Uncertainties of legend conversion and inconsistency analysis of global land cover products---A case study in Heihe River Basin

Abstract:

In this study, the spatial and areal inconsistencies of the most recent global land cover datasets were analyzed quantitatively with widely used common thematic legends (IGBP-17, modified IGBP-9, IPCC-5 and 3 classes) in Heihe River Basin. These products included Finer Resolution Observation and Monitoring of Global Land Cover (hereafter abbreviated as GLC30), Global Land Cover Map (hereafter abbreviated as GlobCover2009) and the IGBP layer of MODIS Collection 5.1 land cover (hereafter abbreviated as MODISLC_IGBP). Before areal and spatial inconsistencies were assessed, uncertainty caused by legend conversion was also discussed quantitatively. Majority filtering and statistical methods were adopted to upscale from high to low spatial resolution classifications and to compute inconsistencies respectively. The results indicated that IGBP-17 and “3 classes” were not suitable as common legends to compare land cover products using LCCS and FAO, because IGBP-17 itself included ambiguous type and too much aggregation in “3 classes” was less meaningful for users due to less information. In general, 5 - 12 kinds of land cover type were suitable quantity to compare different land cover products. In the IGBP-9 and IPCC-5 legend conversion, GLC30 also had “zero” uncertainties, and the GlobCover2009 had the largest uncertainties, the values were 17.3%, 7.44%, 0.46% and 16.2%, 5.95%, 0.40% in the mountainous, oasis and desert area respectively. The hierarchical classification system with explicit attribution information was suggested to global land cover producers by authors, because it can be used to reduce effectively uncertainty of legend conversion caused by arbitrary errors. The order of areal inconsistency showed opposite to overall spatial inconsistency. Areal inconsistency between GLC30 and GlobCover2009 was the least among the three groups comparison, with values less than 19.71%, 8.10%, 1.58% and 27.20%, 8.05% and 1.53% in mountainous, oasis and desert areas, respectively. However, the overall spatial inconsistency between them showed the largest value, with values 18.46% and 18.31%.