cuFSDAF: An Enhanced Flexible Spatiotemporal Data Fusion Algorithm Parallelized Using Graphics Processing Units

Spatiotemporal data fusion is a cost-effective way to produce remote sensing images with high spatial and temporal resolutions using multisource images. Using spectral unmixing analysis interpolation, the flexible spatiotemporal (FSDAF) algorithm suitable for heterogeneous landscapes capable of capturing abrupt land-cover changes. However, extensive computational complexity FSDAF prevents its use in large-scale applications mass production. Besides, domain decomposition strategy causes accuracy loss at edges subdomains due insufficient consideration edge effects. In this study, an enhanced (cuFSDAF) proposed address these problems, includes three main improvements. First, TPS interpolator replaced by accelerated inverse distance weighted (IDW) reduce complexity. Second, parallelized based on compute unified device architecture (CUDA), widely used parallel computing framework graphics processing units (GPUs). Third, adaptive (ADD) method improve enable GPUs varying capacities deal datasets any size. Experiments showed while obtaining similar accuracies up-to-date deep-learning-based method, cuFSDAF reduced time significantly achieved speed-ups 140.3–182.2 over original program. efficiently producing fused both support long-term land surface dynamics. Source code test available https://github.com/HPSCIL/cuFSDAF.