Multi-model ensemble of statistically downscaled GCMs over southeastern South America: historical evaluation and future projections of daily precipitation with focus on extremes

High-resolution rainfall information is of great value, particularly over southeastern South America (SESA) where the observed and projected climate changes pose a substantial threat to socio-economic activities and the hydrological sector. Consequently, this work focuses on the construction of an unprecedented multi-model ensemble of statistically downscaled (ESD) global climate models (GCMs) for daily precipitation. Different statistical techniques were employed - including analogs, stochastic versions of regression-based models involving neural networks and generalised linear models and linear regressions conditioned by weather types - and a variety of CMIP5 and CMIP6 models. In general, most of the models added value in the representation of the main features of daily precipitation, especially in the spatial and intra-annual variability of extremes. The statistical models were sensible to the driving GCMs, although the ESD family choice also introduced differences among the simulations. The ESD projections depicted increases in mean precipitation associated with a rising frequency of extreme events - mostly during the warm season - following the observed trends over SESA. Change rates were consistent among downscaled models up to mid-21st century, when model spread started to emerge. Furthermore, these projections were compared to the available CORDEX-CORE RCM simulations, evidencing a consistent agreement between statistical and dynamical downscaling procedures in terms of the sign of the changes, presenting the main differences in their intensity. Overall, this study evidences the potential of statistical downscaling in a changing climate and contributes to its undergoing development over SESA.

Automated summary

This study focuses on the construction of an unprecedented multi-model ensemble of statistically downscaled global climate models (GCMs) for daily precipitation. The results show that statistical downscaling adds value in representing the main features of daily precipitation and can predict future changes in precipitation. Furthermore, the statistical downscaling projections show consistent agreement with the dynamical downscaling results in terms of change direction but have differences in intensity.