Refined classification of wetland in Hangzhou City based on Google Earth Engine (GEE) and Sentinel-2 imagery
-
摘要: 基于Google Earth Engine(GEE)云平台,使用Sentinel-2卫星影像数据,采用随机森林算法面向杭州城市湿地类型进行遥感分类研究.研究结果表明:在进行湿地类型细致分类时,相比于使用单一特征信息分类,多特征信息的组合可以显著提高分类精度.最佳特征组合包括传统光谱特征、红边光谱特征、变换特征、纹理特征和地形特征,其总体精度为81.2%,Kappa系数为0.75;不同特征信息对于不同湿地类型的提取具有重要作用,传统光谱特征对于滩地和养殖池的识别非常有利,而红边光谱特征和变换特征分别对草本沼泽和湖泊的识别更为有效.此外,红边光谱特征和纹理特征的组合有利于运河/水渠的识别.通过特征对比实验可为湿地精细化分类的案例研究提供参考,湿地分类结果可为杭州城市湿地信息的遥感识别提供数据支撑.
-
关键词:
- Google Earth Engine(GEE) /
- Sentinel-2影像 /
- 随机森林算法 /
- 湿地精细化分类
Abstract: Hangzhou City possesses diverse and abundant wetland resources. However, with the rapid urbanization, the area of wetlands has been shrinking, and issues such as degradation of ecosystem quality and functional decline have become increasingly severe. As a result, conducting a refined classification research on land cover and wetland types in Hangzhou City is crucial for creating effective wetland protection and management policies. In this study, we employed the Google Earth Engine (GEE) cloud platform and Sentinel-2 satellite imagery data to perform a classification of wetland in Hangzhou City using the Random Forest algorithm. Our findings revealed that a combination of multiple feature information substantially enhanced classification accuracy, compared to using single-feature information when executing refined wetland classification. The optimal feature combination encompassed traditional spectral features, red-edge spectral features, transformed features, texture features, and topographic features, yielding an overall accuracy of 81.2% and a Kappa coefficient of 0.75. Different feature information significantly contributed to the extraction of various wetland types. Traditional spectral features were particularly advantageous for identifying tidal flats and aquaculture ponds, while red-edge spectral features and transformed features were more effective for recognizing herbaceous marshes and lakes, respectively. Furthermore, combining red-edge spectral features and texture features proved beneficial for identifying canals and water channels. The feature comparison experiment in this study serves as a reference for future case studies on refined wetland classification, and the wetland classification results provide data support for the remote sensing identification of wetland information in Hangzhou City. -
表 1 杭州市土地覆盖分类体系
一级类型 二级类型 类型描述 GEE影像图例 自然湿地 河流 有流水的天然线性水体 湖泊 有积水且面积超过8公顷的天然多边形水体 草本沼泽 草本植物占优势的自然湿地 滩地 在正常水位与河湖洪水位之间或低于季节性洪水位的滩面 人造湿地 水库/坑塘 有积水且有明显坝体的人工多边形水体 运河/水渠 有流水且有明显坝体或直线边界的人工线性水体 养殖池 用于水产养殖的多边形水体 非湿地 林地 包括森林和灌木的天然木本植被覆盖区域 建设用地 包括城镇、工矿、交通设施等各类住宅区且人工建筑活动形成的区域 耕地 以种植农作物为主的土地区域 草地 天然草本植物覆盖区域 裸地 植被覆盖率低于10%的自然覆盖区域 表 2 特征信息表
特征类型 特征信息 特征描述 光谱特征 影像自带波段
传统光谱特征
红边光谱特征B2、B3、B4、B5、B6、B7、B8、B11、B12、B8A
NDVI、MNDWI、BSI、MSAVI、LSWI、EVI、NDBI
MCARI、PSSRI、NDI45、NDre1、NDre2变换特征 主成分分析
缨帽变换提取影像主成分变换的前3个主成分
亮度、绿度、湿度分量纹理特征 灰度共生矩阵 角二阶距、对比度、相关度、方差、逆差矩、和平均、
和方差、熵、和熵地形特征 坡度、海拔 NASADEM数据的地形特征 表 3 光谱特征描述
