Efficient Large Scale Reflectance Fusion Model for Generating High Temporal Landsat-like Data by Fusing Multi-Satellite Image

The emergence of spatiotemporal data fusion aims to address the inherent contradiction between temporal and spatial resolutions, seeking to generate high-resolution images in both dimensions. We present the Efficient Large-Scale Reflectance Fusion (ELSRF) approach on the Google Earth Engine (GEE) platform for resolving the temporal-spatial resolution trade-off in remote sensing data fusion. ELSRF effectively integrates Landsat and MODIS data, utilizing rubber-sheeting, Bayesian averaging, and weighted regression. Firstly, to bridge sensor disparities, rubber-sheeting and ordinary least squares methods are employed, yielding a fused MODIS-Landsat dataset. Then, Bayesian model averaging is harnessed to seamlessly integrate the initial fused data with climatological means, enhancing both precision and contextual understanding. Finally, ELSRF involves applying a weighted regression model, a critical step in refining the preliminary fusion outcomes. Validation demonstrates its accuracy in comparison with ground truth and existing methods: (1) Favorable accuracy is observed for both Landsat and Sentinel-2 comparisons; (2) ELSRF outperforms STARFM and FSDAF in visual and accuracy aspects; (3) NDVI indices derived from fusion data show reliable performance in validation with MOD13Q1 data. The experimental implementation of ELSRF on the GEE platform facilitates the swift and efficient operation of spatiotemporal fusion at a large scale.