Modelling tree growth rates across Southern African woodlands: a hierarchical mixed-effects approach
Item statusRestricted Access
Embargo end date10/03/2023
Ecosystem products such as fuel, timber, food, and water, coupled with services such as regulation of the climate, mitigation of floods, and maintenance of soil fertility, provided by southern African woodlands (SAWs) play a central role in the livelihoods of millions of both rural and urban dwellers. However, the ability of these woodlands to meet future demand is threatened by their rapid deforestation and degradation. This has led to a proliferation of campaigns to champion their sustainability especially given uncertainties inherent in global change. Key to their sustainability is disentangling how fast trees grow across the region. Growth rate studies can aid policy formulations that herald management of existing woodlands. For example, if growth rates are known for a given stand, it can enable the prediction of maximum sustainable yield of fuel or timber, and how that changes with edaphoclimatic fluctuations. Although investigating growth rates is a continuing concern within sustainability science, studies focusing on the broader southern African region are still limited, leading to major uncertainties on the determinants of tree growth rates across SAWs. Understanding tree growth rates require scientists to unravel the biophysical interactions that influence growth. This information can then be used to develop spatial-temporal predictive models. In this thesis, I combine data from permanent sample plots (PSPs), remote sensing products, and modelling techniques to contribute towards deepening our current understanding of tree growth rates and sustainable harvesting rates across SAWs. The PSP data I use is the subcontinent's largest growth dataset and spans 27 years (1991 – 2018) with 38,276 individual trees on 223 permanent sample plots (PSPs) in 8 countries across southern Africa. Results show a mean (±SEM) diameter increment of 1.62 ± 0.3 mm yr-1 across the region, with a substantial site-to-site variation. The hypothesised drivers of growth rates variation such as soils and climate had little effect at the regional level, although plot-level tree basal area (a proxy for tree-tree competition) consistently reduced growth rates. Examining the determinants of growth for each species, however, revealed strong and often contrasting effects of soils, climate, tree size, and fire. A continuum of responses across species emerges, with contrasting patterns between miombo woodland dominants such as Brachystegia and open savanna species. In the faster-growing miombo dominant group, growth increased with stem size, on poor soils, and was less affected by reduced water availability and seasonality. The open savanna group exhibited the opposite trends, growing relatively faster on good soils and in wetter conditions. These findings challenge conventional understanding of the determinates of tree growth in savannas, and suggest distinct controls in different types of savanna, and thus distinct responses to global change. The observed widespread effect of tree-tree competition may play an important, but hitherto neglected, role in determining community composition and sustaining tree-grass coexistence. It is also likely to moderate the current increase of trees in savanna systems, suggesting the globally important C sink in African savannas is unlikely to be sustained. When I assessed basal area increment (BAI) with the aim of estimating growth rates across unmeasured areas of SAWs, I found BAIs ranging between 0.07 and 0.7 with an estimated regional mean of 0.35 m2/ha/year. I further found that BAI was positively related to rainy season length (RSL) but was negatively affected by climatic water deficit (CWD), with some other factors also playing a more minor role. Considering that this study covered a wide environmental gradient, the ability of the selected model to predict growth on an independent dataset with a correlation coefficient (R) of 0.8, a root mean square error (RMSE) of 0.01 m2/ha/year, a RMSE-observations standard deviation ratio (RSR) of 0.7, a percent bias (PBias) of 7.8%, and a Nash-Sutcliffe Efficiency (NSE) of 0.5 between observations and predictions suggests that the model provides practical usability in predicting BAI for sustainable woodland management across unmonitored areas of SAWs. Model predictions across unmonitored areas of SAWs were found to range between 0.07 and 0.7 with an estimated regional mean of 0.35 m2/ha/year and were consistent with previous literature. By extension, the model has demonstrated the potential to contribute towards deepening our understanding of woodland dynamics across unmonitored areas of SAWs. It is notable however that the model failed to predict with a 1:1 precision therefore, caution needs to be applied in its use. In particular, its skill can be improved by the use of higher resolution ground-based PVs and further plots. The predicted BAI across unmeasured areas can be used to support the analyses of woodland structures, estimation of carbon stocks, and overall economic assessments of woodland policies for sustainable management options. In terms of sustainable harvesting rates, I found a mean annual increment (G) of 0.64 ±0.1 m3/ha across the region. I then adopted an annual allowable cut (AAC) of 80% of the G to ensure sustainability in both ecological and economic terms. I found a mean AAC of 0.51 ±0.1 m3/ha with a range of 0 – 7.4 m3/ha/year across the region. It is notable however that the distribution of AACs is very skewed, with most in the dataset well below 1 m3/ha/year, and a few outlier plots (4 to be specific) above 3 m3/ha/year. At this AAC level, recovery scenarios were found to surpass the volume at the beginning of the inventory within Year 1. At the individual stem level of the 1,025 large trees studied, results show a mean annual diameter growth rate of 1.5 ±0.3 mm translating into a harvest cycle of 233 ±46 years when a widely accepted harvest size of 35 cm is used thus implying that on a given stand with 233 trees, one tree can be harvested every year. In this study, the set AAC was found to suggest that there is potential for any PSP to recover and become resilient as long as strict adherence to the AAC is followed. I, therefore, conclude that adherence to AAC is a pivotal part of good woodland management because it is the primary safeguard against under or over cutting. This is especially true because by harvesting only large trees as prescribed by the AAC, trees in smaller diameter classes are allowed to graduate to higher classes. Further, the process supports the regeneration of large trees which will in the long term ensure woodland sustainability.