Improving the Accuracy of Remote Sensing Land Cover Classification by GEO-ECO Zoning Coupled with Geostatistical Simulation

Abstract

Land cover products obtained from remote sensing image classification inevitably contain a large number of false classification or uncertain pixels because of spectral confusion, image resolution limitation, and ground object complexity. The confusion matrix used to evaluate the classification accuracy cannot reflect the spatial variation. The information provided to users of land cover products is incomplete and uncertain. In this study, a method is presented to evaluate and improve the accuracy of land cover classification products by coupling Geo-Eco zoning and Markov chain geoscience statistical simulation. Validation points collected from various sources are used in the model calculation and accuracy verification of results. The pre-classified image that needs to be improved and Geo-Eco zoning attribute data are used as auxiliary data for co-simulation. Results show that the accuracy of Globeland30 data can be improved by more than 10% by coupling Geo-Eco zoning and Markov chain geostatistical simulation.