Estimating forest carbon stocks in tropical dry forests of Zimbabwe: exploring the performance of high and medium spatial-resolution multispectral sensors

Abstract

Estimation and mapping of forest dendrometric characteristics such as carbon stocks using remote sensing techniques is fundamental for improved understanding of the role of forests in the carbon cycle and climate change. In this study, we tested whether and to what extent spectral transforms, i.e. vegetation indices derived from new generation high-spatial-resolution multispectral sensors (WorldView-2 and GeoEye-1), estimate carbon stocks when compared with medium-spatial-resolution broadband sensors (Landsat 5 Thematic Mapper) based on two savanna woodland types in Zimbabwe. Subsequently, the best ordinary least squares regression model relating carbon stocks and vegetation indices was applied in mapping carbon stocks in two study sites. Based on k-fold cross-validated regression models, vegetation indices computed from new generation high-spatial-resolution multispectral sensors yielded high R 2 values ranging between 82% and 73% (RMSEcv: 5.55–6.87%) for Mukuvisi and 62–73% (RMSEcv: 11.5–13.6) for Malipati compared with Landsat 5 Thematic Mapper derived vegetation indices, which yielded R 2 values between 47% and 49% (RMSEcv: 9.6–10.1%) for Mukuvisi and 22–41% (RMSEcv: 11.5–19.1%) for Malipati. The findings demonstrated that medium-spatial-resolution sensors are less sensitive to attributes of sparsely distributed trees, especially in savanna woodlands, where the size of trees are often less that the spatial resolution of the mediumspatial-resolution sensors. These findings emphasise the importance of new generation high-spatial-resolution multispectral sensors in estimating forest structural attributes, such as carbon stocks in open woodlands.