基于CatBoost的福州市山火风险评估研究  

Assessment of forest fire risk in Fuzhou City based on CatBoost Algorithm

作　　者：于欣 江洪[1] 林静[1] 徐加其 Yu Xin;Jiang Hong;Lin Jing;Xu Jiaqi(Key Lab of Spatial Data Mining and Information Sharing of Ministry of Education,National&Local Joint Engineering Research Center of Satellite Geospatial Information Technology,Academy of Digital China,Fuzhou University,Fuzhou 350108,China)

机构地区：[1]福州大学空间数据挖掘与信息共享教育部重点实验室,卫星空间信息技术综合应用国家地方联合工程研究中心,数字中国研究院(福建),福建福州350108

出　　处：《海南大学学报（自然科学版）》2024年第2期174-185,共12页Natural Science Journal of Hainan University

基　　金：福建省科技计划引导性项目(2021Y0005);福建省水利科技项目(MSK202301)。

摘　　要：以植被、地形、气象、人为活动4类共23个山火影响因子为基础,构建特征数据集并基于CatBoost集成学习方法构建了福州市日度山火风险评估模型.研究表明:2010—2021年福州市山火的发生具有空间聚集性,且山火发生次数存在显著下降趋势;福州市山火的发生受归一化植被指数的影响最大,其次是气象、地形及人为活动因子;集成学习方法对福州市山火预测精度普遍较高,CatBoost山火预警模型在概率预测及火点识别等方面均优于目前常用的RF和XGBoost模型,AUC为0.928,基于该模型得出福州市山火风险由其东北、西南部向中心降低,罗源县、连江县、闽清县、永泰县山火风险相对较高,福州市区山火风险相对较低.本研究可实现福州市山火风险等级评估,对福州市开展针对性山火防控管理工作具有一定参考价值.In the report,based on a total of 23 mountain fire impact factors in four categories:vegetation,topography,meteorology,and human activity,a feature dataset was constructed,and a daily mountain fire risk assessment model for Fuzhou City based on the CatBoost integrated learning method was also constructed.The mountain fires in Fuzhou are spatially clustered and there is a significant downward trend of the number of mountain fires from 2010 to 2021;The occurrence of mountain fires in Fuzhou is most influenced by the normalized vegetation index,followed by meteorological,topographical,and human activity factors;The integrated learning methods were generally more accurate in predicting mountain fires in Fuzhou City,the CatBoost mountain fire warning model outperforms the currently commonly used RF and XGBoost models in terms of probabilistic prediction and fire point identification with an AUC of 0.928.Based on the model,it was concluded that the risk of mountain fire in Fuzhou City decreases from its northeast and southwest to its central part,the mountain fire risk is relatively high in Luoyuan,Lianjiang,Minqing and Yongtai counties,and relatively low in urban areas.Our study can assess the mountain fire risk level in Fuzhou City,and which can be used as a reference value for prevention and control management of mountain fire in Fuzhou City.

关 键 词：山火 时空分布 影响因子 风险评估 集成学习 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置] S762.2[自动化与计算机技术—控制科学与工程]