Tropical forests have increasing sustainability challenges with regard to forest fires where vegetation nature and human influences interact in the determination of fire risk. This study uses a Random Forest classifier to evaluate the vulnerability of forest fires in tropical moist deciduous (Kanger Ghati National Park) and tropical dry deciduous (Udanti Wildlife Sanctuary) forests of Chhattisgarh, India with reliance on various sources of geospatial data. Pre and post-monsoon susceptibility maps indicate that seasonality is apparent: pre-monsoon months (March–May) are more vulnerable with high surface temperature, lower vegetation moisture, and open canopy cover, whereas the susceptibility in post-monsoon months (October–December) is lower as a result of the replenishment of the moisture by rainfall. The tropical dry deciduous forest was defined as more vulnerable to fire and 17% of its area was very highly susceptible in the pre- monsoon period as compared to 3–4% of the moist deciduous forest. No-fire conditions were well predicted but actual fire occurrences were not as well predicted due to the imbalance in the datasets, which suggests that higher-resolution predictors or real-time predictors are needed. The findings underscore the importance of season- and forest type-specific fire management strategies and demonstrate the utility of machine learning based susceptibility mapping for sustainable forest management and disaster risk reduction in fire-prone landscapes. The work highlights how susceptibility not only varies temporally but also in accordance with forest type.
Gurbani et al. (Sun,) studied this question.