Estimation of forest cover change using Sentinel-2 multi-spectral imagery in Georgia (the Caucasus)

Our objective was to use Sentinel-2A multispectral data in order to cost-effectively detect change in forest cover in Georgia (the Caucasus). Generalized additive models (GAMs) were used to fit forest cover measures to Sentinel-2A spectral band values modified using different topographic correction methods. Canopy closure (calculated from upward-looking fisheye photographs taken beneath forest canopy) was the best forest cover measure accounted for by the Sentinel-2 spectral data that were topographically corrected using the Minnaert Correction (R2 = 0.882). Spectral bands best explaining canopy closure were Band 3 (Green), Band 8 (NIR) and Band 12 (SWIR). Our model is able to reasonably detect spatial and temporal changes in canopy closure, even in highly rugged terrain and diverse vegetation cover, and it has potential to be improved to the extent that it can be applied by managers of natural resources. Based on free open source applications in combination with cheap gadgets our approach might play an important role in monitoring the forests of countries with low economic indicators.