Abstract
Hydrological prediction and attribution in mountain to plain transition basins remain challenging because topographic zonation, groundwater surface water connectivity, and human regulation can induce structural bias and weaken closure consistency. This study develops an integrated framework that combines an enhanced partitioned, connectivity weighted, and closure consistent Budyko scheme with interpretable deep learning. Annual connectivity weights are derived from the baseflow index and embedded in the partitioned closure accounting. Climate and human contributions are cross validated using an elasticity method and Shapley decomposition under scenario designs. Key predictors are screened with the optimal Geodetector. Runoff prediction is performed using an ensemble of a Convolutional Neural Network, a Bidirectional Long Short-Term Memory network, and Adaptive Boosting (CNN-BiLSTM-AdaBoost), and SHAP is applied to the prediction model to quantify driver contributions and diagnose threshold type responses. Results show that the annual runoff change point occurs in 1993. Both attribution approaches identify climate change as the dominant driver, with contributions of 82.4% and 85.5%, while human activities contribute 17.6% and 14.5%. SPEI_M, SPEI_P, SnSPI_P and Rx5day are identified as the primary meteorological drivers. The ensemble model achieves NSE values of 0.822 for training and 0.908 for testing. SHAP indicates relative contributions of 29.07% and 21.98% for SPEI_M and SPEI_P, respectively, and event scale precipitation shows a notable compensating effect around Rx5day of 49.2 mm. Spatially, mountainous zones are more sensitive to precipitation, enhanced evapotranspiration in the plains reduces outflow, and the integrated connectivity weights during the target period favor mountainous pathways. Overall, the framework unifies closure consistent attribution and interpretable prediction, providing a robust basis for diagnosing asymmetric hydrological responses and supporting management assessment in transition basins.
扫码关注我们
求助内容:
应助结果提醒方式:
