Scholarlink

Background <p>Under future climate change scenarios, Jiangxi Province was predicted to be wetter and hotter by the 2060s, which may cause significant changes in forest fire activities. However, no comprehensive analysis has been conducted to predict forest fire probability in this region by integrating multiple ecological, social, and economic factors and future climate scenarios. A prediction and analysis of the spatiotemporal patterns of forest fire probability would help make policies to prevent or mitigate the forest fire occurrence in this region.</p> Objective <p>(1) Project the historical spatiotemporal forest fire probability in Jiangxi Province; (2) identify the key driving factors for fire occurrence; and (3) predict and analyze future fire probability under climate change scenarios.</p> Methods <p>Based on MODIS fire point data and multi-source geospatial datasets, including NDVI, land cover type, climate, topography, and socioeconomic factors during 2001–2020, we constructed a random forest (RF) machine-learning model. A total of 54,372 sampling points, including 21,749 fire points and 32,623 non-fire points, were split into a 60% training set and a 40% test set to train and test the RF model, respectively. Model performance was evaluated using the metrics including area under curve (AUC), accuracy (ACC), and threat score (TS). The trained and validated RF model was then applied to predict forest fire probability under different future climate change scenarios.</p> Results <p>(1) The RF model exhibited excellent predictive ability in predicting fire occurrence probability, with AUC = 0.939, ACC = 0.868, and mean TS = 0.741; (2) the top six key driving forces for fire occurrence were identified, which include precipitation, current-month NDVI, current-month temperature, previous-month temperature, previous-month NDVI, and previous-month precipitation; (3) during 2001–2020, high-risk forest area in Jiangxi Province increased from 3.69% to 4.7% of total forest area, moderate-risk area decreased from 39.12% to 36.48%, and low-risk area increased by 1.42%; (4) under the SSP2-4.5 climate scenario, high and moderate risk will increase gradually from 2031 to 2060, reaching 9.97% in 2051–2060, mainly in central-southern Jiangxi; (5) under SSP5-8.5, high-risk area will decrease from 5.7% (2031–2040) to 3.12% (2051–2060) due to increased precipitation.</p> Conclusions <p>Our study identified the dominant role of “precipitation-NDVI” in regulating fire occurrence in subtropical China. The forest fire risk area, especially the high-risk area, will increase under the medium emission scenario (SSP2-4.5), implying that the current climate change trend will cause more severe fire disasters in this region and thus threaten ecological and socioeconomic systems in Jiangxi Province. Therefore, this region will significantly benefit from the realization of China’s carbon neutrality by 2060.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Projecting spatiotemporal forest fire probability in Jiangxi Province under climate change scenarios

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月20日

查看全文
  • Qing Ye,
  • Guofan Lu,
  • Wangyu Peng,
  • Jiahui Peng,
  • Guoai Xie

摘要

Background

Under future climate change scenarios, Jiangxi Province was predicted to be wetter and hotter by the 2060s, which may cause significant changes in forest fire activities. However, no comprehensive analysis has been conducted to predict forest fire probability in this region by integrating multiple ecological, social, and economic factors and future climate scenarios. A prediction and analysis of the spatiotemporal patterns of forest fire probability would help make policies to prevent or mitigate the forest fire occurrence in this region.

Objective

(1) Project the historical spatiotemporal forest fire probability in Jiangxi Province; (2) identify the key driving factors for fire occurrence; and (3) predict and analyze future fire probability under climate change scenarios.

Methods

Based on MODIS fire point data and multi-source geospatial datasets, including NDVI, land cover type, climate, topography, and socioeconomic factors during 2001–2020, we constructed a random forest (RF) machine-learning model. A total of 54,372 sampling points, including 21,749 fire points and 32,623 non-fire points, were split into a 60% training set and a 40% test set to train and test the RF model, respectively. Model performance was evaluated using the metrics including area under curve (AUC), accuracy (ACC), and threat score (TS). The trained and validated RF model was then applied to predict forest fire probability under different future climate change scenarios.

Results

(1) The RF model exhibited excellent predictive ability in predicting fire occurrence probability, with AUC = 0.939, ACC = 0.868, and mean TS = 0.741; (2) the top six key driving forces for fire occurrence were identified, which include precipitation, current-month NDVI, current-month temperature, previous-month temperature, previous-month NDVI, and previous-month precipitation; (3) during 2001–2020, high-risk forest area in Jiangxi Province increased from 3.69% to 4.7% of total forest area, moderate-risk area decreased from 39.12% to 36.48%, and low-risk area increased by 1.42%; (4) under the SSP2-4.5 climate scenario, high and moderate risk will increase gradually from 2031 to 2060, reaching 9.97% in 2051–2060, mainly in central-southern Jiangxi; (5) under SSP5-8.5, high-risk area will decrease from 5.7% (2031–2040) to 3.12% (2051–2060) due to increased precipitation.

Conclusions

Our study identified the dominant role of “precipitation-NDVI” in regulating fire occurrence in subtropical China. The forest fire risk area, especially the high-risk area, will increase under the medium emission scenario (SSP2-4.5), implying that the current climate change trend will cause more severe fire disasters in this region and thus threaten ecological and socioeconomic systems in Jiangxi Province. Therefore, this region will significantly benefit from the realization of China’s carbon neutrality by 2060.