Multisite–Multivariate Calibration and Validation Framework for Uncertainty Reduction in Integrated Hydrological Modelling

Integrated hydrological models provide a powerful framework for simulating the interactions between land surface processes and groundwater systems. However, their application is often hindered by parameter uncertainty, equifinality, and limited observational data, particularly in tropical and data-scarce region. This study presents a multisite–multivariate calibration and validation framework for uncertainty reduction in an integrated hydrological model by simultaneously assimilating observations from multiple locations and multiple hydrological variables. The framework is implemented using the coupled VIC–AMBHAS model and evaluated in the Berambadi watershed, a tropical hard-rock catchment in southern India.

To improve parameter identifiability, time-varying and time-aggregated global sensitivity analyses were first conducted to identify the dominant parameters influencing different hydrological processes across seasons. The model was then calibrated using groundwater levels, remotely sensed evapotranspiration, and soil moisture observations for multiple monitoring locations. Model performance was evaluated using the bounded non-parametric Kling–Gupta Efficiency (KGEnpb) metric. For evapotranspiration (ET), KGEnpb values were greater than 0.14 during calibration and greater than −0.17 during validation, indicating very good to satisfactory model performance. For groundwater levels (GWL), the model performed satisfactorily during the calibration period (KGEnpb > −0.17); however, some grids, particularly in the eastern part of the study area, exhibited unsatisfactory performance. This relatively poorer performance may be attributed to the simplified representation of the aquifer system as a single-layer model, which does not explicitly account for depth varying aquifer properties, particularly specific yield. For surface soil moisture, the model demonstrated very good to satisfactory performance (KGEnpb > −0.17) during both calibration and validation periods. Overall, the results demonstrate that incorporating diverse hydrological observations within a unified multisite–multivariate calibration framework can improve model reliability and predictive capability under varying hydro-climatic conditions. The proposed methodology provides a robust framework for integrated hydrological modelling and can support improved groundwater assessment, climate change impact analysis, and sustainable water resources management, particularly in data-limited environments.