中国1:25万土地覆盖遥感制图精度评价——以鄱阳湖地区为例

Field Accuracy Validation of Land Cover Remote Sensing Mapping at the Scale of 1:250 000 in the Poyang Lake Area of China

全球及区域尺度的土地覆盖数据是陆地表层过程研究的重要基础。土地覆盖遥感制图是全球变化和区域可持续发展研究重要的支撑数据,制图精度评价对于数据生产者和数据用户都具有非常重要的意义。2011年在鄱阳湖地区的野外考察共采集包括定点验证、GPS以及解译标志3种类型的土地覆盖样点321个,本文利用剔除了时间差异影响后的287个土地覆盖样点,将样点的实际土地覆盖类型与遥感制图中相应位置的土地覆盖类型进行对比,并利用分层次评估法,即分别在土地覆盖一级类和二级类两个尺度上,采用正确得1分,错误得0分的计分方法,对2005年中国1:25万土地覆盖遥感制图在鄱阳湖地区的精度进行实地验证。结果表明:(1)在土地覆盖一级类型尺度上总体的制图精度为61.67%。其中,湿地/水体的制图精度为100%,农田的制图精度为98.4%,森林的制图精度为80.0%,聚落和草地的精度均低于20%。(2)在二级类型尺度上总体的制图精度为44.25%。其中,2个草地和3个森林及1个农田的二级类型的分类精度为0,旱地、城镇建设用地和农村聚落的分类精度也很低,分别为21.1%、29.0%和1.7%。实地调查发现,2005年左右的全国1∶25万土地覆盖遥感制图基本上反映了鄱阳湖地区的土地覆盖状况。但是,对于一些具有过渡性质的土地覆盖类型,如森林和草地等,仅依靠遥感技术准确识别区分仍有一定的难度。

Global and regional land cover data is the foundational base for studies on land surface processes and modeling and land cover remote sensing mapping is important information for supporting studies on global change and regional sustainable development.Mapping accuracy assessment has a great significance to both data producers and data users,and to some extent,it is incomplete for the study of remote sensing mapping of land cover without an accuracy validation.From 2007 to 2009,two land cover datasets of China at the scale of 1∶250 000 in the 1980s and circa 2005 were produced jointly by eight institutes of Chinese Academy of Sciences（CAS）,respectively,including Institute of Remote Sensing Applications（IRSA） and Institute of Geographic Sciences and Natural Resources Research（IGSNRR）.In order to evaluate the accuracy of this remote sensing mapping of land cover in circa 2005,a field survey was carried out in the Poyang Lake area in August,2011,and 321 sampling sites of land cover categories including fixed-points validation,GPS points and interpretation keys were collected.Considering the conspicuous impact of time factor on the accuracy validation of remote sensing mapping,this paper removes 34 samples relating to time differences,and takes advantage of the hierarchical assessment method to compare the actual categories of land cover in field and those in remote sensing mapping at 287 survey sites at two scales,i.e.level-1 and level-2 land cover category,respectively,and uses the scoring method for assessing the mapping accuracy,that is,one point is scored if land cover category is classified correctly,and zero if land cover category is classified incorrectly.The results show that：（1） The overall mapping accuracy of land cover data of China in the Poyang Lake area based on 287 sampling sites at the level-1 scale is 61.67%,of which wetland/water body,farmland and forest with 19,124 and 20 sampling sites reaches 100%,98.4% and 80.0%,respectively.（2） The overall mapping accuracy of land cover dataset in 2005 at the level-2 category is 44.25%,of which the accuracy of two grass and three forest and one cropland level-2 categories is 0,and the very low accuracy of dry land,urban and rural settlement is 21.1%,29.0% and 1.7%,respectively.The dataset of land cover at the scale of 1∶250 000 in 2005 reflects the status of land cover in the Poyang Lake area well.However,there are still a lot of difficulties of different land cover categories,especially for atypical and transitional categories,such as forests and grasslands,that can be distinguished accurately only through the technology of remote sensing.