Quantifying a decade of Savanna vegetation structure change with L-band SAR.

Global savannas are the third largest carbon sink with large human populations being highly dependent on their ecosystem services. However, savannas are changing rapidly due to climate change, fire, animal management, and intense fuelwood utilization. In southern Africa, large trees (>5m in height) are under threat while shrub cover (<3m) is increasing. Although the mapping of vegetation cover and biomass with L-band SAR has recently been demonstrated in savannas, there is a lack of studies on the ability to quantify change, due largely to a lack of validation data from repeat airborne LiDAR (ALS). The objective was to test the ability of ALOS PALSAR 1&2, dual-pol data to quantify woody cover and volume change in savannas over two-, eight- and ten-year periods through comparison to ALS, in the Lowveld of South Africa. For each epoch (2008, 2010, 2018), multiple SAR images (HH, HV) were processed to Gamma0 at 15m resolution with multi-temporal speckle filtering. ALS data were processed to canopy cover (%) and volume (height × canopy cover) and then compared to 5×5 aggregated (i.e. 75m) SAR backscatter pixels. For individual years, the R2 of the linear and logarithmic relationships of backscatter vs. cover, and volume ranged from 0.60 to 0.75. The R2 using HH was only 5-8% lower than that of the HV backscatter. A modified water cloud model (WC) was also tested and provided the best results with R2 = 0.67-0.76. The linear relationships between pairwise backscatter change (dB difference) and canopy cover or volume change (2008-2010, 2010-2018, 2008-2018) had low R2 of 0.08-0.31, with a positive correlation between backscatter change and cover change. The WC model was used to predict canopy cover and volume for individual years and the differences were calculated. After binning canopy cover changes at 10% increments, a T-test revealed that these bins showed statistically significant differences in backscatter changes, e.g. -10 to -20% change differed from -10 to 0% and 0 to +10%, and +20 to +30% cover change. Canopy cover change maps can therefore be produced that indicate positive and negative change at 10% increments. Results for volume change were similar. Additional detailed analysis indicated that the loss of small stands of large trees (>5m) could not be detected with the SAR data, while shrub cover increases could.