1. Deep learning modeling of human activity affected wildfire risk by incorporating structural features: A case study in eastern China.
- Author
-
He, Zhonghua, Fan, Gaofeng, Li, Zhengquan, Li, Shaohong, Gao, Ling, Li, Xiang, and Zeng, Zhao-Cheng
- Subjects
- *
WILDFIRE risk , *DEEP learning , *TRANSFORMER models , *LEARNING , *HAZARD mitigation , *WILDFIRE prevention , *SPATIAL orientation - Abstract
• A deep-learning-based data-driven approach is developed for human activity affected wildfire risk prediction; • A combination of state-of-the-art deep-learning techniques, including CNN, LSTM, channel-spatial attention, Vision transformer and ANN; • The approach adopts internal structural features of samples for capturing details like mountain trails in hilly terrains. • Our predicted wildfire risk shows significant accuracy enhancement compared to traditional FWI and ConvLSTM predictions. Wildfire risk prediction is a critical component of disaster prevention and mitigation, often closely associated with local human activities in most regions. Recent studies demonstrate that employing joint modeling techniques using diverse datasets alongside Convolutional Neural Networks-Long Short-Term Memory Networks (ConvLSTM) produces favorable predictive results. However, previous research inadequately explored the different impact of factors across different categories and spatial orientations, and neglected the impact of fuels and human activities inside the samples. This study focuses on the six eastern provinces of China, utilizing a multi-source dataset comprising satellite-monitored wildfire products from 2012 to 2022, along with various factors indicating terrestrial and human activities, simulated meteorological elements and high-resolution vegetation imagery. By introducing channel and spatial attention mechanisms and visual transformer mode, this research optimizes the ConvLSTM wildfire prediction model. Results indicate a noteworthy enhancement, elevating accuracy, Kappa coefficient, and AUC of ROC curves from 91.15%, 80.87%, and 97.01% to 92.79%, 84.48%, and 97.90%, respectively. Consequently, it reinforces the accuracy by increase of the structural features within samples and quantifying the differences in the importance of different factors, which is also validated by prediction application of the samples in the entire year of 2023. Sensitivity analysis reveals that the current model is still highly dependent on the meteorological factors. Notably, the impact of structural features significantly surpasses the influence of terrain and terrestrial ecology elements, which should be considered in further models. Thus, this study has developed a methodology integrating multiple attention mechanisms and sample structural features, which could furnish high-precision daily kilometer-level wildfire risk prediction products. This method could improve the efficiency of prevention and control by improving the accuracy and narrowing the high-risk areas. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF