Machine Learning Based on NDVI and Meteorological Drought Index in Forest Fire Early Warning

Forest fires not only contribute to global warming but also pose a serious threat to human life and property safety. Based on the combination of deep learning and remote sensing technology, efficient monitoring and early warning of forest fires can be realized, to effectively reduce the losses caused by fires. Based on these factors this paper proposes a deep learning method for early warning of fire using NDVI and meteorological drought data. After normalization of the raw data, static and temporal features are constructed and used to generate fire risk target variables. Subsequently, machine learning and deep learning are performed based on the static and time-series features, respectively. In this paper, it is found that (1) the application of three machine learning models, i.e., GirdSearch XGBoost (GX), Fixed XGBoost (FX), Voting Ensemble (Ens) model, and Long Short-Term Memory (LSTM) model, performs well in prediction of both static and time-series features. 2) A path to generate fire risk prediction maps based on fusion algorithms is explored. The overall distribution of fire risk, the distribution of high-risk areas, and geographically linked features were derived. However, this study still has deficiencies in the efficiency of program operation and the breadth of data sources. In the future, the data structure will be optimized to guarantee the efficient running of the program; a wider range of data will be used to improve the accuracy and robustness of the model.