Forest fire prediction models for Guangdong Province based on GIS and multiple machine learning algorithms
Forest fires represent a grave natural disaster,causing extensive damage to forest resources and significant socio-economic losses.Since Guangdong Province is one of the high-risk areas for forest fires in China,there is a need to develop crucial tools for accurate predictions to effectively prevent and mitigate forest fires.Most forest fire prediction studies have traditionally focused on a limited number of factors,often neglecting the complexity of fire oc-currences and suffering from lower prediction accuracy.Hence,this study proposed a high-accuracy forest fire predic-tion method that combines GIS(geographic information system)and machine learning.Five machine learning algo-rithms,namely XGBoost(eXtreme gradient boosting,XGB),Light GBM(light gradient boosting machine),Cat-Boost(categorical boosting),deep neural networks(DNN),and random forests(RF),were utilized as predictive models.The inclusion of meteorological,topographical,vegetation,infrastructure,social,and human data encom-passed 24 feature factors as inputs for the models.By considering a broader range of factors related to forest fire occur-rences,this approach aimed at constructing a more comprehensive daily scale prediction model.Furthermore,the Bayesian automatic hyperparameter optimization method within the Optuna framework was introduced.This automated optimization feature adapted to different data structures,enhancing the accuracy of forest fire prediction models by au-tomatically optimizing parameter combinations.The results demonstrated that the XGBoost model emerged as the top performer with an accuracy rate of 91.30％.The forest fire prediction model was validated using data from 2018,achieving a validation accuracy of 87.81％.By utilizing GIS,a forest fire risk map of Guangdong Province was craf-ted.Furthermore,the developed model exhibited a conspicuously superior accuracy compared to other extant predictive models for forest fires.This study holds the potential to furnish valuable scientific insights for wildfire pre-vention in the forestry sector of Guangdong Province,China.
forest firemachine learningGISXGBoostOptuna hyperparameter optimizationGuangdong Province
万方数据
维普
