Modelling spatial variations in wood volume and forest carbon stocks in dry forests of Southern Africa using remotely sensed data

Abstract

The estimation of forest carbon is important to generate knowledge on the extent to which forests contribute to climate change mitigation. Several studies on the estimation of forest carbon stocks have mainly focused on tropical rainforests. However, only a few studies have focused on dry forests tropical savanna, yet they constitute about 33% of the terrestrial biomes. Moreover, most work on the estimation of forest carbon stocks has traditionally relied on fieldwork which covers only small spatial extents. Work that has global proportions needs a method of estimating forest carbon stocks that covers large spatial extents. To this end, remote sensing provides an opportunity to estimate dendrometric characteristics of forests and woodlands such as wood volume and forest carbon stocks over large spatial extents. In this thesis, we predicted wood volume and forest carbon stocks as a function of remotely sensed vegetation indices. Specifically, we tested whether high spatial resolution satellite imagery (WorldView-2 and GeoEye-1) improves accuracy in wood volume and forest carbon stocks estimation based on two study sites in dry forests in Zimbabwe with contrasting annual rainfall amounts. Firstly, we compared the predictive ability of vegetation indices (i.e., Simple Ratio (SR), Soil Adjusted Vegetation Index (SAVI) and Normalized Difference Vegetation Index (NDVI) derived the high spatial resolution sensors (GeoEye-1 and WorldView-2) for Mukuvisi and Malipati respectively with the indices derived from the medium resolution sensor, i.e., Landsat 5 TM (Thematic Mapper) in predicting wood volume. Secondly, we mapped the spatial variations in wood volume in the two study sites using best predictive model relating wood volume to remotely sensed vegetation indices. Thirdly, we tested whether the inclusion of the red edge band as an explanatory variable to vegetation indices derived from WorldView-2 can improve the estimation of forest carbon stocks in dry forests of Malipati Safari Area. Finally, we mapped the spatial variations in forest carbon stocks in Malipati using best predictive model relating forest carbon stocks to vegetation indices and the red edge band. Our results showed that vegetation indices derived from WorldView-2 and GeoEye-1 significantly (p< 0.05) predicted wood volume better Landsat 5 TM derived vegetation indices Our results also showed that vegetation indices alone as an explanatory variable significantly (p<0.05) predicted forest carbon stocks with R2 ranging between 45% and 63% and RMSE ranging from 10.3% and 12.9%. However, when the reflectance in the red edge band was included the explained variance increased to between 68% and 70% with the RMSE ranging between 9.56% and 10.1%. A combination of SR and reflectance produced the best predictor of forest carbon stocks. We concluded that vegetation indices derived from high spatial resolution improves accuracy in estimating wood volume and forest carbon stocks and thus can be successfully used to map forest carbon stocks in dry forests.