首页|多源多特征集成的南水北调工程丹江库区湿地时空格局演变

多源多特征集成的南水北调工程丹江库区湿地时空格局演变

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
丹江口水库是中国南水北调工程的关键水源区.随着城镇化发展以及大坝二次建设,库区湿地生态系统变化显著,亟需湿地生态科学监测.以丹江库区为例,依托Google Earth Engine(GEE)平台,首先采用已有土地覆盖数据集生成湿地样本集,其次整合Landsat影像、DEM等数据构建多源特征集合,并基于Jeffries-Matusita(JM)距离进行特征优选,使用随机森林(RF)算法实现了 1985～2023年丹江库区湿地制图.结果表明:①本文提出的样本采集流程可有效提高样本质量,为长时序分类样本采集提供参考;②湿地分类特征优选后特征数由37个减为27个,分类总体精度变化不大,优选后的特征应用于丹江库区湿地分类的平均总体精度(OA)以及平均数量和分配分歧指数(QADI)分别为89.53％和0.080 2,说明特征优选有效减少信息冗余,提高影像分类效率;③1985～2023年,丹江库区湿地面积呈波动增加趋势,从1985年的17 839.85 ha扩大到2023年的28 872.48 ha,面积增长38.12％.总体来说,丹江库区湿地生态系统呈现出逐步恢复和优化的良好态势.
Evolution of Spatial and Temporal Patterns of Wetlands in Danjiang Reservoir Area of South-to-North Water Diversion Project,China with Multi-source and Multi-feature Integration
The Danjiangkou reservoir serves as a crucial water source area for China's South-to-North Water Diversion Project.With the progression of urbanization and the secondary construction of the dam,the wetland ecosystem within the reservoir area has undergone significant transformations,necessitating urgent ecological scientific monitoring.Utilizing the Danjiang reservoir area as a case study,the capabilities of the Google Earth Engine(GEE)platform was harnessed.Firstly,it adopts existing land cover datasets to generate a wetland sample set;secondly,it integrates Landsat imagery,digital elevation model(DEM),and other data sources to construct a multi-source feature collection,subsequently performing feature selection based on the Jeffries-Matusita(JM)distance;the Random Forest(RF)algorithm is then employed to achieve wetland mapping in Danjiang reservoir area from 1985 to 2023.The results show that ① the proposed sample collection process effectively enhances the quality of the samples,providing a reference for the acquisition of long-term time-series classification samples;② after wetland classification feature selection,the number of features is reduced from 37 to 27,with minimal impact on the overall classification accuracy;the average overall accuracy and the QADI value of the selected features applied to wetland classification in Danjiang reservoir area are 89.53％and 0.080 2,respectively,demonstrating the efficacy of feature selection in reducing information redundancy and improving image classification efficiency;③ from 1985 to 2023,the wetland area in Danjiang reservoir area exhibits a fluctuating increasing trend,expanding from 17 839.85 ha in 1985 to 28 872.48 ha in 2023,representing an area increase of 38.12％.In general,the wetland ecosystem in Danjiang reservoir area demonstrates a positive trend of gradual recovery and optimization.

remote sensing monitoringwetland classificationfeature selectionrandom forestLandsat imageryspatial and temporal characteristicsDanjiangkou reservoirSouth-to-North Water Diversion Project

王晓峰、马娟、周继涛、尧文洁、涂又、王筱雪

展开 >

长安大学土地工程学院,陕西西安 710054

陕西省土地整治重点实验室,陕西西安 710054

遥感监测 湿地分类 特征优选 随机森林 Landsat影像 时空特征 丹江口水库 南水北调工程

国家自然科学基金项目长安大学中央高校基本科研业务费专项资金项目

72349002CHD300102352201

2024

10.19814/j.jese.2024.02024

地球科学与环境学报
长安大学

地球科学与环境学报

CSTPCD北大核心
影响因子：1.422
ISSN：1672-6561
年,卷(期)：2024.46(5)