Deep Learning-Driven Downscaling for Climate Risk Assessment of Projected Temperature Extremes in the Nordic Region

TL;DR

This work tackles the need for high-resolution Nordic temperature projections by developing an integrative downscaling framework that fuses Vision Transformer ($ViT$), ConvLSTM, and GeoStaNet with an entropy-weighted DL-TOPSIS ranking to select robust models. It leverages bias-corrected NorESM2-LM CMIP6 outputs across ten stations spanning four Köppen zones to deliver station-based, high-resolution projections of $tasmax$, $tasmin$, and $dtr$ under SSP2-4.5 and SSP5-8.5, with Time of Emergence (TOE) analyses identifying early adaptation windows (e.g., subarctic winter ~2032). The results show deep-learning models outperform traditional baselines in accuracy and extreme-event fidelity, revealing pronounced zonal warming and larger diurnal ranges, especially in continental and subarctic areas. The framework thus provides actionable climate-service products for Nordic adaptation planning, including early-warning signals and sector-specific risk assessments, while highlighting the importance of uncertainty quantification and ongoing data-network enhancements.

Abstract

Rapid changes and increasing climatic variability across the widely varied Koppen-Geiger regions of northern Europe generate significant needs for adaptation. Regional planning needs high-resolution projected temperatures. This work presents an integrative downscaling framework that incorporates Vision Transformer (ViT), Convolutional Long Short-Term Memory (ConvLSTM), and Geospatial Spatiotemporal Transformer with Attention and Imbalance-Aware Network (GeoStaNet) models. The framework is evaluated with a multicriteria decision system, Deep Learning-TOPSIS (DL-TOPSIS), for ten strategically chosen meteorological stations encompassing the temperate oceanic (Cfb), subpolar oceanic (Cfc), warm-summer continental (Dfb), and subarctic (Dfc) climate regions. Norwegian Earth System Model (NorESM2-LM) Coupled Model Intercomparison Project Phase 6 (CMIP6) outputs were bias-corrected during the 1951-2014 period and subsequently validated against earlier observations of day-to-day temperature metrics and diurnal range statistics. The ViT showed improved performance (Root Mean Squared Error (RMSE): 1.01 degrees C; R^2: 0.92), allowing for production of credible downscaled projections. Under the SSP5-8.5 scenario, the Dfc and Dfb climate zones are projected to warm by 4.8 degrees C and 3.9 degrees C, respectively, by 2100, with expansion in the diurnal temperature range by more than 1.5 degrees C. The Time of Emergence signal first appears in subarctic winter seasons (Dfc: approximately 2032), signifying an urgent need for adaptation measures. The presented framework offers station-based, high-resolution estimates of uncertainties and extremes, with direct uses for adaptation policy over high-latitude regions with fast environmental change.

Figures (8)

• Figure 1: Geographical distribution of the ten selected Nordic stations, colour-coded by Köppen-Geiger climate classification.
• Figure 2: Workflow of the multi‑stage DL-TOPSIS downscaling framework, from GCM and observations through model ranking to scenario projection.
• Figure 3: Observed seasonal means (top panels) and annual ranges (bottom panels) of tasmax (red), tasmin (blue), and dtr (green) across four Köppen climate zones (Cfb, Cfc, Dfb, Dfc) for 1951-2014.
• Figure 4: DL-TOPSIS composite closeness coefficients for the ten downscaling models, ranked from best (ViT) to worst (GCM Baseline). Deep-learning architectures consistently outperform machine learning and statistical methods across all aggregated criteria.
• Figure 5: Observed (green) and downscaled modelled daily tasmax, tasmin, and dtr at the Stockholm station (Cfb zone). All downscaling models are shown to illustrate relative fidelity across the validation period (2011-2014). Deep-learning models (ViT, ConvLSTM, GeoStaNet) reproduce observed variability and extremes with high temporal consistency, while statistical methods exhibit larger seasonal biases.