Generation of Cloud-Free High Spatial Resolution Optical Images Using Spatio-Temporal Fusion

Environmental monitoring using optical satellite imagery often faces challenges related to limited data availability due to the inherent resolution of the imagery and cloud cover. To address this issue, this study presents a novel image fusion method for constructing cloud-free high spatial resolution optical images. The proposed method combines a cloud removal technique to restore cloud-covered regions and a spatio-temporal fusion process that yields images with both enhanced spatial and temporal resolutions. Machine learning-based regression modeling and Poisson blending are employed to restore cloud-covered regions in low spatial resolution image. Then, spatio-temporal fusion using object-based weighting is implemented to predict images including detailed spatial features and spectral patterns of the prediction date. To demonstrate the effectiveness of the proposed method, a simulation experiment was conducted using Sentinel-2 and PlanetScope images in croplands. Comparative analyses with existing methods revealed the superior prediction accuracy of the proposed method in terms of cloud removal and spatio-temporal fusion. The experimental results indicate that the proposed method can be effectively applied to generate a time series of high spatial resolution optical images for environmental monitoring.