Super-Resolution Land Cover Mapping with Spatial–Temporal Dependence by Integrating a Former Fine Resolution Map

Super-resolution mapping (SRM) is a technique to predict spatial locations of land cover classes at the subpixel scale within coarse resolution remotely sensed image pixels. Due to the lack of information about the spatial pattern of land covers, uncertainty always exists in resultant fine-resolution land cover maps. In the present work, by integrating a former fine-resolution land cover map, the spatial dependence used in existing SRM algorithms is extended into a novel spatial–temporal dependence used in the SRM algorithm (SRM_STD). The spatial–temporal dependence consists of the spatial dependences of former fine-resolution land cover map, the spatial dependences of latter coarse resolution fraction images, and the corresponding dependence between former and latter land cover maps. By considering the spatial–temporal dependences of subpixels, SRM_STD can inherit valuable land cover information from the former fine-resolution land cover map, and reduce the uncertainty of SRM to a large extent. The performance of the proposed SRM_STD algorithm is assessed using a subset of the National Land Cover Database datasets and land cover maps produced by Landsat imagery in an area of rapid urban expansion. The results of two experiments show that the former dependence has little influence on the result, whereas the corresponding dependence plays a crucial role on the result. With a large weight of corresponding dependence, the proposed SRM_STD algorithm can produce fine-resolution land cover maps with higher accuracy than those of hard classification and the pixel swapping algorithm.