Modelling the climate factors affecting forest fire in Sumatra using Random Forest and Artificial Neural Network

Abstract

Carbon emissions produced by forest fires contribute to the global emission increase. The amount of carbon emission may indicate the severity of the fires. In a dry climate condition, forest fires become an unexpected serious problem. This paper investigates the effect of climate variables on forest fires in Sumatra from 1998 to 2018. We employ two methods, Random Forest (RF) and Artificial Neural Network (ANN) to predict the carbon emission in 2019-2021. The total emission over the domain and the fire distribution map are compared in both models. As a result, the RF model is more accurate in predicting the location and intensity in 2019 but overestimates in 2020-2021. This indicates that the RF model gives a slightly better prediction when the carbon emission is high. This result is consistent with the evaluation metrics showing that ANN mostly gives smaller errors. Also, we found that the climate variables are still relevant to describe the carbon emissions through both models with importance scores of more than .