Machine learning based forest fire susceptibility assessment of Manavgat district (Antalya), Turkey

This study primarily aims to produce forest fire susceptibility maps for the Manavgat district of Antalya province in Turkey using different machine learning (ML) techniques. Forest fire inventory data were obtained from the General Directorate of Forestry. The inventory data comprise a total of 545 forest fire ignition points during the years 2013-2021. For model training and validation, 70% and 30% of these points were used, respectively. Average annual temperature, average annual rainfall, aspect, distance to rivers, elevation, distance to settlements, forest type, distance to roads, land cover, plan curvature, slope, solar radiation, tree cover density, topographic wetness index, and wind effect parameters were used in the study. Multicollinearity analysis of these 15 factors showed that they are independent of each other. Tree-based ML models, namely, eXtreme gradient boosting (XGBoost), random forest, and gradient boosting machine, as well as artificial neural networks (ANN) were used to produce forest fire susceptibility maps. The metrics of overall accuracy, precision, recall, F1 score and area under the receiver operating characteristic curve (AU-ROC) were used to evaluate the performance of the ML models. Based on our results, the XGBoost model revealed the most appropriate susceptibility map that could be used for fire prevention measures.

Automated summary

This study aimed to produce forest fire susceptibility maps for the Manavgat district in Antalya province, Turkey using different machine learning techniques. Forest fire inventory data from 2013-2021 were obtained and 15 factors were used in the study. Tree-based ML models and artificial neural networks were used to produce the maps, and performance evaluation was done using various metrics. The XGBoost model was found to be the most suitable for fire prevention measures.