Guinea-Bissau is one of the poorest countries in the world with a large proportion of its population living in rural areas. While industry is limited, over 70% of the territory is covered by forests, which can potentially be used to attract investment through forest-based projects that promote reductions in carbon emissions and sustainable management. These can be leveraged by producing accurate maps of forest aboveground biomass (AGB) at national level and by developing cost-effective mapping methods that allow reliable future updating for management and engagement in international mechanisms such as the United Nations (UN) Collaborative Program on Reducing Emissions from Deforestation and Forest Degradation (REDD) in developing countries.Using data from Japan's Advanced Land Observing Satellite (ALOS) Phased Array L-band Synthetic Aperture Radar (PALSAR), this study compared a semi-empirical and machine learning algorithm, with the latter based on bagging stochastic gradient boosting (BagSGB), for retrieving the AGB of woody vegetation thereby supporting estimation of national carbon stocks. AGB was estimated by using measurements of tree size collected from112 forest plots during two field campaigns (2007 and 2008) as input to published allometric equations.The BagSGB outperformed the semi-empirical algorithm, resulting in a coefficient of correlation (R) between observed and cross-validation predicted forest AGB values of 0.95 and in a root mean square error (RMSE) of 26.62Mgha−1. Furthermore, the BagSGB model produced also a measure of forest AGB prediction variability (coefficient of variation) on a pixel-by-pixel basis, with values ranging from 7 to 250% (mean=42%). An estimate of total forest AGB carbon stock of 96.93Mt C was obtained in this study for Guinea-Bissau, with a mean forest AGB value of 65.17Mgha−1. Although the mean error associated with this forest AGB map is still undesirably high, several issues were addressed. The heterogeneity of forest structural types, presence of palm trees, and dimension and type of field plots were identified as potential source of uncertainty that must be tackled in future studies. This study represents a step forward regarding the information currently available for Guinea-Bissau.
Read full abstract