Geelbek (Atractoscion aequidens) is an important fish species in South Africa's linefishery, a fishing sector defined by its fishing gear of rod and reel or handline. Distributed from Cape Point (34°21'S, 18°29'E) on the south west coast to Kosi Bay (26°51'S, 32°53'E) on the east coast, they are targeted throughout their range by the commercial linefishery, recreational anglers and small-scale fishers. The majority of geelbek are caught on the Agulhas Bank during austral summer. Due to current minimum size limits of 600 mm (total length, TL), well below the 50% size-at-maturity (950 mm TL), the majority of the catches are comprised of immature fish, making the stock vulnerable to growth overfishing. Adults (>5 years) migrate seasonally to spawn off KwaZulu-Natal and congregate in offshore shoals at night. These spawning aggregations allow fishermen to catch large numbers of fish, making geelbek also vulnerable to recruitment overfishing. This study aims to improve understanding of the fishery and population dynamics of geelbek to help inform natural resource management of the geelbek linefishery. A stock assessment of South African geelbek was undertaken to fulfil this aim. For this purpose, spatially and seasonally explicit equilibrium per-recruit and dynamic age-structured operating models were developed for geelbek to account for the dynamic in stock structure as a result of the intra-annual coastal migration and differences in the vulnerability of life history stages to varying fishing pressure along South Africa's coastline. These models were developed using statistical programming environment R. The models were parameterised and calibrated using length and catch data from the National Marine Linefish System (NMLS) and life history parameters sourced from peer-reviewed literature. Per-recruit analyses were performed to estimate current stock-specific fisheries mortality rates and the spawner biomass depletion. These estimates were used as input into the stochastic age-structured simulation model and calibrated using available commercial catch data (1987 - 2011). The stochastic operating model was used to predict the probability of stock recovery and long-term sustainability under eleven alternative fisheries management strategies. The current stock status was estimated at 9.9% (approximately 10%) of the pristine spawner biomass (SB₀) using per-recruit analysis. This was compared to the stock depletion estimates of ~5 and 7% SB₀ from prior assessments conducted in the late 1990s and 1980s. This study indicated that there was a ~50 to 100% increase in spawner biomass over the past twenty years. However, this level of stock depletion is still considered critically low with respect to the previous limit management goal of increasing spawner biomass depletion rates above 25% SB₀, the collapsed limit reference level, advised by Griffiths in 1997. Eleven management strategies were simulated, examining the effects of decreases in harvest rates, closed seasons and areas and changes in minimum size limits, initiated in 2020, and tested over the medium (ten years) to long (twenty years) term. The least efficient management strategy was continuing at the status quo, with a minimum size limit of 600 mm (TL), which predicted only 1% and 2% increase in SB by 2030 and 2040, respectively. The most efficient in terms of a rapid recovery was a full fishery closure 'control scenario' (moratorium), which predicted a recovery to the threshold reference level for sustainable fishing at 40% SB₀ by 2025, and approaching pristine levels by 2040. Increasing the minimum size limit to the size-at-50%-maturity, 950 mm TL, had the second highest recovery rate, reaching 25% SB₀ by 2027, and nearing 40% SB₀ by 2035, at which point its trajectory is asymptotic to 40% SB₀. Decreasing the harvest rate by 50% across all regions and seasons had the third highest recovery rate, reaching 25% SB₀ by 2035, but levelling off thereafter. All the other management strategies resulted in slight stock recoveries, but with all stock trajectories remaining below 14% SB₀ in the long term. Additionally, the impact of various strategies, such as increasing the minimum size limit to the size-at-50%-maturity, 950 mm TL, were unequal, with the east coast experiencing increasingly higher catches over time, whereas the catches for the south south west coast declined drastically throughout the year, and did not improve with time. Such unequal distribution of the impact of management intervention is a consequence of the migratory life history of the geelbek stock. These results provide comprehensive insights into the population dynamics and current impacts on the geelbek stock, suggesting that this species remain severely depleted at ~10% SB₀. Rebuilding the stock to sustainable levels would require serious management intervention.

