Spatial heterogeneity of errors in land cover data (Theory of Biomathematics and Its Applications XVI -Toward quantitative understanding for life Sciences-)

Abstract

Thematic Land cover (LC) maps attempt to describe the Earth's terrestrial surface, encompassing all attributes of the biosphere (International Panel on Climate Change, 2000). LC has been regarded as an important component of the Earth system which physically interacts with climate, topography, human impacts, and their complex interactions. As LC maps are required to cover an area widely from local to global scales, remotely sensed (RS) imagery is often used, that is classified into defined thematic land cover classes by a classification method such as statistical and machine learning models. It is hence important to make an accurate LC classification map for high-quality quantification of the Earth system component. To assess the accuracy of the thematic LC classification map, conventional summary measures of error, such as user's, producer's, and overall accuracies for per-pixel classification, and mean signed deviation (msd), mean absolute error (mae), root mean square error (rmse) and Pearson's correlation coefficient (r) for sub-pixel classification. However, these summary measures of error do not take any spatial information (e.g., spatial heterogeneity) of error into account (Foody, 2005, 2002). A spatially explicit approach for the assessment is helpful to identify spatial characteristics of errors. This study demonstrates one of the spatial measures of error for assessing thematic LC maps. In this paper, a map for forest aboveground biomass (AGB) in the Yucatan peninsula, Mexico, estimated by Rodríguez-Veiga et al. (2016), is assessed.