FSDAF 2.0 : improving the performance of retrieving land cover changes and preserving spatial details

Abstract

Spatiotemporal fusion is a feasible solution to resolve the tradeoff between the temporal and spatial resolutions of remote sensing images. However, the development of spatiotemporal fusion algorithms has not yet reached maturity, and existing methods still face many challenges, e.g., accurately retrieving land cover changes and improving the robustness of fusion algorithms. The Flexible Spatiotemporal DAta Fusion (FSDAF) method proposed by Zhu et al. in 2016 solved the abovementioned problems to some extent. However, FSDAF has two shortcomings that can be further improved: (1) FSDAF is prone to losing spatial details and predicting a “blurrier” image due to the input of coarse pixels containing type change information and a large amount of boundary information for unmixing calculation, and (2) FSDAF does not optimize the areas of land cover change. In this paper, an improved FSDAF method incorporating change detection technology and an optimized model for changed-type areas (FSDAF 2.0) was proposed to improve the aforementioned problems. Based on the existing FSDAF algorithm, FSDAF 2.0 excludes changed pixels and boundary pixels for unmixing calculation, and establishes a model to optimize the changed pixels. Its performance was compared with that of the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM), the original FSDAF, and the enhanced FSDAF that incorporates sub-pixel class fraction change information (SFSDAF). Two sites consisting of landscapes with heterogeneous and large-scale abrupt land cover changes were employed for testing. The results of the experiments demonstrate that FSDAF 2.0 effectively improves the shortcomings of FSDAF, blends synthetic fine-resolution images with higher accuracy than that of the other three methods at two different sites, and strengthens the robustness of the fusion algorithm. More importantly, FSDAF 2.0 has a powerful ability to retrieve land cover changes and provides a feasible way to improve the performance of retrieving land cover changes. Consequently, FSDAF 2.0 has great potential for monitoring complex dynamic changes in the Earth's surface.