Mapping and monitoring fractional woody vegetation cover in the arid savannahs of northern Namibia using machine learning, LiDAR and SAR data.

Globally savannahs are experiencing an increase in woody vegetation. Although Namibia is a very arid country, it has experienced significant bush encroachment. which has decreased livestock productivity significantly. Therefore it is essential to monitor bush encroachment and widespread debushing activities. The aim of study was to develop a system which can accurately map and monitor fractional woody cover (FWC) in arid savannas, at national scales with SAR satellite data (ALOS PALSAR annual global mosaics, 2009, 2010, 2015, 2016), using machine learning models that were trained with diverse airborne LiDAR acquisitions (244032 ha, 2008-2014). Innovative methods were applied to process LiDAR data with highly variable point density to unbiased estimates of FWC. The average R² across all the years with the same ancillary data sets were 0.64, 0.75 and 0.76 at 25m, 50m and 75m respectively. The average absolute errors of FWC estimation at 25m, 50m and 75m were, 0.168, 0.14 and 0.135. All the models overestimated by an average of 10% FWC at the lower cover values (10-20% and 20-30%) and underestimated by an average of 15-20% FWC at high cover values (70-80% and 80-90%). The addition of ancillary variables, elevation, rainfall and both, increased the average R² to 0.75, 0.74, 0.79 respectively. Maps of FWC reflected spatial patterns related to rainfall gradients, vegetation type and structure. However, the inclusion of ancillary data sets resulted in anomalies in the FWC maps in areas with insufficient LiDAR training data coverage. The pair-wise annual FWC change maps reflected debushing activities, wildfires and subsequent increases in FWC. Increases in FWC of 0.2-0.3 occurred over 3-7% of the surface area of Dense Shrublands and Woodlands (2009-2016). However, there is no independent data to verify if this represents bush encroachment or error. The study provided essential insight into to potential and challenges of using this approach at operational scales.