Decadal climate predictions are a source of information to anticipate the evolution of the climate system on timescales from 1 to 10 years ahead. Since these predictions are often produced at coarser resolutions than ideal, statistical downscaling techniques can be employed to enhance their spatial resolution. This study assesses the effectiveness of various statistical downscaling methods applied to multi-model decadal predictions of mean near-surface air temperature and precipitation over Western Europe, for forecast years 1–5. The multi-model ensemble includes predictions from 13 forecasting systems (133 members) contributing to the Decadal Climate Prediction Project (DCPP) of Coupled Model Intercomparison Project Phase 6 (CMIP6). The performance of the different downscaling methods is evaluated against the high-resolution ERA5-Land reanalysis. Methods from Model Output Statistics (MOS) and Perfect Prognosis (PP) approaches are compared, which can be divided in four categories: (i) interpolations, (ii) calibrations, (iii) linear regressions and (iv) analogs methods. The results show that the skill estimates primarily depend on the calibration or linear regression approaches, with small differences from the interpolation method used during the downscaling. For temperature, raw model predictions already exhibit high skill, which is preserved when using calibrations or linear regressions using the own variable as predictor. However, skill tends to decrease when external predictors or analog methods are used. For precipitation, raw predictions generally show low skill and most downscaling methods offer little improvement. Nevertheless, analog approaches can increase skill in specific regions, highlighting their potential under certain conditions. • Interpolation methods achieve the highest performance for predictions of temperature. • Analog methods improve the skill of raw precipitation predictions over some regions. • The performance of some methods is region-dependent, especially for precipitation.
Moreno-Montes et al. (Tue,) studied this question.