基于模态筛选与LSTM-Transformer的渭河流域月径流预报OA

Accurate monthly runoff forecasting is crucial for flood and drought disaster prevention and wa-ter resources regulation in river basins.To improve the accuracy of runoff forecasting under changing en-vironments,we proposed a TP-LT monthly runoff forecasting framework that integrates TVF-EMD de-composition,PCMCI causal analysis,and the LSTM-Transformer hybrid model.Key driving factors in-cluding modal components of the runoff series,meteorological variables,and atmospheric circulation indi-ces were considered to develop a forecasting model that can capture both short-term and long-term de-pendencies in runoff.Then the model was applied to the forecasting of monthly runoff in the Wei River Basin.The results show that the medium-and high-frequency IMF components can reflect the key fluctu-ation characteristics of monthly runoff.Meteorological factors such as precipitation and relative humidity,along with atmospheric circulation indices including Nino3.4 and AO,are the main meteorological driv-ers.The comparative analysis indicates that the LSTM-Transformer model outperforms any individual model.When only meteorological drivers are used as inputs,the NSE of all models during the validation period is below 0.45,indicating insufficient accuracy.After introducing the key modal drivers,the NSE of the LSTM-Transformer model increases to 0.63-0.78,KGE reaches 0.62-0.82,and R exceeds 0.80.Moreover,the RMSE and MAE are significantly reduced compared with all individual models,and the model captures the timing and process of peak flows more accurately during the flood season.In sum-mary,the TP-LT monthly runoff forecasting framework can effectively improve forecasting performance and provide a reliable method for runoff forecasting in river basins under changing environments.