IT-STF: A Fast and Imperfect Input Tolerant Spatio-Temporal Fusion Method

Spatial–temporal fusion (STF) technology has emerged as a cost-effective and convenient solution for obtaining dense time-series high-resolution satellite images. Large-scale applications require a practical STF method that is fast and robust to be used in various climate zones and with various input data qualities. Current spatio-temporal fusion methods require manual selection of cloud-free input images and exhibit inefficiencies in data production, rendering them unsuitable for large-scale applications. To solve this issue, this study proposes the Imperfect input tolerant STF (IT-STF) method, which consists of three parts: quality control of imperfect inputs, fast window-by-window fusion of images, and compensation of spectral and spatial information. This method utilizes time-series data and incorporates a quality control process before spatio-temporal fusion. It also combines the advantages of existing fast fusion methods to compute the fusion coefficients for all bands on a window-by-window basis. The experiments demonstrate that compared with the four traditional methods, the fusion results of the IT-STF method provide higher accuracy under different climates and land cover types in regions with frequent cloudy days. In terms of computational efficiency, IT-STF is comparable to the fast STF method. In terms of large-area data production, IT-STF maintains higher consistency among adjacent slices.