摘要
面向重大工程生态影响评估对植被覆盖高精度与高时效的需求,现有全球、全国地表覆盖数据集在空间尺度、分类体系与时间对齐方面适配性不足。以引江济淮工程区为例,依托Google Earth Engine(GEE),构建融合光谱波段、光谱指数、GLCM纹理、地形因子、年内分位数和IQR的多源特征集,采用随机森林算法,并以特征重要性驱动的特征优选形成高精度分类流程。基于Sentinel-210 m影像与人工解译样本完成训练与独立验证。结果表明:融合多源特征并经优选的分类方法,在2024年的分类结果中展现出高精度,其总体精度(OA)达到85.39%,Kappa系数为0.81,显著优于同期10 m分辨率的Dynamic World产品(OA为75.54%、Kappa系数为0.69),尤其在刻画工程廊道、窄幅岸线等复杂场景的细节方面优势明显;特征重要性显示,年内分位数与IQR的物候特征、水体和土壤与建成区增强指数、SWIR波段与纹理指标对精度提升贡献突出;2024年工程主线已形成连续水体带,沿线护岸、堤坝硬质化形成窄幅不透水带,局部出现耕地和林地向裸地或稀疏植被的转化,值得重点关注。该方法在分类精度、时效性与工程细节识别方面具备显著优势,可为重大工程生态影响评估提供高质量底图支撑。
To address the high-precision and high-timeliness requirements of ecological impact assessment for major engineering projects on vegetation cover,existing global and national land cover datasets exhibit insufficient adaptability in terms of spatial scale,classification system,and temporal alignment.Taking the Yangtze River-Huaihe River Water Diversion Project area as an example,leveraging Google Earth Engine,a multi-source feature set integrating spectral bands,spectral indices,GLCM textures,topographic factors,intra-annual quantiles,and IQR was constructed.A random forest algorithm was employed,and a high-precision classification workflow was formed through feature importance-driven feature selection.Training and independent validation were completed based on Sentinel-210 m imagery and manually interpreted samples.The results demonstrate that the classification method integrating multi-source features and optimized selection achieved high accuracy in the 2024 classification results,with an overall accuracy(OA)of 85.39% and a Kappa coefficient of 0.81,significantly outperforming the contemporaneous 10 m resolution Dynamic World product(OA 75.54%,Kappa 0.69),particularly in capturing details of complex scenarios such as engineering corridors and narrow shoreline areas.Feature importance analysis revealed that intra-annual quantiles and IQR phenological features,water and soil-enhanced indices,SWIR bands,and texture indicators contributed prominently to accuracy improvement.In 2024,the main project line had formed a continuous water body belt,with hardened embankments and dams along the route creating narrow impervious zones,and localized conversions of farmland and woodland to bare land or sparse vegetation,which warrant special attention.In conclusion,this method exhibits significant advantages in classification accuracy,timeliness,and engineering detail recognition,providing high-quality baseline support for ecological impact assessment of major engineering projects.
作者
王亚琼
王培晓
WANG Ya-qiong;WANG Pei-xiao(Department of Resources&Environment,Anhui Vocational&Technical College of Forestry,Hefei 230000,China;State Key Laboratory of Geographic Information Science&Technology,Institute of Geographic Sciences&Natural Resources Research,Chinese Academy of Sciences,Beijing 100101,China;College of Resources&Environment,University of Chinese Academy of Sciences,Beijing 100049,China)
出处
《河南城建学院学报》
2026年第1期72-79,共8页
Journal of Henan University of Urban Construction
基金
国家自然科学基金项目(42401524)
安徽省高校自然科学研究项目(2023AH052985)
安徽省教育厅质量工程项目(2024jyxm1198)。
关键词
植被覆盖
高精度制图
多源特征
随机森林
Sentinel-2
Google
Earth
Engine
引江济淮
vegetation coverage
high-precision mapping
multi-source features
random forest
Sentinel-2 remote sensing data
Google Earth Engine
Yangtze River to Huaihe River Water Diversion Project
