Multi-Scale Quantification of Woody Biomass in Heterogeneous Landscapes: Leveraging Traditional Field Sampling, Spectral Unmixing, and Allometric Modeling in Kalahari Savanna Ecosystems

Savannas are complex arid or semi-arid biomes, covering approximately 20 % of the terrestrial surface of the earth and providing substantial ecosystem services, such as carbon storage, firewood, rangeland and protection against soil erosion. They are systems best described as prone to constant change in productivity, diversity and abundance of species and structure. Due to this complexity, savannas are extremely heterogeneous ecosystems where, despite past and current research, the ongoing processes between the abiotic and biotic environment remain poorly understood. This knowledge gap and its importance are only amplified by increasing anthropogenic influences, increasing human population, and increasingly varying climate change.

This dissertation contributes to the ongoing scientific debate regarding the role of savannas as a carbon sink or source. A key factor in understanding carbon cycling is the role of woody vegetation. Multiple drivers of shrub encroachment were assessed as to their impacts on diversity and abundance of woody vegetation across the rainfall gradient of the Kalahari Transect, located in western Botswana. This study underscores the importance to study the combination of multiple drivers and their effects on species with similar functional and structural traits. Further, allometric relationships were developed to estimate biomass using different morphological parameters, therefore providing a useful tool to estimate biomass on multiple scales. These relationships are useful for informing decisions and management ranging from the plot (management) scale to the local scale (where active remote sensing systems may be deployed). Regional and global carbon monitoring efforts are in the most need of properly calibrated, reliable information derived from regional-scale satellite imagery but to date there remains skepticism on the reliability of those data.

To overcome the challenges associated with the quantification of biomass across large tracts of land, this work took on the challenge to extract biomass using a two-dimensional passive remote sensing system. Extensive field-derived measurements, including fractional cover of green vegetation (GV), non-photosynthetic vegetation (NPV), soil, vegetation structure and species for all woody plants, were collected in both wet and dry seasons over 15 sites along a 950km transect. These in situ data were used to develop, test, and validate a new method, the Spectral Line Point Intercept Transect – SLPT, to obtain spectral information efficiently across long distances. Multiple validation methods were leveraged to test the performance of various spectral unmixing techniques in heterogeneous savanna environments to derive fractional cover of GV, NPV, soil and shade at structurally different sites and in different phenological conditions (wet vs. dry season). The fractional cover of shade derived using Multiple Endmember Spectral Mixture Analysis (MESMA) and the Moderate Imaging Spectroradiometer (MODIS) Enhanced Vegetation Index (EVI) products were used to derive a relationship (φ) that identifies fractional cover of woody vegetation for a given MODIS pixel. By combining the MESMA, φ, and the allometric relationships described above, this research was able to develop a modelling approach that successfully quantifies biomass and woody vegetation cover across the Kalahari region.