Abstract
Satellite-based time-series crop monitoring at the subfield level is essential to the efficient implementation of precision crop management. Existing spatiotemporal image fusion techniques can be helpful, but they were often proposed to generate medium-resolution images. This study proposed a high-resolution spatiotemporal image fusion method (HISTIF) consisting of filtering for cross-scale spatial matching (FCSM) and multiplicative modulation of temporal change (MMTC). In FCSM, we considered both point spread function effect and geo-registration errors between fine and coarse resolution images. Subsequently, MMTC used pixel-based multiplicative factors to estimate the temporal change between reference and prediction dates without image classification. The performance of HISTIF was evaluated using both simulated and real datasets with one from real Gaofen-1 (GF-1) and simulated Landsat-like/Sentinel-like images, and the other from real GF-1 and real Landsat/Sentinel-2 data on two sites. HISTIF was compared with the existing methods spatial and temporal adaptive reflectance fusion model (STARFM), FSDAF, and Fit-FC. The results demonstrated that HISTIF produced substantial reduction in the fusion error from cross-scale spatial mismatch and accurate reconstruction in spatial details within fields, regardless of simulated or real data. The images predicted by STARFM exhibited pronounced blocky artifacts. While the images predicted by HISTIF and Fit-FC both showed clear within-field variability patterns, HISTIF was able to reduce the spectral distortion more significantly than Fit-FC. Furthermore, HISTIF exhibited the most stable performance across sensors. The findings suggest that HISTIF could be beneficial for the frequent and detailed monitoring of crop growth at the subfield level.
Highlights
M ODERN crop production is moving toward precision planting, efficient management, intelligent decisionmaking, and quantitative implementation at field or even subfield levels [1]–[3]
The shapes of spatial matching filter (SMF) characterized by full width at half maximum (FWHM) were mostly unsymmetrical
Given that the maximum value of ΔRMSE was as small as 0.0004, the SMF could be regarded stable across time
Summary
M ODERN crop production is moving toward precision planting, efficient management, intelligent decisionmaking, and quantitative implementation at field or even subfield levels [1]–[3]. The mid-resolution satellite imagery with weekly or bi-weekly revisit frequency, such as Sentinel-2, Landsat, and ASTER, has the potential in monitoring crop growth status over critical growth stages [5], [7], [8], but it is typically difficult to use the imagery for mapping smallholder farms [9]. Those are the main crop management units in many Asian and African countries such as India, China, and Ethiopia, where the size of a typical field (
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
