Cloud cover presents a major challenge to the effective use of high-spatial-resolution satellite imagery in time-series remote sensing applications. Existing spatiotemporal fusion algorithms rely heavily on high-quality inputs, limiting their applicability under widespread and complex cloud conditions. This study investigates an integrated workflow that combines a Modified Neighborhood Similar Pixel Interpolation (MNSPI) technique for cloud removal with three representative spatiotemporal fusion algorithms—Spatial and Temporal Adaptive Reflectance Fusion Model, Enhanced Spatial and Temporal Adaptive Reflectance Fusion Model (ESTARFM), and Flexible Spatiotemporal Data Fusion. Using GF-1 (Gaofen-1) Wide Field of View and Sentinel-2 Level-2 A imagery over Gaoyou Lake and its surrounding plains in Yangzhou, Jiangsu Province, China, we evaluate the performance of the proposed workflow under both simulated and real thick cloud scenarios. The results show that MNSPI performs well under low cloud coverage and preserves spatial structure under challenging, dense, and realistic cloud conditions. Images preprocessed with MNSPI lead to improved accuracy in subsequent fusion results, with ESTARFM demonstrating the best performance in spectral prediction and spatial detail reconstruction. Furthermore, land cover classification based on the fused imagery yields overall accuracy and Kappa values comparable to those of original cloud-free images, confirming the reliability and practical utility of the proposed “cloud removal + fusion” framework. The findings highlight the potential of this method to enhance the effective use of remote sensing data in cloud-prone regions and provide a robust solution for long-term environmental monitoring under complex atmospheric conditions. • We propose a cloud removal → spatiotemporal fusion workflow to generate fine-resolution time series in cloud-prone regions. • MNSPI preprocessing substantially reduces cloud-induced artifacts and improves the radiometric completeness of fine-resolution references. • After cloud removal, STARFM/ESTARFM/FSDAF show consistent performance gains, with clear differences in spectral–texture trade-offs. • The fused products achieve classification accuracy close to cloud-free references, indicating strong downstream usability. • This workflow enables more reliable high-frequency monitoring of dynamic land-surface processes (e.g., crops and coastal wetlands) when cloud-free observations are scarce.
Cui et al. (Sun,) studied this question.