Mapping woody canopies using Sentinel-1 and Sentine-2 data in savannas

Woody vegetation is a central component of savanna ecosystems, acting as a carbon sink and providing ecosystem services for local livelihoods. Accurate monitoring of woody vegetation in savannas is therefore desirable, yet the spatial resolution of satellite data available for monitoring large areas cannot directly capture the scattered nature of trees and shrubs. All studies at regional scale thus estimate the fractional cover of woody plants for a given area, generally characterized by high uncertainty and without information on the spatial distribution of trees and shrubs. With the launch of the Sentinel satellite systems, the spatial resolution of images approaches the size of medium/large tree crowns, providing the opportunity to map the presence/absence of tree canopies, rather than the fraction of woody cover or forested areas. Here, we used a support vector machine (SVM) to classify the presence/absence of woody canopies from Sentinel-1 and -2 data at a 10-m spatial resolution for the entire African Sahel. Training samples for the SVM classifier were collected from Google Earth very high resolution (VHR) images and satellite data were processed in Google Earth Engine. Accuracy assessment was performed based on reference data set derived from independent VHR DigitalGlobe images, showing an overall accuracy of 93% when combining the use of Sentinel-1 and -2 data. The combined use of data from the Sentinels proved to perform significantly better (p<0.05) than the single use of both Sentinel-1 and -2 data. A comparison with existing tree cover maps showed noticeable differences, reflecting the need of new woody cover products adapted to the nature of savanna ecosystems. The Sentinel woody canopy map was able to reproduce the general pattern of scattered woody canopies, but generally overestimated the woody coverage (11.37± 26.13% when aggregating to 250 m resolution) due to the 10×10-m spatial resolution. The high spatial and temporal resolution of cloud-based Sentinel-1 and -2 analysis is a step towards direct woody canopy mapping showing advantages over conventional fractional cover estimates. Ultimately, the direct assessment of woody canopy areas will allow the mapping of temporal dynamics of woody canopies in future studies as time-series expands to multiple years.