published_or_final_version / abstract / Computer Science / Doctoral / Doctor of Philosophy

A dissertation submitted in fulfilment of the academic requirements for the degree of Master of Science in the School of Animal, Plant and Environmental Sciences, University of the Witwatersrand, Johannesburg. June 2016 / The African baobab (Adansonia digitata) is a charismatic and iconic tree. This keystone species has highly specific moisture requirements for recruitment, a very slow life history, low dispersal capabilities, and is exposed to heavy, often destructive utilization by baboons, elephants and humans. These characteristics result in the reduced ability for baobabs to accommodate and respond to changing environmental conditions. During periods of unfavourable conditions, baobabs rely on the persistence of established individuals for the survival of the population. The Kruger National Park (KNP), South Africa falls within the southern-most distribution of the baobab. A perceived lack of recruitment together with high numbers of dead large baobabs recorded in aerial surveys prompted concern over local population decline in the early 1990s. Long-term monitoring is necessary in order to quantify changes occurring within the KNP baobab population. For this reason Dr Ian Whyte sampled 424 baobab trees in northern KNP in 1995/1996 to describe population size-class distributions and elephant damage. Nearly all of these baobabs, with the addition of 486 individuals were resampled in 2001 by Michele Hofmeyr. This study resampled the majority of these 910 baobabs plus 126 individuals within 50m of those previously sampled, culminated in a data set spanning 18 years and most of the dominant vegetation and climatic zones within the baobab’s KNP range. Photographs taken in 1995/6 and repeated in 2013 allowed for simultaneous direct visual comparisons of elephant damage to baobab stems. This study aimed to record the temporal and spatial changes in the demographics of, and damage to, this sample of 1036 baobabs in order to define and evaluate the factors threatening KNP baobab persistence. The overall population stem diameter structure of baobabs in the KNP showed a healthy inverse J-shaped distribution, with a high proportion of smaller trees and a decreasing proportion of larger trees at each sampling snapshot. However no seedling recruitment was observed during intensive searches within 50m of sampled trees, accumulating in a total area of 4km2. Growth was only noticeable in sub-adult trees >0.15m and <1m in stem diameter. Annualised mortality increased three-fold between the periods 1995/6-2001 (0.25%) and 2001-2013 (0.79%). It is speculated that this increase in mortality is due to the cumulative effect of increased elephant damage, together with below average rainfall years of 2002 to 2013. Trees that had previously sustained severe and very severe damage contributed the highest proportion (32%) of mortality. When comparing sizeclasses, the majority of mortality (61%) was recorded within the <1m stem diameter size-class. However, a high proportion of very large trees (21% of the 4- 4.5m and 29% of the 5-5.5m stem-diameter baobabs sampled) had died from unknown causes. With no recruitment of seedlings evident and little recruitment between size classes, mortality alone was not enough to significantly change the baobab population structure between 1995/6, 2001 and 2013. This might change should more very large (>4m stem-diameter) trees die, as these size classes make up less than 2% of the population. At a spatial scale, maximum fire return interval, level of elephant damage and mean annual temperature all had significant impacts on the size-class distribution of baobabs. Longer (>30 years) maximum fire return intervals and higher mean annual temperatures (23°C) supported an inverse J-shaped baobab population structure whilst shorter (<30 years) maximum fire return intervals and lower mean annual temperatures (21-22°C) supported a bell-shaped population structure. Increases in bark damage over time correlated with increased elephant density (primarily attributed to mature male elephant density). Smaller baobabs – those less than <1m in stem diameter, generally escaped elephant utilization altogether, or suffered very severe damage by elephants. Despite a 3-fold increase in mortality and no recruitment over the 18 year study period, the KNP baobab population has maintained its inverse J-shaped population structure. The high proportion of mortality in the >4m stem diameter is however concerning as the persistence of these large baobabs is essential for the survival of the population during unfavourable conditions for recruitment.

