Detection of smallholder agroforestry management disturbances in Sri Lanka using Sentinel-2 time series, Landsat CCDC, and Climate Normalization

Detecting smallholder agroforestry management practices (thinning, pruning, and final harvesting) from satellite time series remains challenging in Sri Lanka’s Dry and Intermediate zone environments, where hydroclimatic variability can induce strong vegetation-index fluctuations. We developed a patch-based, multi-sensor framework to assess management detectability across 227 smallholder agroforestry patches in five districts using an “inside patch vs outside reference” (patch–buffer contrast) design. Sentinel-2 monthly contrast series (NDVI, EVI, NDMI, NBR) were used to quantify index sensitivity to disturbance and early recovery and to detect breakpoints within management-informed windows using BFASTmonitor. Landsat CCDC provided pixel-resolved change timing to characterize within-patch spatial heterogeneity and to evaluate consistency with patch-scale breakpoint dates. Climatic influence was screened using CHIRPS rainfall anomalies to distinguish breakpoints occurring under near-normal conditions from climate-sensitive candidates. Results show that management-related disturbances can be detected for a defensible subset of patches, but robust attribution benefits from multi-index agreement, pixel-level spatial diagnostics, and climate-aware interpretation. The proposed framework supports operational monitoring by prioritizing high-confidence patch candidates for targeted validation.