Bridging CORDEX and CMIP6: Machine Learning Downscaling for Wind and Solar Energy Droughts in Central Europe

TL;DR

This study demonstrates that machine-learning emulators can efficiently downscale coarse CMIP projections to regional CORDEX resolution, enabling detailed assessment of wind and solar energy droughts in Central Europe. By training on CMIP5–CORDEX and applying to CMIP6, the CorrDiff-based emulator reproduces regional drought signals and reveals that higher-resolution data tend to show more frequent energy-drought events than driving GCMs. Application to CMIP6 indicates slightly smaller changes in drought days than in CORDEX, reflecting differences in global-model ensembles, while the emulator provides a scalable framework for impact analyses with large ensembles. Overall, the work highlights the practical value of ML emulators in bridging global and regional climate information for renewable-energy resilience, with robust uncertainty characterization through extensive sampling.

Abstract

Reliable regional climate information is essential for assessing the impacts of climate change and for planning in sectors such as renewable energy; yet, producing high-resolution projections through coordinated initiatives like CORDEX that run multiple physical regional climate models is both computationally demanding and difficult to organize. Machine learning emulators that learn the mapping between global and regional climate fields offer a promising way to address these limitations. Here we introduce the application of such an emulator: trained on CMIP5 and CORDEX simulations, it reproduces regional climate model data with sufficient accuracy. When applied to CMIP6 simulations not seen during training, it also produces realistic results, indicating stable performance. Using CORDEX data, CMIP5 and CMIP6 simulations, as well as regional data generated by two machine learning models, we analyze the co-occurrence of low wind speed and low solar radiation and find indications that the number of such energy drought days is likely to decrease in the future. Our results highlight that downscaling with machine learning emulators provides an efficient complement to efforts such as CORDEX, supplying the higher-resolution information required for impact assessments.

Figures (14)

• Figure 1: Number of co-occurring low wind and solar radiation days in the original RCMs from CORDEX, emulated RCMs, and GCMs from CMIP5 for the periods 2030–2039 (left) and 2090–2099 (right). No drought days are observed in spring and summer. In autumn and winter, both the emulated and original RCMs indicate a decrease in drought days, whereas the GCMs project a slight increase in autumn and a decrease in winter. Markers with black edges represent the ensemble mean, and whiskers denote the standard deviation.
• Figure 2: The colored region corresponds to the RCM region considered, and the frame of the plot marks the GCM region. The red marker indicates the RCM location (lon = $6.55^\circ$E, lat = $46.90^\circ$N) analyzed in \ref{['fig:single-location-eraland']}.
• Figure 3: Drought days at a single location. For the RCM and emulated RCM, the location is lon = $6.55^\circ$E, lat = $46.90^\circ$N. For the GCM, the closest grid point to that is lon = $6.25^\circ$E, lat = $46.25^\circ$N and for ERA5-Land lon = $6.50^\circ$E, lat = $46.90^\circ$N.
• Figure 4: Spatial distribution of co-occurring low wind speed and low solar radiation events in 2090-99 in the CNRM-CM5 global model downscaled with the regional CM5 RegCM4-6 model. The first column shows the number of drought events with location-specific threshold for GCM, RCM, and ML emulator data, illustrating the added spatial detail in RCM and emulator outputs. The second and third columns display the corresponding average solar radiation and average wind speed during drought events, respectively.
• Figure 5: Number of co-occurring low wind and solar days in the original RCMs, emulated RCMs, and GCMs for the periods 2030–2039 (a) and 2090–2099 (b) and the difference between the two time periods (c). The plot is an extension of \ref{['fig:figure1']} and additionally includes CMIP6 results. Spring and summer have been omitted as no drought days occur in any of the datasets.