A landscape shape index-based sampling approach for land cover accuracy assessment

Spatial sampling design is one of the key steps in land cover accuracy assessment, and many traditional sampling approaches may not achieve credible spatial sampling due to the high spatial heterogeneity of land cover. This paper characterizes the spatial heterogeneity with three-level LSIs and determines the subsequent sample sizes and their spatial distributions. The three-level LSIs are rLSI in a region, cLSI for each land cover class and uLSI in each geographic sampling unit in the region. The rLSIs are used to derive appropriate sample sizes in the target regions. The cLSIs are used to assure that larger sample numbers are allocated to land cover classes with higher spatial heterogeneity. The uLSIs provide useful measures for selecting optimal geographic units in which sample sites will be located. This LSI-based sampling approach can derive the sample sizes and determine their distributions in an adaptive way according to the spatial heterogeneity. An experimental case study further demonstrates that the LSI-based sampling approach obtains more appropriate sample sizes for each region, sufficient sample numbers for rare classes, and optimal sample distributions in the geographical space.