Using remote sensing to detect forest degradation caused by small-scale farming in tropical Africa

Deforestation and forest degradation mainly caused by human activities around the world present a serious threat to all the life forms that are dependent on forests, and Nigeria is not an exception in this regard. Illegal farming activities are destroying forest reserves from inside and it is necessary to get an estimate of how much forest area has been converted to farmlands for better forest management and for examining its potential impact on climate change. The aim of the study is to detect crop clearings inside forested areas using Random Forest (RF) and Landsat 8 imagery alongside GIS ancillary data including vegetation indices NDVI, GRVI and topographic variables such as DEM, Slope, and Aspect for better classification results. In order to examine the effect of GIS ancillary data on classification accuracy, two scenarios were designed. In scenario 1 only spectral bands were used for classification, while in scenario 2, GIS ancillary data was also incorporated into the Random Forest (RF) model. A pixel-based supervised Random Forest classifier with appropriate training data was deployed on both scenarios. The results from scenario 2 proved to be more accurate with an overall accuracy of 95.5% and kappa statistics of 0.94, compared to scenario 1, which resulted in an overall accuracy of 92.5% and kappa value of 0.91. The study indicates the importance of GIS ancillary data for accurate classification of different crop type classes. The study also highlights the importance of near-infrared (NIR), shortwave infrared (SWIR) and digital elevation model (DEM) for vegetation analysis in the present study. The blue band also showed importance, especially in the case of classifying oil palm. The results show that the most dominant crops in the forested area are banana, cocoa, and cassava indicating the encroachment of illegal farming activities in the forest reserves. As only medium resolution imagery was available for the present study, in the future similar study with high-resolution imagery could further improve the results. Overall, the study shows that Random Forest along with GIS ancillary can be successfully used for detection of crop clearing in forested areas