Detecting intact forests in a highly fragmented landscape: Complementary insights from Landsat and GEDI

Forest alteration is a global problem that undermines biodiversity and ecosystem resilience, yet operational methods to distinguish intact from altered forests remain limited in highly fragmented regions. We propose an operational framework for detecting intact forests in highly fragmented landscapes using multi-source remote sensing and field data, with a case study of threatened Mediterranean-type deciduous forests in South America. We define intact forests based on long-term spectral-temporal stability. Using Landsat time series (1984–2023) and the LandTrendr algorithm within a Spectral-Temporal Identification Workflow (STIW), we identified forest fragments exhibiting near-zero cumulative canopy loss and recovery. STIW discriminated 3 of 22 fragments as intact, with no detectable canopy alteration since 1984. We then characterized forest structure using 25 m-diameter footprints from GEDI satellite LiDAR across the same 22 fragments (3.3–298 ha). We combined GEDI-based structural metrics with data from 99 forest plots, including 55 plots that overlapped GEDI footprints. We trained a supervised classification model on GEDI metrics using 66 footprints (41 altered and 25 intact) and evaluated its predictive performance at both the footprint and fragment scales. The predictive model achieved 60% accuracy at the GEDI footprint scale and 81.8% at the fragment scale, indicating higher reliability when aggregating information at the fragment level. Key predictors captured shrub-layer development and mid- to upper-canopy structure. Intact forests had greater basal area, higher species richness, and taller, larger trees than altered forests. GEDI metrics also showed that intact forests concentrate more biomass in the upper strata. The synergy between STIW and GEDI provides a robust and scalable framework for mapping intact and altered forest states in highly fragmented landscapes, supporting the identification of high-value reference ecosystems and enabling evidence-based conservation and restoration planning across multiple spatial scales.