Comparison of Five Spatio-Temporal Satellite Image Fusion Models over Landscapes with Various Spatial Heterogeneity and Temporal Variation

Abstract

In recent years, many spatial and temporal satellite image fusion (STIF) methods have been developed to solve the problems of trade-off between spatial and temporal resolution of satellite sensors. This study, for the first time, conducted both scene-level and local-level comparison of five state-of-art STIF methods from four categories over landscapes with various spatial heterogeneity and temporal variation. The five STIF methods include the spatial and temporal adaptive reflectance fusion model (STARFM) and Fit-FC model from the weight function-based category, an unmixing-based data fusion (UBDF) method from the unmixing-based category, the one-pair learning method from the learning-based category, and the Flexible Spatiotemporal DAta Fusion (FSDAF) method from hybrid category. The relationship between the performances of the STIF methods and scene-level and local-level landscape heterogeneity index (LHI) and temporal variation index (TVI) were analyzed. Our results showed that (1) the FSDAF model was most robust regardless of variations in LHI and TVI at both scene level and local level, while it was less computationally efficient than the other models except for one-pair learning; (2) Fit-FC had the highest computing efficiency. It was accurate in predicting reflectance but less accurate than FSDAF and one-pair learning in capturing image structures; (3) One-pair learning had advantages in prediction of large-area land cover change with the capability of preserving image structures. However, it was the least computational efficient model; (4) STARFM was good at predicting phenological change, while it was not suitable for applications of land cover type change; (5) UBDF is not recommended for cases with strong temporal changes or abrupt changes. These findings could provide guidelines for users to select appropriate STIF method for their own applications.

Adaptive-SFSDAF for Spatiotemporal Image Fusion that Selectively Uses Class Abundance Change Information

“…Table 2 shows the quantitative comparison results of the five methods. It appears that UBDF has the worst performance among all the methods, which is consistent with the results reported in [20]. For all the bands, SFSDAF generated lower RMSE and MAD and higher CC, SSIM and PSNR values compared with those of FSDAF.…”

“…To evaluate the performance of the adaptive-SFSDAF algorithm quantitatively and visually, four comparison algorithms are selected as benchmark methods: FSDAF [11], SFSDAF [39], FIT-FC [9], and UBDF [27]. The above four methods were selected due to the following reasons: (1) FSDAF is a robust model at various scales [20]; (2) SFSDAF is a recently developed fusion algorithm based on FSDAF and had better performance than the existing representative fusion methods in all of the experiments reported as it incorporated sub-pixel class fraction change information in the fusion methods; (3) FIT-FC is computationally efficient in comparison to other fusion algorithms in the literature; and (4) UBDF is the most cited model in the unmixing based methods. Among all experiments, for FSDAF, SFSDAF and UBDF, the number of land cover classes was set to 4, for FSDAF, SFSDAF and FIT-FC, the number of similar pixels was set to 20, and the size of the sliding window was set to 16.…”

“…In the spatial domain, such a method estimates the predicted image information through the self-similar characteristics in the scene. Such methods are often referred to as weighted function-based methods [20]. The spatial and temporal adaptive reflectance fusion model (STARFM) is the one of the earliest to establish the satellite fusion model [1], which is simple, flexible and the most widely used.…”

ERTFM: An Effective Model to Fuse Chinese GF-1 and MODIS Reflectance Data for Terrestrial Latent Heat Flux Estimation

“…Firstly, the large uncertainties are mainly attributed to the defectiveness of the prior fine-coarse image pair, which meant that less supplementary information was available. Previous studies also supported this finding, and suggested that the number of fine resolution images that are inputted has a significant influence on the performance of fusion methods [58]. Significantly, the fusion images utilized in our study are GF-1 and MODIS products, which are not identical to the products used in the other studies.…”

An object-based spatiotemporal fusion model for remote sensing images

“…For example, the Fit-FC proposed to retrieve strong temporal changes performs more effectively than STARFM and FSDAF in capturing fast phenological changes (Q. Wang & Atkinson, 2018), but shows lower accuracy than FSDAF in retaining image structures when blending heterogeneous sites (Liu et al, 2019a). The accuracy of learning-based methods decreases when spatial heterogeneity is high and spectral scale differences between low-and high-resolution images are large (Zhu et al, 2016).…”