Predicting forest fire risk based on mining rules with ant-miner algorithm in cloud-rich areas

Abstract

Annually, millions of hectares of forest lands around the world are destroyed by fires. To minimize the fire-caused losses, more studies on the risk prediction of forest fires need to be carried out. For predicting the risk of forest fires in cloud-rich areas (e.g., the southwest of China), the synergetic use of operational forecasting systems and remote sensing-based models is expected to have a consistent performance. Therefore, we proposed in this study a new model based on ant-miner algorithm which has a good capability of solving multivariable and non-linear problems in the synergetic modeling of multi-source data. Based on historical fire data during 2000–2018 in Chongqing city, its performance was tested, and then was compared with that of other three models (i.e., meteorological data-, Artificial Neural Network-, and Support Vector Machine-based models). Results showed that, without interference from human factors, the risk predictions of proposed model were more objective. And, its mined-rules were easier to understand and also portable across multiple GIS platforms. Moreover, the proposed model has a better performance at predicting risk levels (i.e., overall accuracy was 79.02% and Kappa coefficient was 0.678) and the spatial distribution of its predictions were more detailed. This research indicated that the ant-miner algorithm-based model was more effective and reliable, and it could be used for constructing the operational system of risk predictions for forest fires in cloud-rich areas.