Spatio-temporal fusion assisted by gap filling under different cloud conditions

In recent years, the spatio-temporal fusion has gained growing attention for increasing the quality of remote sensing images. However, the images used for spatio-temporal fusion are usually limited by cloud pollution. Gap filling can provide spatially complete and temporally proximate auxiliary data to alleviate this challenge, but factors such as cloud size can affect the process. To address this issue, this paper explored the influence of cloud conditions on the filling then spatiotemporal fusion (FSTF) method, termed C-FSTF. Specifically, we used the virtual image pairs for spatio-temporal fusion (VIPSTF-SW) method based on a modified neighborhood similar pixel interpolator (MNSPI) and conducted experiments on Landsat-MODIS image pairs from three regions to evaluate fusion performance. The results show that: 1) By using the reflectance information of clear pixels in cloudy images, C-FSTF achieves larger prediction accuracy than discarding cloudy images directly. 2) The fusion accuracy decreases with increasing overlap of cloud pixels at fine resolution and longer gap filling intervals. 3) When the cloud distribution is more scattered, C-FSTF performs more stably due to more clear-sky pixels at the cloud edges.