Abstract ID: 3.7713 | Pending | Talk/Oral | TBA | TBA

Forest fires in mountainous regions, particularly in the Indian Himalayan Region, are becoming an increasingly inevitable and destructive phenomenon. Over recent years, the frequency, size, and intensity of these fires have grown significantly, exacerbating the challenges faced in fire management. Despite the increased financial investments and infrastructure support, existing fire management strategies at both state and central levels have remained largely ineffective. This study emphasizes the need for restructuring policies, such as the National Action Plan for Forest Fires and the Forest Fire Prevention and Management Scheme, to develop more science-based and practical approaches. The analysis of fire incidents, using near-real-time data from the Moderate Resolution Imaging Spectroradiometer (MODIS) and fire spot information from NASA’s Fire Information Management System, will inform the policy review and highlight the necessity for a comprehensive, forest-type inclusive fire management framework. Building upon this context, the study further investigates the spatiotemporal patterns and predictive modeling of forest fires in Uttarakhand using a dataset from the Global Fire Atlas (2003–2016). This dataset includes key fire attributes such as fire start and end dates, duration, and burned area, alongside environmental variables derived from Google Earth Engine (GEE). These variables encompass elevation, slope, aspect, temperature trends, NDVI, precipitation, and proximity to roads. Through data preprocessing, including handling missing values and encoding categorical variables, and spatial autocorrelation analysis using Moran’s I and z-scores, the study identifies weak but statistically significant spatial clustering of fire duration and size. A range of machine learning (ML) and deep learning models, including Random Forest Regression, LightGBM, XGBoost, CatBoost, LSTM networks, and ARIMA, were used to predict fire duration. Model performance was evaluated using mean absolute error (MAE) and root mean squared error (RMSE), revealing the varying predictive capabilities of these algorithms. The results of this study provide valuable insights for optimizing fire management strategies, offering a basis for better resource allocation, risk mitigation, and environmental protection.