Selection of Global Climate Models (GCMs) for Climate Change Analysis Using t-Distributed Stochastic Neighbour Embedding (t-SNE): Implications of Future Bioclimatic Shifts on Forest Trees in Tunisia and Algeria

Global climate models (GCMs) are crucial for analysing climate change. With more than 50 models available from the Coupled Model Intercomparison Project Phase 6 (CMIP6), selecting the most appropriate models for accurate assessments of climate impacts is a major challenge. The variability of climate projections among these models further complicates this process. This study addresses this issue by proposing a framework for selecting representative GCMs using t-distributed stochastic neighbour embedding (t-SNE) in Tunisia and Algeria. To implement this framework, spatial grid data representing baseline conditions (1970–2000) and future climate scenarios of 20 GCMs under SSP2-4.5 were compiled from the WorldClim database. Six bioclimatic variables were selected for the analysis. For the temperature-related variables, we calculated the differences between the future projections (2050 and 2070) and the baseline climate, as well as the relative changes for the precipitation-related variables. We then performed t-SNE to reduce the dimensionality of the resulting data, testing perplexity values from 3 to 6 to determine the optimal setting. The k-means clustering algorithm was then applied to the reduced data to group GCMs with similar climate projections into clusters, and the quality of clustering was assessed using the silhouette score. From each cluster, we identified the GCM that was closest to the centroid and served as a representative model. Using these selected GCMs, we calculated the Emberger and modified thermicity indices and assessed the bioclimatic shifts in the distribution of forest trees. As a result, we identified three clusters of GCMs, each representing different patterns of climate projections. In particular, the model UKESM1-0-LL, representing cluster 3, systematically projected a more pronounced shift towards drier conditions in the future for most forest species compared to the other models. This study highlights the importance of selecting representative GCMs to capture the variability in climate projections, which is essential for reliable assessments of climate change impacts on forest ecosystems.