Evaluation of four Spatiotemporal Gap-Filling Methods in Crop Phenology Monitoring

Abstract

The high spatial resolution land surface phenology (LSP) monitoring is often limited by the temporal discontinuity of high spatial resolution observations. Many gap-filling methods have been proposed for LSP monitoring, however, a thorough intercomparison and evaluation is still lacking. Four widely-used methods, including the Spatial and Temporal Adaptive Reflectance Fusion Model (STARFM), the Flexible Spatiotemporal DAta Fusion (FSDAF), Multi-year based model, and Spatiotemporal Shape Matching Model (SSMM) were selected to extract green-up date (GUD) based on the Harmonized Landsat and Sentinel-2 (HLS) dataset. The results of the four methods show consistency with those of PhenoCam sites (R2>0.64) and there is a lag phenomenon. Compared within 3×3 VIIRS pixel window, the difference between mean VIIRS GUDs and aggregated 30m GUDs is small and the mean absolute difference is less than 7 days. Comprehensively, SSMM has high consistency (R2=0.75) and smaller bias (Bias=7.3days), which shows more potential in LSP monitoring.