Accurately predicting runoff (Q) and Suspended Sediment Concentration (SSC) is crucial for the environmental and geological evolution of the Yellow River Delta, a region with numerous oil fields and wetlands. However, accurate prediction of Q and SSC in the Yellow River Delta, characterized by the intricate interplay of high-frequency and low-frequency environmental factors, poses significant scientific challenges. This study introduces a novel framework focused on predicting Q and SSC exclusively using irrigation data. In this framework, Spatial Autocorrelation measures potential interactions among irrigation points, while Q-learning Swarm Optimization identifies high-impact irrigation areas. Ridge Regression (RR) is then applied to predict Q and SSC using the irrigation data. To comprehensively assess factors beyond irrigation, Empirical Mode Decomposition decomposes the residuals. The periodicity of each Intrinsic Mode Function (IMF) is ascertained using a zero-crossing method. IMF1, IMF2, and IMF3 are identified as seasonal signals, capturing the alternation between dry and rainy seasons, as well as inter-annual changes. Subsequently, Gated Recurrent Units (GRU) are further employed to extract this IMF information, expanding the dataset using bootstrap techniques. In the culmination of this research, the proposed framework emerges as an amalgamation of the RR model and the GRU models. Numerical examples substantiate the efficacy of this approach, demonstrating its ability to provide one-month ahead forecasts for Q and SSC with a Mean Absolute Percentage Error of 15% and 12%, respectively.