Under climate change, robust quantification of future precipitation and runoff evolution is critical for reservoir operation and water resources management. This study investigates the upstream basin of the Longtan Reservoir in Southwest China. A multi-model ensemble framework based on CMIP6 outputs is constructed, integrating bilinear interpolation downscaling, Delta bias correction, and Bayesian Model Averaging (BMA) to generate precipitation projections under SSP1-2.6, SSP2-4.5, and SSP5-8.5 scenarios for 2026–2100. A hybrid LSTM-Transformer model is then developed to simulate runoff responses and diagnose hydroclimatic nonstationarity. Evaluation against observations at 26 rain gauges during 2017–2025 demonstrates that the ensemble precipitation reproduces daily variability well, with a mean correlation of 0.84, confirming its suitability for driving hydrological simulations. The projected precipitation field maintains a stable spatial structure characterized by an east–west gradient, indicating persistent topographic and large-scale climatic controls. However, under high-emission forcing, both precipitation intensity and extreme-event frequency increase substantially, with SSP5-8.5 exhibiting the strongest amplification. All scenarios show increasing trends in annual precipitation, with accelerated growth under high forcing conditions. The LSTM–Transformer model achieves robust performance in runoff simulation, with NSE values of 0.89 (training) and 0.78 (testing), and successfully captures nonlinear rainfall–runoff dynamics and flood peak evolution, indicating strong generalization under nonstationary forcing conditions. Runoff projections reveal a consistent increase in annual mean inflows relative to the historical period (2010–2025), yet the response is non-monotonic with respect to emission intensity. The SSP2-4.5 scenario produces the highest annual runoff, despite not exhibiting the largest precipitation increase. This counterintuitive response is primarily attributed to more favorable hydroclimatic organization, characterized by a balanced precipitation structure and concentrated, sustained flood-season runoff contributions, which enhances rainfall–runoff conversion efficiency. In contrast, SSP5-8.5 exhibits stronger precipitation extremes but a more fragmented runoff regime, with reduced baseline flow contributions and enhanced intra-annual variability, reflecting intensified hydroclimatic instability. The SSP1-2.6 scenario remains constrained by insufficient precipitation input, limiting sustained runoff generation. Further analysis highlights that runoff responses exhibit strong nonlinearity across scenarios and cannot be explained solely by precipitation magnitude or extreme indices. Instead, runoff dynamics are jointly controlled by precipitation temporal–spatial organization and catchment-scale rainfall–runoff transformation efficiency, reflecting increasing system complexity under climate forcing. Despite scenario differences, the intra-annual runoff regime remains monsoon-dominated; however, high-emission conditions significantly intensify hydroclimatic contrasts, characterized by enhanced flood peaks, strengthened dry–wet separation, and increased seasonal asymmetry. Overall, future hydroclimatic conditions in the Longtan Reservoir upstream basin are projected to shift toward higher mean runoff, stronger extremes, and amplified temporal variability. These changes become progressively more pronounced with increasing emission levels, posing substantial challenges for reservoir flood regulation and adaptive water resources management under deepening climate nonstationarity.
更多
分享
引用
