LANDSCAPE INDICES BASED TWO-RANK SAMPLING FOR LAND COVER VALIDATION

摘要

A web-based validation is necessary for assessing the accuracy of Globalland30. As one of the obstacles in validation of global land cover data, it is a challenge to effectively select a reasonable sample dataset. Global land cover is heterogeneous and complex. However, some sampling plans based on probability and mathematical statistics don't consider the landscape heterogeneity. To address this disadvantage of the two-rank acceptance sampling plan(TRASP) for validation of land cover data, landscape indices are used to improving this sampling plan. Landscape indicator is an typical method of quantifying landscape heterogeneity. The landscape indices based sampling combined landscape indicators and TRASP, with an innovation on the two side of computing samples' size and distributing samples. Firstly, landscape shape index(LSI) is introduced in the equation of TRASP for the sample size. Further, the validation area is divided into some small grids, and LSI is used to find more valuable hotspots from these grids to distributing samples. The theory and formulas are presented in this paper, and an example application is provided in which the sample of a land-cover map is chosen.