草业科学2025，Vol.42Issue(8)：1901-1910,10.DOI:10.11829/j.issn.1001-0629.2024-0422
利用随机森林与遥感数据估算内蒙古东部草地土壤有机碳含量
Estimation of soil organic carbon content in grasslands of eastern Inner Mongolia using random forest and remote sensing data
摘要
Abstract
In recent years,the issue of carbon storage in grassland ecosystems has become a central focus of ecological and environmental research,as soil health continues to decline in many areas.Soil organic carbon(SOC),a key indicator of soil function,plays a vital role in supporting plant growth,improving soil quality,regulating water,ensuring energy security,and conserving biodiversity.Currently,research on SOC content largely relies on remote sensing satellite imagery or hyperspectral data to derive relevant indices and estimate SOC using ecosystem models.This study focuses on the grasslands of eastern Inner Mongolia,integrating satellite imagery and hyperspectral data,and using multiple vegetation and soil indices as feature variables to develop a machine learning model for estimating SOC content and its spatial distribution in the study area.The findings reveal that the vegetation and soil indices derived from Landsat 8 strongly correlate with those calculated from ASD FieldSpec 4(portable field spectroradiometer)data.Specifically,the normalized difference vegetation index(NDVI),difference vegetation index(DVI),optimized soil adjusted vegetation index(OSAVI),normalized differential vegetation index(RDVI),and enhanced vegetation index(EVI)are all significantly positively correlated(P＜0.01).Among the soil salinity indices,soil salinity index 3(SI3)shows a significant positive correlation,while soil salinity index 1(SI1)and soil salinity index 2(SI2)do not.A random forest model constructed using vegetation indices(NDVI,DVI,OSAVI,RDVI,EVI),along with SI3 and the brightness index(BI),outperforms a support vector machine model,achieving an R2 of 0.92 and a root mean square error(RMSE)of 1.27 g·kg-1.However,the model based solely on Landsat 8 indices tends to overestimate SOC content,whereas the model combining Landsat 8 and hyperspectral indices better captures the spatial distribution of SOC.The SOC content in eastern Inner Mongolia's grasslands displays spatial heterogeneity,with higher levels in the west and lower levels in the east.Grasslands show the highest SOC content,while sandy areas exhibit the lowest.This study establishes a new approach for SOC estimation and offers accurate technical support for large-scale SOC assessments.关键词
草地生态系统/土壤有机碳/卫星遥感/高光谱遥感/土壤指数/植被指数/随机森林Key words
grassland ecosystem/soil organic carbon(SOC)/satellite remote sensing/hyperspectral remote sensing/soil indices/vegetation indices/random forest引用本文复制引用
贾子玉,庄前友,红梅,张振豪,程云湘..利用随机森林与遥感数据估算内蒙古东部草地土壤有机碳含量[J].草业科学,2025,42(8):1901-1910,10.基金项目
蒙古高原生态学与资源利用教育部重点实验室开放基金课题(KF2023004) （KF2023004）
中国科学院战略性先导科技专项(A类)(XDA26000000) （A类）
内蒙古自治区自然科学基金项目(2021MS03032) （2021MS03032）
