Hydrological models are subject to natural variability in forcing conditions, yet traditional global sensitivity analysis (GSA) often overlooks these uncertainties, potentially limiting the validity of results to specific conditions. To address this, we propose an approach that accounts for the uncontrollable nature of rain forcing variability. We treat the model parameters' Sobol' indices as random variables dependent on forcing variability and use polynomial chaos expansion metamodels to reduce the computational cost of their estimation. We apply this methodology to study soil moisture sensitivity in the physically-based distributed hydrological model PESHMELBA. Our results show that parameter rankings vary with forcing conditions. To consolidate these diverging GSA results, we propose a unique ranking based on aggregated sensitivity indices that accounts for parameter contributions across the entire domain of forcing conditions. This approach enhances the robustness of GSA to natural variability in forcings, thereby improving the reliability of subsequent GSA-based decisions.