基于众源地理数据的GlobeLand30人造地表数据精度验证

Employing Crowdsourced Geographic Information for GlobeLand30 Artificial Surface Validation

GlobeLand30数据已被广泛应用于生态、环境、气候等研究领域,对其进行精度验证是保证GlobeLand30产品推广应用的首要前提。本文提出了一种基于众源地理数据的GlobeLand30人造地表精度验证方法:首先利用随机抽样的方法布设样本点,其次应用LDA(latent Dirichlet allocation)主题模型计算众源数据文档的地表覆盖主题,最后通过构建样本点与GlobeLand30数据产品之间的混淆矩阵,以实现人造地表数据的精度验证。研究区域为我国27个省级行政区,众源地理数据选用POI数据,对本文提出的方法进行了实例验证。结果表明:2000年和2010年的人造地表总体精度分别为87.63%和88.81%,这与GlobeLand30人造地表第三方评价精度86.97%相近,验证了本文方法的有效性。

GlobeLand30 is the world's first 30-meter resolution globe-land cover dataset,and has been widely used in ecological,environmental,climate and other research fields.Ensuring the product accuracy and data quality is the premise of the promotion and application of GlobeLand30.In this paper,we propose an artificial surface validation method based on crowdsouring geographic data.Firstly,the random sampling method is used to allocation the samples.Secondly,the crowd-souring geographic data is analyzed to calculate the topic of land cover type using LDA topic model.Finally,the accuracy of artificial surface is evaluated by confusion matrix.The 27 provincial administrative regions of China as study area,and POI are used to validate and analyze the accuracy of GlobeLand30 in 2000 and 2010.Experiment results represented promises in land cover validation,and given that artificial surface shows an overall accuracy of 87.63%and 88.81%in 2000 and 2010 respectively,which was close to the validation result of 86.97%given by the third party institutions.