An understanding of past and current weather conditions can aid in identifying trends and changes that have occurred in weather patterns. This is particularly important as certain weather conditions can have both a positive and a negative impact on various activities in any region. Together with an ever-changing climate it has become markedly noticeable that there is an upward trend in extreme weather conditions. The aim of this study is to evaluate the efficacy of univariate and multivariate extreme value theory models on climate data in the Western Cape province of South Africa. Data collected since 1965 from five weather stations viz. Cape Town International Airport, George Airport, Langebaanweg, Plettenberg Bay and Vredendal was modelled and analysed. In the multivariate analysis, multiple variables are modelled at a single location. Block maxima, threshold excess and point process approaches are used on the weather data, specifically on rainfall, wind speed and temperature maxima. For the block maxima approach, the data is grouped in n-length blocks and the maxima of each block form the dataset to be modelled. The threshold excess and point process approaches use a suitably chosen threshold whereby observations above the threshold are considered as extreme and therefore form the dataset used in the models. Under the threshold excess approach, only observations that exceed the threshold in all components are able to be modelled, whereas exceedances in one and all components simultaneously can be handled by the point process approach. While the probability of experiencing high levels of rainfall, wind speed and temperature individually and jointly are low, a few conclusions were drawn based on the comparison of the performance of the models. It was found that models under the block maxima approach did not perform well in modelling the weather variables at the five stations in both the univariate and multivariate case as many useful observations are discarded. The threshold excess and point process approaches performed better in modelling the weather extremes. Similar results are achieved between these two approaches in the univariate analysis and there is no outright distinction that favours one approach over the other. In terms of the multivariate case, which is restricted to two variables, the point process approach was able to provide estimates with increased accuracy as in many cases there are more extremes in one component individually than in both components. Specifically, the negative logistic and negative bilogistic models suitably capture the dependence structure between maximum wind speed versus maximum rain- fall and maximum wind speed versus maximum temperature at the five weather stations. The results from the point process models showed very weak dependence between wind speed and rainfall maxima as well as between wind speed and temperature maxima which may warrant the inclusion of additional variables into the analysis and even a spatial component which is not included in this study.

Includes bibliographical references.

Inlcudes bibliographical references.

Bibliography: leaves 92-94. / Modern portfolio theory was developed by Harry Markowitz more than forty years ago and is now considered to be an indispensable tool in portfolio construction. Sharpe introduced the index models as a simplification of the original Markowitz formulation, as this required fewer parameters to be estimated. One of the premises underlying the Sharpe Index model was that the returns of shares and market proxies followed a normal distribution and under this assumption, model parameters could best be estimated using Ordinary Least Squares (OLS) regression. More recent empirical evidence has however cast doubt over the assumption of normality and has suggested that market returns tend to be non-normal. If the assumption of normality is no longer upheld then OLS or Maximum Likelihood Estimates may no longer produce the best estimates of parameters and hence may compromise optimal portfolio constructions. In addition to the parameter estimation problems, at the time of the initial formulation of the portfolio models, managers were not allowed to participate in short sales (selling a share one does not own). This practice is now quite common in most developed markets and any portfolio formulation model needs to be generalised to allow for this. Empirical studies have shown that share returns and the returns of market proxies do not follow a normal distribution but rather seem to follow a skew distribution and has long tails. It is the aim of this thesis to explore robust regression procedures which should be an improvement of OLS when the data does not come from a normal distribution. The robust regression procedures will then be used to estimate the parameters used in the Sharpe Index models and examine whether or not they aid in the portfolio construction process. The robust procedures will be applied to the classical portfolio formulations as well as the generalised models.

In our research, we set out to model, understand and evaluate the business process at a start-up actuarial firm which employs Report Writers (RWers) who specialise in quantifying actuarial matters. We simulated various "what-if" and extreme scenarios relating to (1) the impact of qualitative variables (stress, morale and health) on RWer productivity, (2) hiring policies for RWers who have various skills sets, (3) the allocation of RWers to various roles within the process, (4) the impact that a high turnover of experienced RWers has on productivity, (5) the impact of introducing a flexible working arrangement (flexitime). This was done through business process modelling and simulations. The business process we modelled was governed by numerous potentially complex inter-relationships between variables and inter-relationships, which we believed could lead to potentially significant feedback loops. The models we built were then simulated over a period of 3 to 7 years to gain insights into the behavioural trends of the firm's business process over time when subject to "what-if" scenarios and policy implementations. The model simulations allowed us to get an understanding of the behaviour of processes over time, and the key variables and relationships involved in bringing about such behaviour as certain variables were subjected to changes in levels, as set out in our objectives. We made use of relevant literature, expert opinion, past data, questionnaires and cognitive mapping techniques to build simulation models. Guided by methodologies used in literature on modelling qualitative variables, bearing in mind the dangers in modelling for them, we modelled for the complex inter-relationships between qualitative and quantitative variables.

