Spatio-Temporal Assessment of Forest Degradation Using Multi-Index Landsat Time Series for Restoration Prioritization in Bhutan
Description
Forest degradation is difficult to detect because gradual changes in forest condition can occur without complete canopy loss. This study assessed Spatio-temporal forest degradation and restoration priorities in Bhutan from 2005 to 2024 using Landsat-derived NDVI, NDMI, NBR, and NDFI. Pixel-wise temporal trends were evaluated using ordinary least-squares regression (p < 0.05), and the four index-specific classifications were integrated using a ≥3 of 4 (75%) consensus criterion. Of 18,356.70 km² of analyzed persistent forest, 1.99% showed strong multi-index evidence of degradation, 64.03% were classified as stable/no significant trend, 12.97% improved, and 21.01% showed no consensus. Validation using 901 stratified random samples yielded an area-adjusted overall accuracy of 52.31%, indicating a trade-off between conservative multi-index agreement and sensitivity to subtle degradation. Restoration priority within mapped degraded forests was assessed by integrating erosion susceptibility, fire recurrence, human pressure, conservation importance, and forest-type restoration sensitivity using Simple Additive Weighting, with the resulting index classified into low, moderate, high, and critical priorities. The framework provides a spatially explicit approach for linking long-term forest degradation detection with targeted restoration prioritization in Bhutan, while explicitly accounting for uncertainty in forest-condition trajectories.
Files
Steps to reproduce
Annual Landsat 7 ETM+ (2005 - 2012) and Landsat 8 OLI (2013 - 2024) Collection 2 Level-2 surface-reflectance imagery was processed in Google Earth Engine using a fixed post-monsoon observation period (15 October - 30 November). Clouds, cloud shadows, cirrus, snow, and radiometrically saturated pixels were masked using QA_PIXEL and QA_RADSAT, and annual median composites were generated. Terrain illumination effects were reduced using SRTM 30-m DEM-based C-correction. Landsat 7 spectral-index observations were harmonized to the Landsat 8 reference scale before temporal analysis. NDVI, NDMI, NBR, and NDFI were calculated annually for the persistent-forest domain. Persistent forest was defined as locations classified as forest in the 1995, 2010, 2016, and 2022 national forest-cover reference datasets. Pixels were retained for temporal analysis when at least 15 annual observations were available, including at least three observations during both 2005-2009 and 2020-2024. Pixel-wise ordinary least-squares regression was applied independently to each spectral index. Significant negative trends were defined as slope < 0 and p < 0.05; significant positive trends as slope > 0 and p < 0.05; and no significant trend as p ≥ 0.05. The final forest-condition classification required agreement from at least three of the four indices (≥75%). Restoration priority was subsequently calculated only within mapped degraded forest using erosion susceptibility, fire recurrence, human pressure, conservation importance, and forest-type restoration sensitivity. Criteria were standardized to 0-1 and integrated using Simple Additive Weighting with baseline weights of 0.30, 0.20, 0.20, 0.15, and 0.15, respectively. Weight robustness was evaluated using an equal-weight scenario and ±20% one at a time perturbations. Analytical scripts and associated output tables included with this deposit can be used to reproduce the principal results.
Institutions
- Korea UniversitySeoul, Seoul