drought (SDI≤-2) in the low water season, reservoir regulated streamflow accumulation accounted for 25–44.4 % (100–800 m³/s out of Q_max=1800 m³/s). Moreover, the XGBoost model can provide a new way for streamflow reconstruction and forecast, particularly in the quantification of the reservoir groups under climate change. Reservoir storage efficiency was more obvious in the midstream, especially after 2015 (with normalized difference (S_{reconstruction} – S_observation)> 0 more frequent), while released more in the downstream after 2010 (with normalized difference<0 more frequent). The streamflow reconstruction simulation had a stronger correlation with the catchment streamflow than large-scale atmospheric circulation indices. In summary, our study demonstrates models forecast reliability in streamflow reconstruction under climate change and reservoirs, providing a new reasoning approach for extreme streamflow event early warning resulting from large reservoir operations.