| 基于合成影像和多变量的博斯腾湖流域土壤有机碳含量估测 |
| 摘要点击 115 全文点击 7 投稿时间：2024-06-08 修订日期：2024-08-31 |
| 查看HTML全文
查看全文 查看/发表评论 下载PDF阅读器 |
| 中文关键词 影像合成 光谱和环境变量 Boruta算法 随机森林（RF）模型 土壤有机碳（SOC） 博斯腾湖流域 |
| 英文关键词 image synthesis spectral and environmental variables Boruta algorithm random forest （RF） model soil organic carbon（SOC） Bosten Lake Basin |
|
| 中文摘要 |
| 选择合适的多时相遥感影像合成方法以及建模变量对于土壤有机碳含量的估测及其空间分布反演具有重要作用. 以新疆博斯腾湖流域土壤有机碳含量为研究对象，按照最小值、中值以及均值Sentinel-2多时相卫星影像合成方法生成光谱变量，同时引入气候和地形等环境变量作为建模变量. 结合Boruta变量筛选算法和随机森林（RF）模型分析探究不同影像合成方法以及变量集合对耕层土壤有机碳含量估测的影响及差异. 结果表明：①环境变量结合光谱变量能够较好地估测土壤有机碳含量，环境变量中的气候变量对博斯腾湖流域土壤有机碳含量的建模估测发挥着关键作用；②相对于全变量集合，经过Boruta变量筛选算法后的特征变量模型估测精度要更好；③均值合成的影像光谱变量结合环境变量的建模效果最好，最优模型的估测精度R2为0.97，RMSE为2.919 g·kg-1，RPD为5.319. 使用Boruta变量筛选算法对多时相均值合成光谱变量与环境变量所建立的RF模型能够准确地实现博斯腾湖流域土壤有机碳含量的空间反演估测，为该流域土壤有机碳含量的准确估测提供技术支持. |
| 英文摘要 |
| The selection of appropriate multi-temporal remote sensing image synthesis methods and modeling variables plays an important role in the estimation of soil organic carbon content and its spatial distribution inversion. Taking the soil organic carbon content in Xinjiang Bosten lake basin as the research object， spectral variables were generated according to the minimum， median， and mean Sentinel-2 multi-temporal satellite image synthesis method， and environmental variables such as climate and topography were added as modeling variables. The effects of different image compositing methods and variable sets on the estimation of soil organic carbon content in tillage were analyzed using the Boruta variable screening algorithm and random forest （RF） model. The results showed that： ① The combination of environmental and spectral variables significantly improved SOC content estimation， with climate variables among the environmental factors playing a crucial role in the Bosten lake basin. ② Models using variables selected by the Boruta algorithm outperformed those using the full variable set， demonstrating that the Boruta algorithm effectively enhanced model accuracy. ③ Among all models， the one using mean-synthesized spectral variables combined with environmental variables yielded the best results， with the optimal model achieving an estimation accuracy in which R2 was 0.970， RMSE was 2.919 g·kg-1， and RPD was 5.319. The integration of multi-temporal mean-synthesized spectral variables， environmental variables， Boruta algorithm， and random forest modeling provides a reliable approach for estimating the spatial distribution of topsoil SOC content in the Bosten Lake Basin， offering a technical reference for accurately assessing soil organic carbon storage in arid regions. |
