Predict Wildfires in Kalimantan using MODIS Products with Linear Regression, Gradient Boosting and Decision Tree algorithms
Syamsul Syahab Mangun, Kusrini Kusrini, Mulia Sulistiyono · 2023
Wildfires in Kalimantan pose a serious threat to the environment and ecological sustainability. Previous research may have limitations in addressing common problems associated with forest fire prediction such as uncertainties in weather data and other factors that may affect forest fires, so in an attempt to address these issues, this research focuses on using MODIS (Moderate Resolution Imaging Spectroradiometer) products and applying linear regression, Gradient Boosting Regressor (GBR), and Decision Tree Regressor (DTR) algorithms to predict forest fires in Kalimantan. Data collected by MODIS, such as surface temperature, thermal anomalies, and other environmental variables, are used as inputs for the prediction model. Linear regression algorithms were used to analyze the relationship between these variables and develop a forest fire prediction model. The results showed that both models, GBR obtained a Root Mean Square Error (RMSE) value of 9.76 with an accuracy value of 91.36% while DTR obtained an RMSE value of 9.85 with an accuracy value of 91.21%, which both models have high accuracy with a good level of variability explanation, indicated by R2 score close to 1.0. and therefore the GBR model may be preferred as a forest fire prediction model in Kalimantan based on the evaluation results obtained.