Improving spatiotemporal data fusion method in multiband images by distributing variates

Several spatiotemporal data fusion methods have been developed to generate continuous fine-resolution satellite imagery using widely available datasets. This study introduces the Residual Distribution-based Spatiotemporal Data Fusion Method (RDSFM), designed to enhance fusion accuracy. RDSFM addresses residuals caused by spatial and temporal variations, utilizing the IR-MAD algorithm to estimate subpixel distribution weights based on multivariate data collected over time. Compared to existing methods, RDSFM offers several key advantages: (1) It accurately predicts seasonal variations in bands such as red and NIR, (2) It effectively handles heterogeneous landscapes and shifting land cover, and (3) It requires only one high-resolution reference image as input, minimizing data requirements. The effectiveness of RDSFM was validated using real satellite images and benchmarked against methods like unmixing-based data fusion (UBDF). Experimental results show that RDSFM successfully captures seasonal changes in coarse-resolution bands, particularly in red and NIR, making it especially useful for vegetation analysis. Additionally, RDSFM demonstrates strong performance in managing heterogeneous landscapes and areas with dynamic land cover, as confirmed by both visual and quantitative assessments.