Deciphering Reservoir Storage Variability: A Multi-Reservoir Assessment of Hydroclimatic Drought Responses in Northwestern Türkiye

Climate change, rising water demand, and land use change are increasing the complexity of reservoir management. This study evaluates reservoir storage responses to hydroclimatic drought across 12 reservoirs in the Marmara, Susurluk, Meriç–Ergene, and Northern Aegean basins of northwestern Türkiye. Data from 27 meteorological stations were used to calculate the SPI, SPEI, and RDI at 1-, 3-, 6-, 9-, and 12-month scales. Reservoir storage during 2013–2022 was evaluated seasonally using Random Forest (RF), Linear Models (LMs), and Generalized Additive Models (GAMs) under leave-one-year-out cross-validation (LOYO-CV). The agricultural-area NDVI and reservoir characteristics provided contextual information rather than direct measures of water withdrawal. Drought indices at 6–12-month scales were selected as dominant predictors for 11 of the 12 reservoirs. RF was the best performing model in 20 of 48 reservoir–season combinations, compared with 16 for LMs and 12 for GAMs, although performance varied among reservoirs and seasons. Basin mean and reservoir local drought series were strongly correlated (mean r = 0.88), whereas Local-IDW representation did not systematically improve performance. Nevertheless, dominant predictors changed in eight reservoirs, highlighting local hydroclimatic heterogeneity. The agricultural NDVI showed considerable temporal and spatial variability but was not interpreted as evidence of reservoir-specific withdrawals or anthropogenic causation. Overall, reservoir drought sensitivity reflects cumulative hydroclimatic conditions and pronounced seasonal and reservoir-specific variability. The integrated cross-validation and spatial sensitivity framework provides a rigorous basis for reservoir drought assessment and adaptive water resource management.