Assessment of Large-Scale Reservoirs' Impact on the Local Precipitation

Reservoir operations have complex and profound impacts on local climate, particularly precipitation. Quantifying this impact is challenging because it requires the reconstruction of natural precipitation prior to reservoir operation. Instead of assuming that the natural variability of the contrast region and the study region is identical, this study develops an interpretable machine learning model to investigate relationships between precipitation-influencing factors and precipitation itself, including both stable components (sum of trend and seasonality from STL decomposition) and random components (residuals after removing trend and seasonality), which is then used to forecast natural precipitation in the absence of reservoir operation. The application in the contrast region verifies the forecast's accuracy, even in mountainous areas. The proposed method is used to analyze the impact of three large-scale reservoirs along the Yangtze River on local precipitation, collectively having a total storage capacity of 17.86 × 10⁹ m³. The results indicate that reservoir operation leads to a 14% increase in the trend and seasonal components of precipitation, which would be underestimated by previous methods. In addition, there is a noticeable shift in the precipitation center toward the reservoir. Further comparisons suggest that reservoir operation shifts the key influencing factors of local precipitation patterns from those characterized by high variability to those characterized by low variability. Changes in soil water retention capacity likely play a significant role in these precipitation changes. We also found a significant positive coupling between soil moisture and precipitation in the study area, which has been a focal point of recent research. These findings provide new insights into the mechanisms through which reservoir construction impacts precipitation.