Climate change is fundamentally transforming fire regimes in boreal forest ecosystems. This study aims to evaluate the spatial coherence of fire risk maps generated by Bayesian network (BN) models under SSP2-4.5 (intermediate scenario) and SSP5-8.5 (high-emission scenario) climate projections for the 2050 and 2080 projection periods. The research integrates remote sensing data, topographic variables, climate projection data, and historical fire records from the northwestern Canadian boreal zone to construct a multivariate Bayesian network model. Spatial coherence is measured using Cohen's Kappa statistic, Moran's I spatial autocorrelation index, structural similarity index (SSIM), and area-based overlap ratio. Results indicate that SSP2-4.5 exhibits high spatial coherence between 2050 and 2080 (Kappa = 0.74 ± 0.03), while SSP5-8.5 shows markedly lower coherence (Kappa = 0.58 ± 0.05). A northward shift of fire risk classes is observed, particularly pronounced under SSP5-8.5. The Bayesian network's capacity to explicitly represent uncertainty through conditional probability tables produces more reliable spatial projections compared to logistic regression and random forest approaches. This study provides evidence regarding the spatial reliability of scenario-based fire risk maps for climate adaptation planning and develops actionable recommendations for forest management policies.
Kaan Alper (Sat,) studied this question.