关键词: 甘肃省, Logistic回归模型, 自然因素, 人为因素, 火险区划

Abstract:

Objective: The objective of this study is to explore the main influencing factors of forest fires in Gansu Province and the geographical division of forest fire risk in the Province. Method: Based on the historical fire data of Gansu Province from 2000 to 2021, ArcGIS 10.8 was used to extract the spatial information of topography, meteorology, vegetation, human activities, and socio-economic factors, and the Logistic regression model was used to determine the main driving factors of forest fires in Gansu Province. The standardized regression coefficient was used to test the relative importance of each driving factor to the occurrence of forest fires, and the ROC curve was used to test the fitting effect of the model. According to the prediction results of the model, the probability of forest fire occurrence was classified into hierarchical regions. Result: 1) The spatial logistic forest fire risk model established was well fitted. The AUC value of the five training samples was all greater than 0.970, and the prediction accuracy was between 89.6%?90.6%, and the AUC value of the whole sample was 0.972, and the prediction accuracy was 89.1%, indicating that the model was suitable for the predicting of forest fire occurrence in Gansu Province. 2) Elevation, slope, mean monthly temperature, vegetation coverage, and distance to the railway were positively correlated with the probability of forest fires, while mean monthly precipitation, distance to the highway, and distance to the settlement were negatively correlated with the probability of forest fires. 3) According to the standardized regression coefficients, the importance of each driving factor to the occurrence of forest fires in Gansu Province was ranked as follows: vegetation coverage (15.31) > mean monthly temperature (1.647) > slope (1.055) > distance to the railway (1.015) > elevation (1.007) > distance to the highway (0.985) > distance to the settlement (0.852) > mean monthly precipitation (0.808). 4) According to the probability of forest fires, the forest fire risk in Gansu Province was divided into five levels at equal intervals: very low fire danger area (0?0.2), low fire danger area (0.2?0.4), medium fire danger area (0.4?0.6), high fire danger area (0.6?0.8), and very high fire danger area (0.8?1), and the corresponding forest areas were 462 640, 850 160, 1 611 551, 1 715 681 and 705 050 hm², respectively. Conclusion: Vegetation coverage, mean monthly temperature, slope, distance to the railway, elevation, distance to the highway, distance to the settlement, and mean monthly precipitation are the main driving factors of forest fire occurrence in Gansu Province. The probability of forest fire occurrence is generally high in the southeast and low in the northwest, among which the extremely high fire danger areas are mainly distributed in Gannan Tibetan Autonomous Prefecture, Linxia Hui Autonomous Prefecture, Lanzhou and some areas of Longnan, and the high fire danger areas are mainly distributed in the southeast of Longnan and Tianshui, the western part of Pingliang, Lanzhou and some areas of Linxia Hui Autonomous Prefecture.

Key words: Gansu Province, logistic regression model, natural factors, human factors, fire hazard zoning