Quantifying Combined Impacts of Downscaled Climate Parameters and Land Use Dynamics on Streamflow through Artificial Intelligence

Existing forecasting models often neglect combined, non-stationary feedback mechanisms between downscaled climate parameters (DCP) and landscape dynamics. This study bridges this gap by developing an Integrated Multi-Agent Framework that couples statistically downscaled CMIP6 climate projections with dynamic LULC simulation to forecast daily streamflow from 2021 to 2030 in northern Iraq. The framework employs a Multi-Agent System in which DCP, LULC, and River Discharge (RD) agents interact. The hydrological response was modeled using an ensemble of machine learning algorithms, with the Gradient Boosting Regressor emerging as the superior model that achieved a Nash-Sutcliffe Efficiency (NSE) of 0.92 and Kling-Gupta Efficiency (KGE) of 0.84. Sensitivity analysis revealed a “Climate Dominance” effect: despite a projected 55% loss in tree cover and 11.6% urban expansion, the overwhelming precipitation deficit masked the hydrological signal of land cover changes. These findings demonstrate that integrating multi-agent simulation with machine learning significantly improves forecast reliability in data-scarce regions.