Estimating biomass in the mountain regions of Bwindi Impenetrable National Park, Uganda using radar and optical remote sensing

Abstract

Field measured estimates of aboveground biomass (AGB) for 15 transects in Bwindi Impenetrable National Park (BINP), Uganda were used to generate a number of prediction models for estimating aboveground biomass (AGB) over the full extent of BINP. AGB estimates were extrapolated from the field data using dual-polarization radar satellite data alone, optical satellite data, and a combination of both. The effectiveness of the dual-polarization radar remote sensing data alone was limited due to the difficulties of geocoding and terrain correction in this mountainous region, producing problems with layover and shadowing. The optical-only method demonstrated that perhaps thermal bands may be more sensitive to biomass in tropical forests than visible bands. The radar and optical combined method, generated using the non-parametric algorithm Random Forest (RF) in R, provided the lowest RMSE error (~120 Mg ha-1). The analysis also demonstrated that a number of radar backscatter variables had greater utility for generating a predictive model of biomass than many optical bands in this mountainous region. The combined optical and radar remote sensing model was used to produce a final AGB map over the full 331 km2 extent of BINP; AGB in BINP was estimated at 89.1 million Mg ± 3.9 million Mg, with a mean carbon density of 44.5 million Mg C ± 60 Mg C ha-1.