| Summary: | While monitoring the effectiveness of forest conservation programs requires accurate data on (changes in) forest cover, many countries still lack the ability to map local forest inventory, especially in the drylands of Africa where forest areas are very sparsely covered. In this paper, we present a high resolution tree cover estimation of twelve gazetted forests in Burkina Faso using Random Forest-based supervised classification and Sentinel-2 satellite imagery sensed between March and April 2016. The methodology relies on ground truth sample points labeled manually over 10-m resolution images displaying a composite of near infrared (NIR), red and green bands extracted from Sentinel-2 multi-spectral satellite data to estimate tree cover with an average balanced accuracy rate of 80 percent. The output is a collection of rasters with binary values representing the combination of 10, and down-sampled 20 and 60-m bands indicating an estimate of the existence of trees or lack thereof, usable as a baseline for deforestation monitoring. Keywords: Burkina Faso, Image classification, Forest cover, Sentinel-2, Google Earth engine, Sahel
|
|---|
