A Two-Step Spatio-Temporal Satellite Image Fusion Model for Temporal Changes of Various LULC

This paper proposes a two-step spatio-temporal fusion model (TSTFM) for generating synthetic satellite remote sensing images with high-spatial and high-temporal resolution (HSaHTeR) based on one pair of prior images, which contain one low-spatial but high-temporal resolution (LSaHTeR) image and one high-spatial but low-temporal resolution (HSaLTeR) image. Considering both phenology and type surface temporal changes, the two steps in TSTFM are adopted to handle these two kinds of changes respectively, which are based on weighted mean and example-based image super-resolution approaches accordingly. In addition, a relative radiometric normalization process is conducted before performing the two-step spatio-temporal fusion (STF) process, which aims to calibrate radiometric differences of different kinds of satellite sensors. The proposed method was tested on two sets of test data: surface with mainly LULC phenology changes and surface with primarily LULC type changes. Experimental results show that TSTFM can capture both phenology and type changes efficiently and precisely even with one-pair prior images, and it can also maintain its robustness when facing extremely complex LULC.