Integrating RADAR and optical imagery improve the modelling of carbon stocks in a mopane-dominated African savannah dry forest

Abstract

This study examined the integration of two satellite data sets, that is Landsat 7 ETM+ and ALOS PALSAR (Advanced Land Observing Satellite Phased Array type L-band Synthetic Aperture RADAR) in estimating carbon stocks in mopane woodlands of north-western Zimbabwe. Mopane woodlands cover large spatial extents and provide ecosystem benefits to the rural economies and grazing resources for both livestock and wildlife. In this study, artificial neural networks (ANN) were used to estimate carbon stocks based on spectral metrics derived from Landsat 7 ETM+ and ALOS PALSAR. To determine the utility of the two satellite-derived metrics, a two-pronged modelling framework was adopted. Firstly, we used spectral bands and vegetation indices from the two satellite data sets independently, and subsequently, we integrated the metrics from the two satellite data sets into the final model. Results showed that the ALOS PALSAR (R2 = 0.75 and nRMSE = 0.16) and Landsat ETM+ (R2 = 0.78 and nRMSE = 0.14) derived spectral bands and vegetation indices comparatively yielded accurate estimations of carbon stocks. Integrating spectral bands and vegetation indices from both sensors significantly improved the estimation of carbon stocks (R2 = 0.84 and nRMSE = 0.12). These findings underscore the importance of integrating satellite data in vegetation biophysical assessment and monitoring in poorly documented ecosystems such as the mopane woodlands.