This thesis develops methodologies for continuous estimation of hydrological variables which infill missing daily rainfall data and the forecast of weekly streamflows from a watershed. Several mathematical programming formulations were developed and used to estimate missing historical rainfall data. Functional relationships were created between radar precipitation and known rain gauge data then are used to estimate the missing data. Streamflow predictions models require highly non-linear mathematical models to capture the complex physical characteristics of a watershed. An artificial neural network model was developed for streamflow prediction. There are no set methods of creating a neural network and the selection of architecture and inputs to a neural network affects the performance. This thesis addresses this issue with automated input and network architecture selection through optimization. MATLABª scripts are developed and used to test many combinations and select a model through optimization. / by Ricardo Eric Brown. / Thesis (M.S.C.S.)--Florida Atlantic University, 2012. / Includes bibliography. / Electronic reproduction. Boca Raton, Fla., 2012. Mode of access: World Wide Web.

The focus of this thesis is extremal dependence among spatial observations. In particular, this research extends the notion of the extremogram to the spatial process setting. Proposed by Davis and Mikosch (2009), the extremogram measures extremal dependence for a stationary time series. The versatility and flexibility of the concept made it well suited for many time series applications including from finance and environmental science. After defining the spatial extremogram, we investigate the asymptotic properties of the empirical estimator of the spatial extremogram. To this end, two sampling scenarios are considered: 1) observations are taken on the lattice and 2) observations are taken on a continuous region in a continuous space, in which the locations are points of a homogeneous Poisson point process. For both cases, we establish the central limit theorem for the empirical spatial extremogram under general mixing and dependence conditions. A high level overview is as follows. When observations are observed on a lattice, the asymptotic results generalize those obtained in Davis and Mikosch (2009). For non-lattice cases, we define a kernel estimator of the empirical spatial extremogram and establish the central limit theorem provided the bandwidth of the kernel gets smaller and the sampling region grows at proper speeds. We illustrate the performance of the empirical spatial extremogram using simulation examples, and then demonstrate the practical use of our results with a data set of rainfall in Florida and ground-level ozone data in the eastern United States. The second part of the thesis is devoted to bootstrapping and variance estimation with a view towards constructing asymptotically correct confidence intervals. Even though the empirical spatial extremogram is asymptotically normal, the limiting variance is intractable. We consider three approaches: for lattice data, we use the circular bootstrap adapted to spatial observations, jackknife variance estimation, and subsampling variance estimation. For data sampled according to a Poisson process, we use subsampling methods to estimate the variance of the empirical spatial extremogram. We establish the (conditional) asymptotic normality for the circular block bootstrap estimator for the spatial extremogram and show L2 consistency of the variance estimated by jackknife and subsampling. Then, we propose a portmanteau style test to check the existence of extremal dependences at multiple lags. The validity of confidence intervals produced from these approaches and a portmanteau style test are demonstrated through simulation examples. Finally, we illustrate this methodology to two data sets. The first is the amount of rainfall over a grid of locations in northern Florida. The second is ground-level ozone in the eastern United States, which are recorded on an irregularly spaced set of stations.

Advanced lightweight structural materials like composites are being increasingly utilized in various engineering applications due to high specific strength and stiffness with tailorable properties. Even though composites have the advantage of modulating a large number of design parameters to achieve various application-specific requirements, this concurrently brings the challenge of dealing with inevitable uncertainties during manufacturing and service-life conditions. This dissertation focuses on practically relevant modelling of random spatial variability coupled with the influence of damage to quantify the effect of source-uncertainties following an efficient surrogate based framework. Layer-wise random variable based approach and the random field based approaches of uncertainty modelling are investigated to quantify the stochastic dynamics and stability characteristics of in a probabilistic multi-scale framework. A novel concept of stochastic representative volume element is proposed to consider the spatially varying structural attributes effectively. A physically relevant random field based modelling approach with correlated material properties is adopted based on the Karhunen-Loève expansion. To understand the relative influences, sensitivity of the stochastic input parameters are analyzed for the global structural responses of composite laminates considering micro and macro mechanical properties separately. Besides the conventional sources of uncertainty in material and structural properties, another source of uncertainty is considered in the form of noise. Besides probabilistic analysis, this dissertation proposes a fuzzy representative volume element based approach for modelling spatial variability in non-probabilistic analysis for the cases where statistical distributions of the stochastic input parameters are not available. The results reveal that stochasticity affects the system performance significantly. A notable difference in the global stochastic behaviour is identified depending upon the adopted uncertainty modeling approach. Thus, it is imperative to appropriately model the sourceuncertainties during the analysis and design process. The dissertation provides comprehensive insights on the effect of source-uncertainties on composites following an efficient, yet practically relevant modelling approach.

