Added value of downscaling multivariate drought indices for seasonal climatic water-balance assessments

Abstract Reliable climatic water-balance information at seasonal timescales is essential for informing agricultural and hydrological risks. However, coarse-resolution global climate models (GCMs) often struggle to represent local extremes and derived water-related indices, and their predictive skill for these quantities is typically even more limited. This study assesses the added value of downscaled standardized precipitation evapotranspiration index (SPEI) for seasonal climatic water-balance assessments, relative to SPEI derived directly from coarse-resolution seasonal forecasts, and compares the results against downscaled precipitation. We downscaled European Centre for Medium-Range Weather Forecasts SEAS5.1 (∼100 km) precipitation and minimum and maximum temperature outputs to the Copernicus European Regional ReAnalysis resolution (∼5 km) and computed SPEI at this scale. The downscaling was applied over Catalonia and evaluated both across the full domain and at two local locations with different topographic characteristics. Before computing SPEI, we tested a range of bias correction, linear regression, logistic regression, and analog approaches with multiple predictor configurations and methodological setups to identify the most effective approach for transferring forecast quality from native-resolution fields to higher resolution. The analogs method emerges as the most skillful, although the optimal predictor combination varies across variables. Nevertheless, using a common predictor set for multivariate downscaling is necessary, as selecting analogs from the same calendar day preserves inter-variable consistency when computing SPEI. At local scale, downscaled and GCM-based SPEI exhibit no significant differences in correlation or ranked probability skill score, indicating that downscaling effectively transfers large-scale information without systematic skill changes. However, regional maps derived from downscaled SPEI are more spatially detailed and informative, capturing local variability missed in coarser GCM-based fields. Downscaled SPEI outperforms downscaled precipitation in correlation, while categorical skill is similar at short accumulation windows but increases with window length as the index incorporates past observations. Overall, the results highlight the practical value of statistical downscaling for supporting local decision-making, and indicate room for further improvement.