Earth Observation models help management of tropical dry savannah forests in the Okavango-Zambezi transfrontier conservation zone (KAZA) region of Southern Africa.

The Kavango Zambezi Transfrontier Conservation Area (KAZA) is the World’s largest conservation area with an enclosed area the size of Sweden (519,912 km²), and is characterized by savannah forest, woodland and protected lands. KAZA is situated at the heart of the area most vulnerable to climate change in Africa, and forest loss and degradation are major concerns which directly impact wildlife species distributions and a growing human populations. In particular, detailed knowledge of current vegetation density change and forest area estimates throughout the conservation area is sorely missing, which hampers all efforts to mitigate the threats against KAZA and its unique ecosystems. A combination of remotely sensed data and plot-based estimates can provide forest area estimates and above ground biomass (AGB). Previous AGB mapping efforts in Africa focused on tropical humid forests, with little attention on tropical and subtropical savannah forest. The aim of the current study was to establish a link between remote sensing spectral data derived from Landsat 8 and ground characteristics to improve precision of AGB and forest area estimates in savannah forest. We used 114 sample plots distributed on 6 clusters collected over the 2019 winter growing season in Chobe National Park of Botswana and Landsat 8 spectral variables.

Restricting analysis to sampling dates, before the onset of fire burning and leaf yellowing resulted in increased estimation accuracy. We found a linear relationship between above ground biomass and Landsat 8 derived spectral variables (p < 0.001 and p < 0.005). The normalized difference vegetation index (NDVI) and Green-Red Difference Index (GRVI) exhibited a strong correlation with AGB than other indices (R2 = 0.76) and (R² = 0.67), respectively. An improvement in the correlation is seen when AGB (t/ha) and variables relationship is performed in the woodland/forest cluster (n=74), excluding the shrubland and grassland. The AGB of savannah forest in the study area based on spatial analysis was 111.6 Mg/ha. A root-mean-square error (RMSE) value from predicted and observed AGB was 25.6 Mg/ha. The high total AGB value from savannah forest in the study area highlight the importance of the savannah-forest mosaic as a biomass storage and carbon pool. Overall, spectral variables and indices, particularly the NDVI and GRVI and Landsat 8 band 5 (NIR), would be useful predictors and suitable auxiliary information of AGB in the savannah forest. The results of this study show that taking into account stratification/clustering of different vegetation types and senescence period can greatly increase the accuracy of AGB estimation. This results will allow us to build new models to quantify savannah forest change and long-term trends using Landsat time series from 1980 to 2019. Time series modelling will help inform how changing climate threaten the biodiversity of the KAZA region and be able to respond to these threats with appropriate, evidence-based strategies and measures.