Harbour porpoise (Phocoena phocoena) are the most abundant cetacean in UK waters, and are likely to be affected by a variety of marine industries and activities. This research uses data collected by acoustic recorders (C-PODs) and aerial video surveys to investigate patterns in porpoise detection. The findings can be split into five key themes, and are used to support the development of spatial management and survey recommendations. 1. Porpoise detection changes based on time of day in different habitats, indicating possible differences in diel habitat use and highlighting potential issues with visual or video data collection methods for assessing distribution. 2. Porpoise exhibit seasonal shifts in detection, yet year-round data are often lacking, therefore seasonal changes in distribution are often unknown. 3. The highest proportions of buzzes (associated with foraging) are not detected in areas with the highest relative density of porpoise. I propose that porpoise use different foraging strategies in different habitats which are not equally detectable by acoustic recorders. 4. Porpoise distribution may be influenced by the distribution of perceived risk from predator / competitor species (dolphins). Temporal partitioning of sites may arise either from porpoise actively avoiding times when bottlenose dolphins are expected to be present, or from porpoise and bottlenose preferences for different environmental conditions. 5. The choice of spatial modelling method can influence the fine-scale predictions of areas with the highest density. Improving our understanding of top and mesopredator ecology is informative for management strategies. Each of the points raised above should be considered when determining management strategies to minimise the impact from fisheries, offshore developments and other industrial activities on harbour porpoise.

Ho, Sin Yu. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2011. / Includes bibliographical references (leaves 49-52). / Abstracts in English and Chinese. / Chapter 1 --- Introduction --- p.1 / Chapter 1.1 --- Autoregressive moving average time series --- p.1 / Chapter 1.2 --- Integrated stationary time series --- p.3 / Chapter 1.3 --- Some existing methods of identifying d --- p.4 / Chapter 1.4 --- Introduction to Cressie's --- p.6 / Chapter 1.5 --- Outline of thesis --- p.6 / Chapter 2 --- Variogram and Polyvariogram --- p.7 / Chapter 2.1 --- Introduction to variogram --- p.7 / Chapter 2.2 --- Polyvariogram of order b --- p.8 / Chapter 3 --- Testing Procedure --- p.10 / Chapter 3.1 --- Testing for an integrated white noise series --- p.10 / Chapter 3.2 --- Testing for an integrated ARM A series --- p.11 / Chapter 3.3 --- Testing for an integrated linear process --- p.12 / Chapter 4 --- Simulation Results --- p.14 / Chapter 4.1 --- Choice of series length n and r --- p.14 / Chapter 4.2 --- Integrated ARMA series --- p.21 / Chapter 4.3 --- Integrated linear process --- p.39 / Chapter 4.4 --- Comparisons with some methods in literatures --- p.43 / Chapter 4.5 --- An illustrative example --- p.45 / Chapter 5 --- Concluding Remark --- p.48 / Bibliography --- p.49

