Sub-pixel land-cover mapping with improved fraction images upon multiple-point simulation

► Outputs of soft classification inherently contain uncertainty. ► Multiple-point simulation (MPS) through training image characterizes the spatial structural properties of a surface object/class. ► Quality of the fraction image is improved by fusing MPS result. ► Accuracies of sub-pixel mapping (SPM) simulated images then increase. ► Estimation of net primary productivity (NPP) is an application of this proposed method.