基于随机森林算法的沾益区石漠化遥感评价

饶 昕; 李舒婷; 索 默; 傅俊涛; 岳丽兴; 陈 斌; 张祖海

文章摘要

饶昕,李舒婷,索默,傅俊涛,岳丽兴,陈斌,张祖海.基于随机森林算法的沾益区石漠化遥感评价[J].林业调查规划,2025,50(2):155-161

基于随机森林算法的沾益区石漠化遥感评价

Remote Sensing Evaluation of Rocky Desertification in Zhanyi DistrictBased on Random Forest Algorithm

DOI：

中文关键词: 石漠化等级随机森林算法遥感估测特征优选 Landsat 8 沾益区

英文关键词: rocky desertification grade random forest algorithm remote sensing estimation feature optimization Landsat 8 Zhanyi District

基金项目:云南省科技厅2023 年第二批科技成果转化专项资金(530000231100001753720).

作者	单位
饶昕	云南省林业调查规划院昆明分院,云南昆明 650299
李舒婷	云南省林业调查规划院昆明分院,云南昆明 650299
索默	云南省林业调查规划院昆明分院,云南昆明 650299
傅俊涛	云南省林业调查规划院昆明分院,云南昆明 650299
岳丽兴	云南省林业调查规划院昆明分院,云南昆明 650299
陈斌	云南省林业调查规划院昆明分院,云南昆明 650299
张祖海	云南省林业调查规划院昆明分院,云南昆明 650299

摘要点击次数: 67

全文下载次数: 71

中文摘要:

以曲靖市沾益区为研究区,以Landsat 8 为遥感影像,以236个石漠化特征点为地面样地,基于随机森林算法和Pearson 相关性分析进行特征优选,建立植被综合盖度和基岩裸露度估测随机森林回归模型,建立石漠化评价方法,以验证后的模型估测和评价研究区石漠化等级及分布情况。结果表明, 植被综合盖度模型优选特征为NIR、Albedo、NDWI、NDVI、QSAVI, 建模精度为R²= 0.789,RMSE=4.349;基岩裸露度模型优选特征为EVI、Cirrus、BSI、NDRI、RVI,建模精度为R²= 0.754,RMSE=3.462。研究区遥感估测的各石漠化等级规模和分布结果总体上与小班调查结果一致,表明此方法具有一定准确性,未来可通过优化参数、使用多源遥感等方式提高精度。

英文摘要:

This paper selected Zhanyi District of Qujing City as the research area, Landsat 8 as the remote sensing image and 236 rocky desertification feature points as the ground sample. Based on the random forest algorithm and Pearson correlation analysis, the feature selection was carried out, and a random forest regression model for the estimation of comprehensive vegetation coverage and bedrock exposure was established. The verified model was used to estimate and evaluate the grade and distribution of rocky desertification in the study area. The results showed that the optimal selection features of the comprehensive vegetation coverage model were NIR, Albedo, NDWI, NDVI and QSAVI, and the modeling accuracy was R²=0.789 and RMSE=4.349. The optimal characteristics of the bedrock exposure model were EVI, Cirrus, BSI, NDRI and RVI, and the modeling accuracy was R²=0.754 and RMSE=3.462. The scale and distribution of rocky desertification estimated by remote sensing in the study area were generally consistent with the results of sub-compartment survey, indicating that this method had a certain accuracy, and the accuracy could be improved by optimizing parameters and using multi-source remote sensing in the future.

查看全文查看/发表评论下载PDF阅读器

关闭