A Comparative Analysis of Downscaled Multi-model Decadal Climate Predictions over the Iberian Peninsula

• DOI: 10.5194/ems2024-434
• Publication Date: 2024-07-05T08:06:31Z
• Authors: Sara Moreno-Montes; Carlos Delgado-To...; Eren Duzenli; Núria Pérez-Zanón; Raül Marcos-Matam...; Albert Soret
• Keywords: Peninsula; Climate simulation

Decadal climate predictions are a source of information to anticipate the evolution system from 1 10 years ahead. Whereas both projections and decadal contain about external forcings, their main difference is that also include on phase internal variability. To achieve this, models initialised once per year with observation-based initial conditions.  This study assesses effectiveness various statistical downscaling methods applied multi-model mean near-surface air temperature precipitation for forecast 1-5 over Iberian Peninsula. The ensemble combines 13 systems contributing Climate Prediction Project (DCPP) component Coupled Model Intercomparison Phase 6 (CMIP6). performance different determined by comparing quality against raw, coarse-resolution using four deterministic or probabilistic metrics: Anomaly Correlation Coefficient (ACC), Root Mean Square error Skill Score (RMSSS), Ranked Probability (RPSS) Continuous (CRPSS). assessment carried out high-resolution ERA5Land reanalysis as reference dataset, it performed in leave-one-out cross-validation mode order emulate real-time conditions not overestimate actual skill. Three kinds have been examined. first type is  based calibrating interpolated raw (i.e. correcting biases value variance, among others). second involves building linear regressions predictors: (i) large-scale indices (e.g. Atlantic Multi-decadal Variability, AMV; North Oscillation, NAO) (ii) model data (basic regression) (iii) combination 9 nearest neighbours data. Finally, third approach search past analog days dataset. The results show skill estimates primarily depend calibration regression approaches, small differences interpolation method used during downscaling. While maintains spatial distribution compared predictions, types can change it. For temperature, high skill, which maintained after applying calibration, basic regression. However, reduced calculating predictors. precipitation, rather low, do generally increase such skill. On other hand, AMV index predictor one shows most improvement some regions.