A spatio-temporal analysis of baboon damage using Sentinel-2 imagery and Extreme Gradient Boosting

Abstract

The use of remote sensing for forest health monitoring has increased in popularity over the years, improving quality, spatial and spectral resolutions. However, revisit times of satellites is too slow for real-time detection. The need exists for high resolution monitoring, to quantify biotic damage, which is more difficult to detect due to the diversity of severity. Sentinel-2 MSI data, with 10 - 60 m spatial resolution and 443 - 2190 nm spectral range was used. Through strategical bands and derived vegetation indices, an accuracy of 89,34% with a Kappa of 0.85 using XGboost, was achieved. The study found: (1) Sentinel-2 identified baboon damage with high accuracy, with field measured data (2) different damage severities could be distinguished. SAVI and SIPI were most influential variables, due to good discrimination of structural variations in the canopy. The method provides an accurate and repeatable framework for rapid damage mapping at a sub-compartment scale.