Hydrological Non-stationarity in SWAT+ model Simulations under Changing Climate and Land Use Cover

Hydrological models are widely used for water resources planning and management under changing environmental conditions. Therefore, it is important to assess the capability of these models to capture non-stationarity arising from climate variability and land-use change. The primary objective of this study is to evaluate the ability of the Soil and Water Assessment Tool (SWAT+) model to capture hydrological non-stationarities in the rainfall–runoff (r–r) mechanism caused by long-term fluctuations in precipitation, temperature, and Land Use/Land Cover (LULC) in Sugar Creek at Milford, Illinois (IL), USA. The analysis begins with the identification and characterization of non-stationarities in long-term hydroclimatic variables (1951–2020) using statistical change-point detection techniques. Temporal variations in the r–r relationship are examined using Analysis of Covariance (ANCOVA), which provides a formal statistical framework to test the linear dependence of the r–r mechanism on its driving variables. However, as hydrological responses are often nonlinear and governed by interacting drivers, ANCOVA alone is insufficient to fully explain how the relative influence of multiple dynamic variables evolves over time. To address this limitation, a machine-learning-based SHAP (Shapley Additive Explanations) analysis is employed to quantify the time-varying contributions of precipitation, temperature, and LULC fractions to streamflow observation, enabling an interpretable decomposition of the changing drivers. Complementing these data-driven analyses, SWAT+ simulations under dynamically varying climate and LULC conditions are analyzed to evaluate the model’s ability to capture hydrological non-stationarity. To examine model structural sensitivity, controlled perturbations of precipitation (±20%), temperature, and LULC are applied, and the resulting changes in major hydrological components are quantified using precipitation elasticity, temperature sensitivity, and LULC sensitivity indices. These diagnostics reveal shifts in process dominance—such as infiltration, percolation, evapotranspiration, and streamflow generation—under altered climatic and land-use regimes. Model calibration is conducted separately for pre-change and post-change periods to assess whether SWAT+ maintains parameter stability across different hydroclimatic states. Variations in optimal parameter values across climate and LULC scenarios are analyzed to quantify parameter uncertainty under non-stationary conditions. Overall, the results reveal substantial temporal variability in parameter sensitivity and demonstrate that fluctuations in precipitation, temperature, and LULC induce nonlinear hydrological responses that challenge the stationarity assumptions embedded in the SWAT+ model structure. The findings underscore the need for dynamic parameterization strategies to more accurately represent evolving watershed processes under changing climate and land-surface conditions.