Abstract
Lake Urmia, one of the world's largest hypersaline lakes, has experienced severe drought in recent decades. This study investigates the combined impacts of agricultural expansion and climate variability on river inflows from 1985 to 2020. A hybrid framework incorporating statistical models and Convolutional Neural Networks was employed to estimate river discharge and disentangle the effects of hydroclimatic and anthropogenic drivers. Results indicate a persistent snow drought beginning in the late 1990s, concurrent with exceeding fourfold increase in irrigated lands. Scenario-based analysis, restoring key parameters to pre-1999 levels revealed that reverting agricultural water use was the dominant factor driving changes in river inflows, accounting for approximately 66% (95% CI: 56%-76%) of the total impact. In contrast, restoring precipitation and evaporation contributed 25% (95% CI: 18%-33%) and 9% (95% CI: 7%-12%), respectively, while restoring both simultaneously explained 34% (95% CI: 26%-43%) of the change. These results underscore the primary role of agricultural water demand amplified by declining snowpack and climatic shifts in altering basin hydrology. The findings highlight the urgent need for integrated water resource management, with a focus on climate adaptation, snowpack monitoring, and sustainable agricultural practices to address ongoing environmental degradation and ensure long-term water security.