Downscaling CORDEX Through Deep Learning to Daily 1 km Multivariate Ensemble in Complex Terrain

Abstract High spatio‐temporal resolution near‐surface projected data is vital for climate change impact studies and adaptation. We derived the highest statistically downscaled multivariate ensemble currently available: daily 1 km until end of century. Deep learning models were employed to develop transfer functions precipitation, water vapor pressure, radiation, wind speed, and, maximum, mean minimum temperature. Perfect prognosis particular statistical downscaling methodology applied, using a subset ReKIS set Saxony as predictands, ERA5 reanalysis during‐training predictors CORDEX‐EUR11 predictors. The performance was validated with VALUE framework, yielding highly satisfactory results. Particular attention given three major perfect assumptions, which several tests carried out thoroughly discussed. From latter, we corroborated their fulfillment high degree, thus, projections are considered adequate relevant modelers. In total, 18 runs RCP85, RCP45, 4 RCP26 under both stochastic deterministic approaches. This could drive more accurate diverse in region. Generally, climatologies agreement coarser projections. Nevertheless, particularities observed some projections, list caveats potential users given. Due scalability presented methodology, further possible applications additional datasets proposed. Lastly, improvement prospects discussed toward ideal subsequent iteration methodology.