Assessing the potential of Sentinel-2 MSI sensor in detecting and mapping the spatial distribution of gullies in a communal grazing landscape

In this study, we evaluate the potential of the recently launched Sentinel-2 MSI multispectral sensor in mapping the spatial distribution of gullies in Okhombe valley, KwaZulu-Natal, South Africa. The study further investigates possible environmental factors that contribute towards gully initiation and development. Analysis was done using a robust machine learning algorithm: Support Vector Machine (SVM). Additionally, possible environmental factors (i.e. slope steepness, percent vegetation cover, and Topographic Wetness Index, Stream Power Index, and hillslope profile and planform curvature) that could have an influence on the extent of the gullies were also derived. An overall land cover classification accuracy of 94% was achieved, while the overall classification accuracy for gullies was 77%. It was observed that bands located in the visible regions of the EM, red edge and SWIR could optimally discriminate gullies from other land cover types. Additionally, the findings of the study indicate no significant difference between the environmental variables across the different estimated gully volumes and indicate a weak correlation between the variables of R= 0.190526. Overall, the findings of the study demonstrate the importance of using the free and readily available multispectral Sentinel-2 MSI data in conjunction with robust non-parametric Support Vector Machine classifier in mapping the spatial distribution of gullies.
Source: Physics and Chemistry of the Earth, Parts ABC - Category: Science Source Type: research