Woody savannah tree structural assessment in the greater Kruger National Park region, South Africa, using multi-seasonal polarimetric synthetic aperture radar (SAR) and optical data product approaches

Abstract

With a mean net primary productivity of 7.2 tC/ha/year and a minimum woody coverage ranging from 10 to 30%, savannahs account for approximately 40% of the global carbon store. The savannah woody component impacts the fire regime, biomass production, nutrient cycling, soil erosion, the water cycle and the anthropogenic services (e.g. fuelwood provision) vital for the rural populace. The structural parameters which make up this vital woody component can be directly measured using active remote sensing sensors such as LiDAR and SAR due to their responsiveness to vegetative structure and high canopy penetration ability. The aim of this work is to model regional scale woody tree structural attributes [specifically woody canopy volume (CVOL), woody volume (TWV) and woody cover (TOT COV)] for the management of South African savannas. This goal was achieved by testing multiple dataset scenarios consisting of multi-seasonal and fully polarized RADARSAT-2 C-band satellite SAR data, airborne LiDAR derived tree structural metrics and Rapid Eye optical products in an integrated modelling approach. According to results, SAR data acquired in the middle of the dry season generated the best models in comparison to other seasons but ideally a dataset spanning all seasons were preferable to obtain the best modelled results (CVOL ~ R2 = 0.71; RMSE = 18487.0, SEP = 18947.2; TWV ~ R2 = 0.61, RMSE = 3797.7, SEP = 3936.9; TOT COV ~ R2 = 0.66, RMSE = 8.78, SEP = 8.94). HV and HH Polarized intensities was found to contribute the most to the overall success of the models with the woody canopy volume metric being predicted with the highest accuracies.