Benchmarking Uninitialized CMIP6 Simulations for Inter-Annual Surface Wind Predictions

This study investigates the potential of uninitialized global climate projections for providing 12-month (inter-annual) wind forecasts in Europe in light of the increasing demand for long-term climate predictions. This is important in a context where models based on the past climate may not fully account for the implications for climate variability of current warming trends, and where initialized 12-month forecasts are still not widely available (i.e., seasonal forecasts) and/or consolidated (i.e., decadal predictions). To this aim, we use two types of simulations: uninitialized climate projections from CMIP6 (Coupled Model Intercomparison Project Phase 6) and initialized 6-month seasonal forecasts (ECMWF’s SEAS5), using the latter as a benchmark. All the predictions are bias-corrected with five distinct approaches (quantile delta mapping, empirical quantile mapping, quantile delta mapping, scaling bias-adjustment and a proprietary quantile mapping) and verified against weather observations from the ECA&D E-OBS project (684 weather stations across Europe). It is observed that the quantile-mapping techniques outperform the other bias-correction algorithm in adjusting the cumulative distribution function (CDF) to the reference weather stations and, also, in reducing the mean bias error closer to zero. However, a simple bias -correction by scaling improves the time-series predictive accuracy (root mean square error, anomaly correlation coefficient and mean absolute scaled error) of CMIP6 simulations over quantile-mapping bias corrections. Thus, the results suggest that CMIP6 projections may provide a valuable preliminary framework for comprehending climate wind variations over the ensuing 12-month period. Finally, while baseline methods like climatology could still outperform the presented methods in terms of time-series accuracy (i.e., root mean square error), our approach highlights a key advantage: climatology is static, whereas CMIP6 offers a dynamic, evolving view of climatology. The combination of dynamism and bias correction makes CMIP6 projections a valuable starting point for understanding wind climate variations over the next 12 months. Furthermore, using workload schedulers within high-performance computing frameworks is essential for effectively handling these complex and ever-evolving datasets, highlighting the critical role of advanced computational methods in fully realizing the potential of CMIP6 for climate analysis.