<p>The Himalayan region, known for its rich biodiversity and vast carbon sequestration potential, faces increasing threats from forest fires. This study employed four machine learning algorithms i.e., Random Forest (RF), Support Vector Machine (SVM), Boosted Regression Trees (BRT), and Generalized Linear Model (GLM) to predict forest fire susceptibility across five countries (India, Bhutan, Nepal, China (Tibet) and Pakistan in the Himalayas. Twelve key ignition variables, ranging from elevation to wind speed were utilized to develop robust prediction models. The evaluation metrics, including area under the curve (AUC), coefficient of determination (R²), Total Sum of Squares (TSS), Deviance, and SHapley Additive exPlanations (SHAP) values, were used to access the accuracy of the results. The result revealed that RF outperformed other models in distinguishing susceptible areas. The present study also estimated the vulnerable carbon pools (CP) in the region, uncovering significant variability across the study area. The study revealed that India, Bhutan, and Nepal serve as significant carbon reservoirs, underscoring their pivotal role as carbon sinks. Hence, it is imperative to formulate strategies for mitigating or averting forest fires in this region. The findings underscore the urgent need for targeted conservation and management strategies to safeguard this vital ecosystem from escalating forest fire threats. This research provides valuable insights for policymakers and conservationists, striving to preserve the ecological integrity of the Himalayan region.</p> Graphical Abstract <p>The graphical abstract presents an integrated framework for predicting forest fire vulnerability across the Himalayan region using geospatial analysis and machine learning techniques. The process begins with the geolocation of historical forest fires, derived from remote sensing data and validated through field surveys, which collectively form a comprehensive fire inventory. This dataset is then divided into training and testing subsets (70 − 30 split) to develop and validate the model. The next phase involves incorporating a suite of ignition parameters, including topographic, climatic, and biophysical variables, which serve as inputs for a predictive modeling framework. This model is used to generate spatial fire risk predictions, as shown in the predictive map illustrating different vulnerability zones across the Himalayas. Model performance is assessed using SHAP values (to understand feature importance) and ROC curve (to evaluate classification accuracy). Finally, the results are synthesized into spatially explicit vulnerability maps and bar charts, which highlight regional differences in susceptibility and support targeted forest management interventions. This approach not only enhances our understanding of forest fire dynamics in a climate-sensitive region but also provides a scientific basis for early warning systems and adaptive land-use planning.</p>