Temporal standardization of Sentinel-1 time series for annual and continuous tropical forest loss detection

Moist tropical forest extent monitoring and mapping are at the center of ecosystem preservation in the 21st century. Satellite remote sensing has proven to be essential for assessing forest loss in wide and remote areas at a continental scale. Nevertheless, yearly assessments using optical images tend to face challenges in African moist tropical forests, where the semi-persistent cloud coverage prevents short-term and small-scale events from being correctly detected. The Sentinel-1 (S1) Synthetic Aperture Radar (SAR) C-band sensor provides relevant data for addressing the challenge of continuous forest monitoring, due to its all-weather imaging capability, which ensures 12-day revisit data acquisition. This study introduces a method to temporally standardize Sentinel-1A (S1A) time series to provide a dataset dedicated to change detection–here applied to moist tropical forest loss monitoring in the Congo Basin. Temporal standardization converts the geophysical measurement of ground range detected (GRD) data into pixel-specific z-values or statistical deviations. This standardization prevents the detection of stable low values as forest loss and focuses detection on the decrease of signal backscattering from a historic trend. Annual and continuous detection methods are jointly developed from the standardized dataset and subsequently compared. This study also defines a comprehensive performance metric ensemble, which is designed to assess rare class detection performance in many aspects and guide the z-value thresholding selection across a variety of S1 data compositing. The best performing annual index is produced by combining the 10th-percentile VV and VH composites from an annual times series. The optimal threshold allows the detection of 96.5% of annual forest loss area, with an annual false positive rate in stable forest extent (FPRstable) of 0.04%. For continuous detection, three successive standardized observations are processed in a sliding temporal window, serving as a consolidation period of 24 days–half of the operational SAR-based current systems. This method detects 98.3% of forest loss area, with an annual FPRstable of 0.05%. The study also observes and corrects the geometric distortion inherent in SAR-based tropical forest loss detection systems. These findings and related statistical methods highlight the benefit of temporal standardization, providing C-SAR analysis-ready datasets for operational moist tropical forest monitoring.