生态环境学报2024，Vol.33Issue(7)：1027-1035,9.DOI:10.16258/j.cnki.1674-5906.2024.07.004
淮北矿区多种类型植被地上生物量反演研究
Research on Biomass Inversion of Multiple Vegetation Types on the Surface of Mining Areas
摘要
Abstract
Biomass is an important component of vegetation carbon,and is crucial for evaluating the effectiveness of ecological restoration in mining areas.In the Zhahe mining area of Huaibei Mining,ecological destruction and restoration coexist with a rich variety of surface vegetation types including trees,herbs,and crops.Traditional single-species biomass models are difficult to apply in areas where multiple vegetation types overlap.To accurately invert the biomass of the composite vegetation in this region,this study used Sentinel satellite data and constructed a spectral index feature dataset using vegetation indices and band operation methods.Different features and model combinations were used to construct biomass inversion models for each region.In the feature selection stage,two methods,random forest and correlation analysis,were used to evaluate the features comprehensively.Ultimately,five key spectral indices(enhanced vegetation index,normalized difference water index,improved normalized difference water index,Sentinel-2 red edge position,and land chlorophyll index)and five self-constructed spectral features based on band operations(1/B1,1/B2,1/B7,B1/B2,B5/B6)were selected.Based on these selected features,five different feature combination schemes were designed,and biomass inversion models were constructed separately for forest and composite vegetation data by combining the traditional equation regression and machine learning models.The results showed that compared with traditional regression models,machine learning models have higher accuracy in biomass inversion.Self-constructed spectral features based on band operations can improve the accuracy of biomass inversion using machine learning models.Among them,the support vector regression(SVR)model combined with self-constructed spectral features and original bands achieved the highest accuracy in inversion results,with a determination coefficient R2 of 0.74 and a root mean square error of 8.14 kg·m-2 for the model validation set.Applying the SVR model to biomass inversion in the study area yielded results highly consistent with local vegetation distribution characteristics.The results of this study not only provide data support for the evaluation of ecological restoration in mining areas but also provide reference and experience for subsequent studies on similar ecosystems.关键词
矿山/多植被交叉覆盖区/综合生物量/机器学习/反演模型/植被类型Key words
mining area/multi-vegetation overlap area/composite biomass/machine learning/inversion model/vegetation type分类
天文与地球科学引用本文复制引用
杨可明,彭里顺,张燕海,谷新茹,陈新阳,江克贵..淮北矿区多种类型植被地上生物量反演研究[J].生态环境学报,2024,33(7):1027-1035,9.基金项目
国家科技基础资源调查专项(2022FY101905) （2022FY101905）
企、事业单位委托项目(2023-129) （2023-129）
国家自然科学基金项目(41971401) （41971401）
