Uthappa A R, Das Bappa, Raizada Anurag, Kumar Parveen, Jha Prakash, Prasad P V Vara
ICAR-Central Coastal Agricultural Research Institute, Ela, Old Goa, India.
ICAR-Central Coastal Agricultural Research Institute, Ela, Old Goa, India.
J Environ Manage. 2025 Apr;379:124777. doi: 10.1016/j.jenvman.2025.124777. Epub 2025 Mar 9.
Forest fires have significantly increased over the last decade due to shifts in rainfall patterns, warmer summers, and long spells of dry weather events in the coastal regions. Assessment of susceptibility to forest fires has become an important management tool for damage control before the occurrence of fires, which often spread very rapidly. In this context, the current study was undertaken with the aim to map forest areas susceptible to fire in the state of Goa (India) using remote sensing (RS) and geographic information system () derived variables through an analytical hierarchy process (AHP) and machine learning techniques namely random forest (RF), support vector machine (SVM), extreme gradient boosting (XGB). Nine variables viz. Elevation (m), slope (%), aspect, topographic wetness index (TWI), forest cover types, average normalized difference vegetation index (NDVI), distance to road (m), distance to settlement (m), and land surface temperature (LST, °C) were used to map susceptible areas in five different classes. The map classified forest areas into different susceptibility levels, with significant variations observed across different models. The study emphasized the importance of machine learning techniques for forest management and fire risk assessment. Validation of the susceptibility map showed excellent performance of the models, with the random forest model exhibiting the best performance. The forest fire susceptibility map generated using RF indicated that a large area (44.15%) of forest cover in Goa is very highly susceptible to fire followed by highly susceptible (21.35%) and a moderately susceptible area of 15.62%. SHapley Additive exPlanations (SHAP) analysis using RF identified forest type, distance from settlement, slope and NDVI as important variables affecting forest fire susceptibility. In the study area, an extended dry period with no post-monsoon rainfall makes the forest highly susceptible to fire. In view of the large area potentially susceptible to forest fire, there is an urgent need to implement preventive measures for fire control in the identified zones.
在过去十年中，由于降雨模式的变化、夏季气温升高以及沿海地区长期干旱天气事件，森林火灾显著增加。评估森林火灾易发性已成为火灾发生前进行损害控制的重要管理工具，因为火灾往往蔓延迅速。在此背景下，本研究旨在利用遥感（RS）和地理信息系统（GIS）派生变量，通过层次分析法（AHP）和机器学习技术（即随机森林（RF）、支持向量机（SVM）、极端梯度提升（XGB））绘制印度果阿邦易发生火灾的森林区域地图。使用了九个变量，即海拔（米）、坡度（%）、坡向、地形湿度指数（TWI）、森林覆盖类型、平均归一化植被指数（NDVI）、距道路距离（米）、距居民点距离（米）和地表温度（LST，°C），将易发生火灾区域划分为五个不同等级。该地图将森林区域划分为不同的易发性等级，不同模型之间存在显著差异。该研究强调了机器学习技术在森林管理和火灾风险评估中的重要性。对易发性地图的验证表明模型表现出色，随机森林模型表现最佳。使用随机森林生成的森林火灾易发性地图显示，果阿邦很大一部分森林覆盖面积（44.15%）极易发生火灾，其次是高度易发生火灾的区域（21.35%）和中度易发生火灾的区域（15.62%）。使用随机森林的SHapley加性解释（SHAP）分析确定森林类型、距居民点距离、坡度和归一化植被指数是影响森林火灾易发性的重要变量。在研究区域，季风后无降雨的延长干旱期使森林极易发生火灾。鉴于存在大面积潜在易发生森林火灾的区域，迫切需要在已确定区域实施火灾控制预防措施。
