Regional climate projections using a deep-learning-based model-ranking and downscaling framework: Application to European climate zones

TL;DR

The paper presents a two-stage framework for regional climate projections over Europe by first ranking 32 CMIP6 GCMs with a DL-TOPSIS method that learns dynamic weights for multiple performance metrics, including extremes such as TXx and TNn. Top-ranked GCMs are then downscaled to 0.1° resolution using four deep-learning architectures—CNN-LSTM, ConvLSTM, ViT, and GeoSTANet—each capturing spatiotemporal dependencies, with GeoSTANet delivering the best performance in extreme temperature representation (RMSE ≈ $1.57\,^{\circ}\mathrm{C}$, KGE ≈ $0.89$, NSE ≈ $0.85$, $r\approx0.92$). Across Köppen-Geiger zones and seasons, no single GCM dominates; instead, the framework identifies a subset of models that excel in different contexts, and pairing these with advanced downscaling reduces biases and improves distributional fidelity (PDF Overlap). The results demonstrate that dynamic, data-driven model weighting combined with transformer-based downscaling offers a scalable, high-accuracy pathway to produce regional climate projections suitable for impact assessments and adaptation planning. This approach can be extended to additional variables and regions, with future work including multi-variable downscaling and uncertainty quantification to further enhance decision-relevant climate services.

Abstract

Accurate regional climate forecast calls for high-resolution downscaling of Global Climate Models (GCMs). This work presents a deep-learning-based multi-model evaluation and downscaling framework ranking 32 Coupled Model Intercomparison Project Phase 6 (CMIP6) models using a Deep Learning-TOPSIS (DL-TOPSIS) mechanism and so refines outputs using advanced deep-learning models. Using nine performance criteria, five Köppen-Geiger climate zones -- Tropical, Arid, Temperate, Continental, and Polar -- are investigated over four seasons. While TaiESM1 and CMCC-CM2-SR5 show notable biases, ranking results show that NorESM2-LM, GISS-E2-1-G, and HadGEM3-GC31-LL outperform other models. Four models contribute to downscaling the top-ranked GCMs to 0.1$^{\circ}$ resolution: Vision Transformer (ViT), Geospatial Spatiotemporal Transformer with Attention and Imbalance-Aware Network (GeoSTANet), CNN-LSTM, and CNN-Long Short-Term Memory (ConvLSTM). Effectively capturing temperature extremes (TXx, TNn), GeoSTANet achieves the highest accuracy (Root Mean Square Error (RMSE) = 1.57$^{\circ}$C, Kling-Gupta Efficiency (KGE) = 0.89, Nash-Sutcliffe Efficiency (NSE) = 0.85, Correlation ($r$) = 0.92), so reducing RMSE by 20% over ConvLSTM. CNN-LSTM and ConvLSTM do well in Continental and Temperate zones; ViT finds fine-scale temperature fluctuations difficult. These results confirm that multi-criteria ranking improves GCM selection for regional climate studies and transformer-based downscaling exceeds conventional deep-learning methods. This framework offers a scalable method to enhance high-resolution climate projections, benefiting impact assessments and adaptation plans.

Figures (6)

• Figure 1: Spatial representation of Köppen-Geiger climate zones classification across Europe.
• Figure 2: Methodology workflow for model ranking and downscaling.
• Figure 3: CMIP6 model ranking heatmap across European climate zones and seasons.
• Figure 4: Overall Performance of CMIP6 Models in Europe.
• Figure 5: Spatial distribution of top-ranked CMIP6 models in Europe.