基于Sentinel数据的沅陵县针叶林可燃物载量估测研究

Estimation of Fuel Load in Coniferous Forests of Yuanling County Based on Sentinel Data

森林可燃物是引发森林火灾的重要因素之一,准确估测森林可燃物载量对于制定火灾防控策略、提高火灾预警能力以及保护生态环境具有重要意义。以湖南省沅陵县Senitnel-1A和Sentinel-2A影像为数据源,通过提取多源数据的不同类型遥感因子,结合地面调查获取的样地可燃物载量信息,采用前向特征筛选法和4种机器学习模型[多元线性回归(multiple linear regression,MLR)、k最近邻(k-nearest neighbor,kNN)、支持向量机回归(support vector regression,SVR)、随机森林(random forest,RF)]构建了针叶林可燃物载量反演模型,并对研究区内针叶林可燃物载量进行反演。结果表明:①基于Sentinel-1A数据提取的VH极化后向散射系数与针叶林可燃物载量有较高的相关性;②相比于单一数据源,联合Sentinel-1A和Sentinel-2A数据可显著提高针叶林可燃物载量估测精度,最优模型R^(2)分别提高了0.19、0.29,rRMSE分别降低4.66、6.94个百分点,RMSE分别降低了6.13、9.13 t/hm^(2),平均rRMSE分别降低了5.17、5.75个百分点,最优模型为SVR模型,其R^(2)=0.5,rRMSE=27.71,RMSE=36.47 t/hm^(2)。Sentinel-1A数据的加入有利于针叶林可燃物载量估测精度的提升。

Forest combustible material,as one of the key factors contributing to forest fires,holds significant importance in formulating fire prevention and control strategies,enhancing fire warning capabilities,and safeguarding ecological environments.This study utilizes Sentinel-1A and Sentinel-2A imagery as data sources in Yuanling County,Hunan Province.Through the extraction of various types of remote sensing factors from multiple data sources and integrating ground survey data on combustible load,the study employs forward feature selection and four machine learning models(Multiple Linear Regression,MLR;k-Nearest Neighbor,kNN;Support Vector Regression,SVR;Random Forest,RF)to construct models for estimating combustible load in coniferous forests.The results indicate that:①The VH polarization backscatter coefficient extracted based on Sentinel-1A data has a high correlation with the fuel load of coniferous forests;②Compared to single data sources,the combination of Sentinel-1 and Sentinel-2 data contributes to improved accuracy in estimating combustible load in coniferous forests.The optimal models exhibit an increase in R^(2) by 0.19 and 0.29,a decrease in r RMSE by 4.66 and 6.94 percentage points,a decrease in RMSE by 6.13 t/hm^(2) and 9.13 t/hm^(2),and an average decrease in rRMSE by 5.17 and 5.75 percentage points,respectively.The best-performing model is the SVR model,with R^(2)=0.5,rRMSE=27.71,and RMSE=36.47 t/hm^(2).The incorporation of Sentinel-1A data contributes to the enhancement of accuracy in estimating combustible load in coniferous forests.