Remote sensing environmental change in southern African savannahs : a case study of Namibia

Savannah biomes cover a fifth of Earth’s surface, harbour many of the world’s most iconic
species and most of its livestock and rangeland, while sustaining the livelihoods of an
important proportion of its human population. They provide essential ecosystem services and
functions, ranging from forest, grazing and water resources, to global climate regulation and
carbon sequestration. However, savannahs are highly sensitive to human activities and climate
change. Across sub-Saharan Africa, climatic shifts, destructive wars and increasing
anthropogenic disturbances in the form of agricultural intensification and urbanization, have
resulted in widespread land degradation and loss of ecosystem services. Yet, these threatened
ecosystems are some of the least studied or protected, and hence should be given high
conservation priority. Importantly, the scale of land degradation has not been fully explored,
thereby comprising an important knowledge gap in our understanding of ecosystem services
and processes, and effectively impeding conservation and management of these biodiversity
hotspots.
The primary drivers of land degradation include deforestation, triggered by the increasing
need for urban and arable land, and concurrently, shrub encroachment, a process in which the
herbaceous layer, a defining characteristic of savannahs, is replaced with hardy shrubs. These
processes have significant repercussions on ecosystem service provision, both locally and
globally, although the extents, drivers and impacts of either remain poorly quantified and
understood. Additionally, regional aridification anticipated under climate change, will lead to
important shifts in vegetation composition, amplified warming and reduced carbon
sequestration. Together with a growing human population, these processes are expected to
compound the risk of land degradation, thus further impacting key ecosystem services.
Namibia is undergoing significant environmental and socio-economic changes. The most
pervasive change processes affecting its savannahs are deforestation, degradation and shrub
encroachment. Yet, the extent and drivers of such change processes are not comprehensively
quantified, nor are the implications for rural livelihoods, sustainable land management, the
carbon cycle, climate and conservation fully explored. This is partly due to the complexities
of mapping vegetation changes with satellite data in savannahs. They are naturally spatially
and temporally variable owing to erratic rainfall, divergent plant functional type phenologies
and extensive anthropogenic impacts such as fire and grazing. Accordingly, this thesis aims to
(i) quantify distinct vegetation change processes across Namibia, and (ii) develop
methodologies to overcome limitations inherent in savannah mapping. Multi-sensor satellite
data spanning a range of spatial, temporal and spectral resolutions are integrated with field
datasets to achieve these aims, which are addressed in four journal articles.
Chapters 1 and 2 are introductory. Chapter 3 exploits the Landsat archive to track changes in
land cover classes over five decades throughout the Namibian Kalahari woodlands. The
approach addresses issues implicit in change detection of savannahs by capturing the distinct
phenological phases of woody vegetation and integrating multi-sensor, multi-source data.
Vegetation extent was found to have decreased due to urbanization and small-scale arable
farming. An assessment of the limitations leads to Chapter 4, which elaborates on the
previous chapter by quantifying aboveground biomass changes associated with deforestation
and shrub encroachment. The approach centres on fusing multiple satellite datasets, each
acting as a proxy for distinct vegetation properties, with calibration/validation data consisting
of concurrent field and LiDAR measurements. Biomass losses predominate, demonstrating
the contribution of land management to ecosystem carbon changes.
To identify whether biomass is declining across the country, Chapter 5 focuses on regional,
moderate spatial resolution time-series analyses. Phenological metrics extracted from MODIS
data are used to model observed fractional woody vegetation cover, a proxy for biomass.
Trends in modelled fractional woody cover are then evaluated in relation to the predominant
land-uses and precipitation. Negative trends slightly outweighed positive trends, with
decreases arising largely in protected, urban and communal areas. Since precipitation is a
fundamental control on vegetation, Chapter 6 investigates its relation to NDVI, by assessing
to what extent observed trends in vegetation cover are driven by rainfall. NDVI is modelled as
a function of precipitation, with residuals assumed to describe the fraction of NDVI not
explained by rainfall. Mean annual rainfall and rainfall amplitude show a positive trend,
although extensive “greening” is unrelated to rainfall. NDVI amplitude, used as a proxy for
vegetation density, indicates a widespread shift to a denser condition.
In Chapter 7, trend analysis is applied to a Landsat time-series to overcome spatial and
temporal limitations characteristic of the previous approaches. Results, together with those of
the previous chapters, are synthesized and a synopsis of the main findings is presented.
Vegetation loss is predominantly caused by demand for urban and arable land. Greening
trends are attributed to shrub encroachment and to a lesser extent conservation laws, agroforestry
and rangeland management, with precipitation presenting little influence. Despite
prevalent greening, degradation processes associated with shrub encroachment, including soil
erosion, are likely to be widespread. Deforestation occurs locally while shrub encroachment
occurs regionally. This thesis successfully integrates multi-source data to map, measure and
monitor distinct change processes across scales.