Unfolding the contribution of environmental and anthropogenic variables in forest fire over western Himalayan fire regime

Bar, Somnath, Parida, Bikash Ranjan and Uma Shankar, B. (2021) Unfolding the contribution of environmental and anthropogenic variables in forest fire over western Himalayan fire regime. In 2021 IEEE India Geoscience and Remote Sensing Symposium, InGARSS 2021 - Proceedings. IEEE. pp. 557-560 . (doi:10.1109/InGARSS51564.2021.9792002).

Abstract

In last few decades, a surge of uncontrolled wild and forest fire has been observed over biomes, mostly from tropical and subtropical regions. The present study has disentangled the contribution of different environmental and anthropogenic factors in forest fire over the western Himalayan (Uttarakhand and Himachal Pradesh) fire regime, which is an active fire hotspot in India. Fire-CCI v5.1 data was used to labelled fire and non-fire pixel. The climatic (e.g. maximum and minimum temperature, precipitation, solar radiation, vapour pressure, wind speed, water vapour deficit, soil moisture and palmer drought index), physiographic (elevation, slope, aspect and roughness), anthropogenic (population density and human modification) and locational (latitude and longitude) variables were utilized to unfold their contribution in forest fire by the aid of Random Forest (RF) a machine learning technique. After parameterization, a 10-fold cross-validation RF model was built over the whole dataset and the average overall accuracy, precision, recall, F-1 score and overall accuracy were estimated as 0.94 (±0.002), 0.86 (±0.003), 0.91 (±0.002) and 0.91 (±0.002), respectively. Furthermore, the whole dataset (2005-2018) was divided into two parts, training set (2005-2017) and testing (2018), to get a robust model. The testing accuracy (overall accuracy = 0.82, precision =0.79, recall = 0.95, F1 score = 0.86 and area under curve (AUC) = 0.95) suggested a reliable performance of RF model in forest fire classification (fire and non-fire). The contributions of the selected variables were retrieved from the feature importance of the RF model. The maximum temperature exhibited the highest importance, followed by elevation, minimum temperature and location variable (latitude and longitude). The population density and human modification (gHM) are moderately contributing to western Himalayan forest fire. Keywords: Forest fire; Western Himalaya; Random Forest

This record has no associated files available for download.

Contributors

Author: Somnath Bar

Download statistics