|
| 融合多源算法集成投票模型的森林火灾风险评估研究 |
| Forest Fire Risk Assessment Based on an Integrated Voting Model with Multi-Source Algorithms |
| 投稿时间：2025-08-10 修订日期：2026-03-16 |
| 查看/发表评论 下载PDF阅读器 |
| DOI： |
| 基金项目: |
|
| 中文关键词: 森林火灾 风险评估 层次分析法(AHP) Sentinel-2 随机森林 |
| 英文关键词:Forest fire Risk assessment Analytic Hierarchy Process (AHP) Sentinel-2 Random Forest |
| 摘要点击次数: 428 |
| 全文下载次数: 0 |
| 中文摘要: |
| 随着全球气候变化和人类活动的加剧,森林火灾的发生频率和强度不断增加,给生态环境和社会经济带来了严重影响。本研究基于多时相Sentinel-2卫星遥感数据、野外实测植被可燃物含水率及辅助多源地理数据,构建了融合随机森林特征选择与多类型机器学习算法(随机森林、线性回归、XGBoost)的集成投票模型,实现对植被可燃物含水率的高精度反演。在此基础上,结合NASA-FIRMS平台提供的火点置信度数据与多源环境因子,构建多元线性回归模型,定量分析地表温度(LST)、植被含水量(FMC)、地形及人类活动等因素对火灾发生的影响机制。表明居民区距离(权重22.95%)、FMC(权重18.88%)、坡度(18.85.%)及距道路距离(18.41%)为影响火灾风险的主要因子。进一步采用AHP层次分析法与火点置信度模型对研究区进行火灾风险等级划分,将区域划分为低等、中低等、中高等和高等四类风险区。实验结果表明,基于综合投票模型的FMC在测试集上取得了最优性能(RMSE=0.4401,误差率7.35%),优于单一模型,有效提升了预测稳定性和泛化能力。此外,基于火点置信度的森林火险评估模型预测中高等风险区面积占比68.85%,集中了79.1%的火点；高等风险区占13%,火点占比为12.43%。通过使用从NASA-FIRMS网站获取近两年的177个历史活跃火点的数据进行验证,利用ROC曲线下面积(AUC)来估计森林火灾风险图的表现。结果表明,AUC值为0.747,说明所建立的森林火灾风险评估系统对梅州市具有良好的适用性,可为其他地区的森林火灾风险评估指数体系的建立提供科学参考。 |
| 英文摘要: |
| With the intensification of global climate change and human activities, the frequency and intensity of forest fires have been increasing, causing severe impacts on ecological environments and socio-economic systems. This study utilizes multi-temporal Sentinel-2 remote sensing data, field-measured vegetation fuel moisture content (FMC), and auxiliary multi-source geospatial data to develop an integrated voting model that combines random forest-based feature selection with multiple machine learning algorithms (Random Forest, Linear Regression, XGBoost) for high-precision FMC retrieval. Furthermore, by incorporating fire point confidence data from the NASA-FIRMS platform and multiple environmental factors, a multiple linear regression model is established to quantitatively analyze the influence mechanisms of land surface temperature (LST), FMC, topography, and human activities on fire occurrence. Results indicate that distance to residential areas (weight: 22.95%), FMC (18.88%), slope (18.85%), and distance to roads (18.41%) are the primary factors influencing fire risk. Subsequently, the Analytic Hierarchy Process (AHP) and fire point confidence-based model are employed to classify fire risk levels in the study area into four categories: low, moderate-low, moderate-high, and high risk. Experimental results demonstrate that the integrated voting model achieves optimal performance in FMC prediction (RMSE = 0.4401, error rate = 7.35%) on the test set, outperforming individual models and significantly improving prediction stability and generalization capability. Moreover, the fire risk assessment model based on fire point confidence predicts that moderate-high and high-risk zones account for 68.85% of the total area and concentrate 79.1% of the fire points, with high-risk zones covering 13% of the area and containing 12.43% of the fire points. Validation is conducted using 177 historical active fire points collected from the NASA-FIRMS website over the past two years. The performance of the forest fire risk map is evaluated using the Area Under the ROC Curve (AUC), yielding an AUC value of 0.747. This indicates that the proposed forest fire risk assessment system is highly applicable for Meizhou City and provides a scientific reference for establishing fire risk assessment frameworks in other regions. |
| 关闭 |
|
|
|
