Key points are not available for this paper at this time.
• A novel two-stage spatiotemporal approach for reconstructing NDVI time series. • Integration of spatiotemporal filling and multistep temporal smoothing. • Interval-based comparison enables adaptation to variable planting schedules. • Robust reconstruction of large gaps caused by persistent cloud cover. • Improved prediction of crop phenology and enhanced agricultural monitoring. Vegetation indices derived from satellite observations, particularly the NDVI, are widely used for monitoring crop development and estimating yields. In tropical regions such as Indonesia, however, persistent cloud cover causes large gaps, undermining the reliability of NDVI time series and limiting their agricultural utility. This study proposes a two-stage spatiotemporal reconstruction framework to enhance NDVI time series under such conditions. The approach combines spatiotemporal filling with multistep temporal smoothing through an interval-based strategy that accommodates diverse planting schedules and performs effectively on shorter time series. Experiments with simulated datasets demonstrated that the method consistently outperformed benchmark techniques, achieving the lowest reconstruction errors compared to the Savitzky–Golay filter and cubic spline smoothing. Applied to Sentinel-2 NDVI from 2,460 rice field sites across three Indonesian regions, the method enhanced rice growth phase classification with a random forest model, improving precision by up to 19%, recall by 24%, and F1-score by 21%. These results demonstrate the framework’s robustness in handling large cloud-induced gaps and its potential for near-real-time agricultural monitoring.
Suseno et al. (Tue,) studied this question.