指数简称 指数全称 计算公式 NDVI Normalized Difference Vegetation Index (B8-B4)/ (B8+B4) MNDWI Modified Normalized Difference Water Index (B3-B11)/(B3+B11) BSI Bare Soil Index ((B4 + B11) - (B8 + B2)) / ((B4 + B11) + (B8 + B2)) MSAVI Modified Soil Adjusted Vegetation Index $ [2\times \mathrm{B}8\mathrm{ }+\mathrm{ }1-\sqrt{{\left(2\times \mathrm{B}8+1\right)}^{2}-8\times \left(\mathrm{B}8-\mathrm{B}4\right)} $] /2 LSWI Land Surface Water Index (B8-B11)/(B8+B11) EVI Enhanced Vegetation Index 2.5×(B8-B4)/(B8+6.0×B4-7.5×B2+1) NDBI Normalized Difference Built-up Index (B11-B8)/ (B11+B8) MCARI Modified Chlorophyll Absorption Ratio Index [(B5-B4)-0.2×(B5-B3)]×(B5-B4) PSSRI Pigment Specific Simple Ratio Index B7/B4 NDI45 Normalized Difference Index (B5-B4)/(B5+B4) NDre1 Normalized Difference red-edge1 (B6−B5)/(B6+B5) NDre2 Normalized Difference red-edge2 (B7−B5)/(B7+B5) 表 4 特征组合方案
实验方案 特征组合 实验方案 特征组合 1 TF 9 TF、RESF、TRF 2 TF、TSF 10 TF、RESF、TEF 3 TF、RESF 11 TF、TRF、TEF 4 TF、TRF 12 TF、TSF、TRF、TEF 5 TF、TEF 13 TF、RESF、TRF、TEF 6 TF、TSF、RESF 14 TF、TSF、RESF、TEF 7 TF、TSF、TRF 15 TF、TSF、RESF、TRF 8 TF、TSF、TEF 16 TF、TSF、RESF、TRF、TEF 注:TF、TSF、RESF、TRF、TEF分表表示地形特征、传统光谱特征、红边光谱特征、变换特征、纹理特征. 表 5 不同分类方案的土地覆盖类型分类总精度对比
实验方案 OA (%) Kappa 实验方案 OA (%) Kappa 1 TF 71.8 0.56 9 TF、RESF、TRF 85.6 0.75 2 TF、TSF 81.2 0.69 10 TF、RESF、TEF 82.2 0.70 3 TF、RESF 79.8 0.67 11 TF、TRF、TEF 83.0 0.71 4 TF、TRF 80.6 0.68 12 TF、TSF、TRF、TEF 88.8 0.80 5 TF、TEF 75.2 0.62 13 TF、RESF、TRF、TEF 85.0 0.75 6 TF、TSF、RESF 83.8 0.72 14 TF、TSF、RESF、TEF 87.4 0.78 7 TF、TSF、TRF 84.8 0.73 15 TF、TSF、RESF、TRF 85.0 0.74 8 TF、TSF、TEF 84.0 0.72 16 TF、TSF、RESF、TRF、TEF 93.0 0.88 注:TF、TSF、RESF、TRF、TEF分表表示地形特征、传统光谱特征、红边光谱特征、变换特征、纹理特征,OA为总体精度,Kappa为Kappa系数,红色越深代表精度越低,绿色越深代表精度越高,下同. 表 6 不同分类方案的湿地类型分类总精度对比
实验方案 OA (%) Kappa 实验方案 OA (%) Kappa 1 TF 64.4 0.53 9 TF、RESF、TRF 73.7 0.66 2 TF、TSF 68.9 0.60 10 TF、RESF、TEF 72.5 0.64 3 TF、RESF 70.0 0.62 11 TF、TRF、TEF 69.7 0.60 4 TF、TRF 71.2 0.63 12 TF、TSF、TRF、TEF 72.5 0.65 5 TF、TEF 66.3 0.56 13 TF、RESF、TRF、TEF 77.4 0.71 6 TF、TSF、RESF 72.2 0.64 14 TF、TSF、RESF、TEF 73.4 0.66 7 TF、TSF、TRF 70.7 0.62 15 TF、TSF、RESF、TRF 75.6 0.69 8 TF、TSF、TEF 68.5 0.60 16 TF、TSF、RESF、TRF、TEF 81.2 0.75 -
[1] 裴理鑫,叶思源,何磊,等. 中国湿地资源与开发保护现状及其管理建议[J]. 中国地质,2023,50(2):459 [2] 符超文/摄,金韬文/摄. 共同努力,谱写全球湿地保护新篇章:《湿地公约》第十四届缔约方大会综述[J]. 绿色中国,2022(21):10 [3] 李智,张慧,刘光,等. 近50年来杭州城西湿地时空变化分析[J]. 科技通报,2018,34(8):255 doi: 10.13774/j.cnki.kjtb.2018.08.052 [4] BROWN C F,BRUMBY S P,GUZDER-WILLIAMS B,et al. Dynamic World,Near real-time global 10 m land use land cover mapping[J]. Scientific Data,2022,9:251 doi: 10.1038/s41597-022-01307-4 [5] 邓雅文,蒋卫国,王晓雅,等. 基于随机森林算法和知识规则的国际湿地城市精细湿地分类:以常德市为例[J]. 遥感学报,2023,27(6):1426 [6] NOEL,GORELICK. Google earth engine:planetary-scale geospatial analysis for everyone[J]. Remote Sensing of Environment,2017,202:18 doi: 10.1016/j.rse.2017.06.031 [7] 郝斌飞,韩旭军,马明国,等. Google Earth Engine在地球科学与环境科学中的应用研究进展[J]. 遥感技术与应用,2018,33(4):600 [8] 王小娜,田金炎,李小娟,等. Google Earth Engine云平台对遥感发展的改变[J]. 