Sentinel-1 and Sentinel-2 responses to Hansen-reported forest loss in the Peruvian Amazon: Timing and spatial sensitivity

Satellite signals preceding recorded tropical forest loss are rarely tested against locations without loss. We compared the timing of Sentinel-1 co-polarised (VV) and cross-polarised (VH) backscatter and three Sentinel-2 indices (Normalized Burn Ratio, NBR; Normalized Difference Moisture Index, NDMI; Normalized Difference Red Edge index, NDRE) with Hansen Global Forest Change loss years in the Ucayali Region of the Peruvian Amazon (2019–2025). Hansen loss years served as an annual, satellite-derived reference rather than independent ground truth. Monthly time series at 819 locations with Hansen loss in 2021–2024 at the central 30 m pixel were compared with 2,588 non-loss locations using before-after analysis, superposed epoch analysis and single-change-point breakpoint detection with a pseudo-year null model. Sensitivity to spatial matching was tested with 50–250 m buffers, and canopy height from the Global Ecosystem Dynamics Investigation was examined exploratorily. All variables declined around the Hansen loss year relative to non-loss locations, most strongly for Sentinel-2 (Cliff’s δ = − 0.66 to − 0.76; Sentinel-1: −0.27 to − 0.30). Negative breakpoints occurred at 74–82% of loss locations for the Sentinel-2 indices, mostly in the loss year. Breakpoints preceding the loss year occurred at 9.5–19.0% of loss locations, exceeding the non-loss comparison only for NBR (+4.9 percentage points) and NDRE (+3.8) at the primary penalty. Apparent pre-loss rates increased with buffer size and depended on breakpoint selection, and exploratory GEDI analysis found no detectable association between canopy height and breakpoint timing. Pre-loss satellite signals should be tested against non-loss baselines before being interpreted as early warning.