Abstract: To address the limitations of low prediction accuracy and poor generalization ability in traditional runoff forecasting methods, this study proposes a hybrid runoff forecasting model integrating Successive Variational Mode Decomposition (SVMD), multi-strategy Improving Cooperation Search Algorithm (ICSA), and spatiotemporal error integrated correction.Firstly, the method decomposes the runoff time series into relatively independent subsequences using SVMD.Secondly, each subsequence is then forecast using Least Squares Support Vector Regression (LSSVR), with its parameters optimized by a CSA enhanced via sinusoidal initialization, dynamic communication, and random walk mutation strategies, significantly improving global search capability and convergence stability.Finally, the initial forecasts are summed and further refined by spatiotemporal error integrated correction to further reduce residual errors and ensure reliability.Validated in Chitan Reservoir, Fujian Province, China, the model demonstrably outperformed traditional methods (LSTM, ELM, SVR, LSSVR), it achieved higher RMSE, MAE, CC and NSE values values.The NSE values for the forecast period of 1~4 days are 0.986, 0.982, 0.976 and 0.967 respectively, showing higher accuracy and stability.Validity tests confirmed its capability to accurately capture complex nonlinear runoff relationships and reduce prediction bias, providing a valuable solution for high-precision runoff forecasting under changing conditions.