遥感学报,2022,26(2):299 doi: 10.11834/jrs.20211317 [9] YANG,LIU. Vietnam wetland cover map:using hydro-periods Sentinel-2 images and Google Earth Engine to explore the mapping method of tropical wetland[J]. International Journal of Applied Earth Observation and Geoinformation,2022,115:103122 doi: 10.1016/j.jag.2022.103122 [10] ARSALAN,GHORBANIAN. Improved land cover map of Iran using Sentinel imagery within Google Earth Engine and a novel automatic workflow for land cover classification using migrated training samples[J]. ISPRS Journal of Photogrammetry and Remote Sensing,2020,167:276 doi: 10.1016/j.isprsjprs.2020.07.013 [11] 田晓晖,孙庆来,李国志,等. 杭州市湿地资源保护对策[J]. 湿地科学与管理,2020,16(3):33 doi: 10.3969/j.issn.1673-3290.2020.03.10 [12] 胡云丽,陈斌. 杭州市四季气候变化特征分析[J]. 浙江气象,2020,41(2):1 doi: 10.16000/j.cnki.zjqx.2020.02.001 [13] 田颖,陈卓奇,惠凤鸣,等. 欧空局哨兵卫星Sentinel-2A/B数据特征及应用前景分析[J]. 北京师范大学学报(自然科学版),2019,55(1):57 [14] LÓPEZ-CALATAYUD N,MÁRQUEZ-ROMANCE A M,GUEVARA-PÉREZ E. Wetland management modeling in the frame of Ramsar convention[J]. Journal of Applied Water Engineering and Research,2022,10(4):296 doi: 10.1080/23249676.2021.1982028 [15] 郑建华. 基于Landsat TM/OLI遥感影像的杭州土地利用变化时空特征研究[D]. 杭州:浙江农林大学. [16] 王翀,林慧龙. 中国内陆天然湿地的类型特征及分布规律-Ⅰ类的划分[J]. 草业学报,2012,21(1):262 doi: 10.11686/cyxb20120134 [17] HILL M J. Vegetation index suites as indicators of vegetation state in grassland and savanna:an analysis with simulated SENTINEL 2 data for a North American transect[J]. Remote Sensing of Environment,2013,137:94 doi: 10.1016/j.rse.2013.06.004 [18] 张磊,宫兆宁,王启为,等. Sentinel-2影像多特征优选的黄河三角洲湿地信息提取[J]. 遥感学报,2019,23(2):313 [19] PAUL,MACINTYRE. Efficacy of multi-season Sentinel-2 imagery for compositional vegetation classification[J]. International Journal of Applied Earth Observation and Geoinformation,2020,85:101980 doi: 10.1016/j.jag.2019.101980 [20] 马慧娟,高小红,谷晓天. 随机森林方法支持的复杂地形区土地利用/土地覆被分类研究[J]. 地球信息科学学报,2019,21(3):359 doi: 10.12082/dpxxkx.2019.180346 [21] 杨仙保,张王菲,孙斌,等. 基于GEE和Sentinel-2时序数据的呼伦贝尔沙地及其周边植被类型识别研究[J]. 遥感技术与应用,2022,37(4):982 [22] WANG X Y,JIANG W G,PENG K F,et al. A framework for fine classification of urban wetlands based on random forest and knowledge rules:taking the wetland cities of Haikou and Yinchuan as examples[J]. GIScience & Remote Sensing,2022,59(1):2144 [23] 何云,黄翀,李贺,等. 基于Sentinel-2A影像特征优选的随机森林土地覆盖分类[J]. 资源科学,2019,41(5):992 [24] LAONAMSAI J,JULPHUNTHONG P,SAPRATHET T,et al. Utilizing NDWI,MNDWI,SAVI,WRI,and AWEI for estimating erosion and deposition in Ping River in Thailand[J]. Hydrology,2023,10(3):70 doi: 10.3390/hydrology10030070 [25] 姚茂林,江洪,张丽玉. 基于Sentinel-2影像红边光谱指数与特征优选的竹林提取研究[J]. 海南大学学报(自然科学版),2022,40(4):373 [26] 闫国东,左雪漫,陈瑾,等. 基于多特征优选的Sentinel-2遥感影像林分类型分类[J]. 森林工程,2023,39(3):12 doi: 10.3969/j.issn.1006-8023.2023.03.002 [27] 程丽娜,钟才荣,李晓燕,等. Sentinel-2密集时间序列数据和Google Earth Engine的潮间带湿地快速自动分类[J]. 遥感学报,2022,26(2):348 doi: 10.11834/jrs.20211311 [28] 梁爽,宫兆宁,赵文吉,等. 基于多季相Sentinel-2影像的白洋淀湿地信息提取[J]. 遥感技术与应用,2021,36(4):777 [29] 徐娜,项超超. 基于Sentinel-2影像的河流湖泊遥感提取[J]. 测绘与空间地理信息,2023,46(5):120 doi: 10.3969/j.issn.1672-5867.2023.05.032 [30] SIYAMTHANDA G,TIMOTHY D,DOMINIC M. Leveraging Google Earth Engine platform to characterize and map small seasonal wetlands in the semi-arid environments of South Africa[J]. Science of the Total Environment,2022,803:150139 doi: 10.1016/j.scitotenv.2021.150139 -