Includes bibliographical references. / The niche and niche breadth of Dwarf Chameleons, Bradypodion, was assessed in terms of broad scale climatic factors. A niche-based modelling method was then used to construct present and future habitat suitability maps for 2050 and 2080, for species in the genus. Additionally, the relationship between environment and morphology was analysed for a representative Bradypodion species, the Cape Dwarf Chameleon, B. pumilum. The niche and niche breadth of species and phylogenetic clades were analysed and described via an ordination technique, the outlying mean index (OMI) analysis. Maxent (v2.3), a presence only niche modelling method, proved very useful in the construction of present and future habitat suitability maps for species within the genus. For analysis of the correspondence between environment and morphology for B. pumilum, regression trees were employed. Rainfall seasonality and maximum annual temperatures were shown to strongly effect the current distributions of the genus Bradypodion at both the species and clade level. Additionally, as closely related species inhabited similar environmental niches, the genus was shown to display a degree of niche conservatism. All species and clades were shown to respond to climate change scenarios for 2050 and 2080, but responses were individualistic. However, most demonstrated range contractions under predicted climate scenarios. Additionally, a strong correlation (p < 0.05) was found between the morphology of B. pumilum and its environment. Environmental factors explained over 40% of the variation in snout-vent length and tail length, and over 20% of the variation in head width and head height, thus supporting the hypothesis of a correspondence between vegetation and morphology in Bradypodion. These results have provided an understanding of the relationship between Bradypodion and their environments that could provide valuable information regarding their ecology. Additionally, the habitat suitability maps for 2050 and 2080 could prove useful in the construction of any future conservation plans for these species. Furthermore, the results support the hypothesis of a correspondence between environmental factors and morphological traits within the genus Bradypodion.

The question of what determines the observed geographical distributions of animal species is one of great interest to natural scientists. In general terms, these distributions are known to be related to a number of environmental factors, but the nature of the relationships and the relative importance of the different factors remain poorly understood. This investigation sets out to obtain more insight into these questions by using statistical methods to explore the relationships between the observed geographical distributions of bird species in Swaziland and a number of environmental variables. In 1985 I set out to compile a bird atlas of Swaziland. Data were accumulated between 1985 and 1991 and the bird atlas submitted to a publisher in May 1993. The text of the bird atlas is included here as Appendix 2 because it provides the essential background to this dissertation. The introductory section of the atlas includes a description of the data gathering process. The analysis included in the bird atlas was aimed at a popular readership and was not intended to form part of an academic dissertation Chapter 1 describes the use of logistic regression to investigate the relationships between the bird distribution patterns and a set of environmental variables and to predict the distributions. Chapter 2 describes the results of applying the methods described in Chapter 1 for all of the bird species for which data were available. Because this paper was aimed at a different readership, some repetition of material contained in Chapter 1 was unavoidable. In Chapter 3, biplot techniques were used to obtain a graphical representation of the bird atlas data.

Reinforcement learning methods have become more efficient in recent years. In particular, the A3C (asynchronous advantage actor critic) approach demonstrated in Mnih et al. (2016) was able to halve the training time of the existing state-of-the-art approaches. However, these methods still require relatively large amounts of training resources due to the fundamental exploratory nature of reinforcement learning. Other machine learning approaches are able to improve the ability to train reinforcement learning agents by better processing input information to help map states to actions - convolutional and recurrent neural networks are helpful when input data is in image form that does not satisfy the Markov property. The specific required architecture of these convolutional and recurrent neural network models is not obvious given infinite possible permutations. There is very limited research giving clear guidance on neural network structure in a RL (reinforcement learning) context, and grid search-like approaches require too many resources and do not always find good optima. In order to address these, and other, challenges associated with traditional parameter optimization methods, an evolutionary approach similar to that taken by Dufourq and Bassett (2017) for image classification tasks was used to find the optimal model architecture when training an agent that learns to play Atari Pong. The approach found models that were able to train reinforcement learning agents faster, and with fewer parameters than that found by OpenAI’s model in Blackwell et al. (2018) - a superhuman level of performance.

Includes bibliographical references. / Every species of bird has at least one species of endoparasite or ectoparasite, if not several. Thus it cannot be taken for granted that a particular individual avian species is free from haemosporidian infections. Birds have the richest diversity of intracellular, single cell, protozoan parasites with complicated life cycles, changing between the vector as definitive host (blood-feeding arthropod) and bird as intermediate host. Avian haemosporidian parasites belong to the order Haemospororida, to which the families Haemoprotidae, Plasmodiidae, Garniidae and Leucocytozoidae belong, consisting of the genera Haemoproteus, Plasmodium, Fallisia and Leucocytozoon.