Thesis (MA. Agricultural Management (Animal Production)) -- University of Limpopo, 2014 / Body measurements have been used in animals to estimate body weight especially in rural areas where scales are not readily available. The study was conducted in Sekhukhune District of the Limpopo province, South Africa where the goats between the ages of 20 and 30 months were used. Body weight and four morphological traits were taken on each goat in the morning before they were released for grazing in a total of 613 indigenous goats of both sexes (62 male and 551 female). Body weight was measured in kilogram (kg) using a balance scale, hip height and shoulder height measurement (cm) were done using a graduated measuring stick and the body length and heart girth were measured in centimetre (cm) using a tape rule. Means, standard deviations (SD) and coefficients of variation (CV) of body weight and morphological traits were calculated. Pair wise correlations among body weight and morphological traits were also determined and range from 0.81-0.91 for male and 0.72-0.89 for female goats, respectively and were significant (P<0.01). The direct effect of heart girth on body weight was the strongest in both sexes (path coefficient of 0.58 and 0.62 in males and females, respectively). Body length (males) and body length and hip height (females) also positively (P<0.05) influenced body weight. The direct effects of other linear type traits on body weight in both sexes were non-significant as revealed by t-test. These traits were indirectly realized mostly via heart girth. Thus, they were removed from the final regression equations to obtain much more simplified prediction models. The optimum multiple regression equation for male goats included heart girth and body length, In female goats, however, the v variables included in the final prediction model were heart girth, body length and shoulder height. The forecast indices obtained in this study could aid in weight estimation, selection and breeding programmes.

Thesis (MA. Agricultural Management (Animal Production)) -- University of Limpopo, 2014. / Body measurements have been used in animals to estimate body weight especially in rural areas where scales are not readily available. The study was conducted in Sekhukhune District of the Limpopo province, South Africa where the goats between the ages of 20 and 30 months were used. Body weight and four morphological traits were taken on each goat in the morning before they were released for grazing in a total of 613 indigenous goats of both sexes (62 male and 551 female). Body weight was measured in kilogram (kg) using a balance scale, hip height and shoulder height measurement (cm) were done using a graduated measuring stick and the body length and heart girth were measured in centimetre (cm) using a tape rule. Means, standard deviations (SD) and coefficients of variation (CV) of body weight and morphological traits were calculated. Pair wise correlations among body weight and morphological traits were also determined and range from 0.81-0.91 for male and 0.72-0.89 for female goats, respectively and were significant (P<0.01). The direct effect of heart girth on body weight was the strongest in both sexes (path coefficient of 0.58 and 0.62 in males and females, respectively). Body length (males) and body length and hip height (females) also positively (P<0.05) influenced body weight. The direct effects of other linear type traits on body weight in both sexes were non-significant as revealed by t-test. These traits were indirectly realized mostly via heart girth. Thus, they were removed from the final regression equations to obtain much more simplified prediction models. The optimum multiple regression equation for male goats included heart girth and body length, In female goats, however, the v variables included in the final prediction model were heart girth, body length and shoulder height. The forecast indices obtained in this study could aid in weight estimation, selection and breeding programmes

Dans la première partie nous étudions l'apprentissage et le rappel dans des réseaux de neurones à une couche (modèle de Hopfield). Nous proposons un algorithme d'apprentissage qui est capable d'optimiser la 'stabilité', un paramètre qui décrit la qualité de la représentation d'un pattern dans le réseau. Pour des patterns aléatoires, cet algorithme permet d'atteindre la borne théorique de Gardner. Nous étudions ensuite l'importance dynamique de la stabilité et d'un paramètre concernant la symétrie de la matrice de couplages. Puis, nous traitons le cas où les couplages ne peuvent prendre que deux valeurs (inhibiteur, excitateur). Pour ce modèle nous établissons les limites supérieures de la capacité par un calcul numérique, et nous proposons une solution analytique. La deuxième partie de la thèse est consacrée à une étude détaillée - du point de vue de la physique statistique - du problème du voyageur de commerce. Nous étudions le cas spécial d'une matrice aléatoire de connexions. Nous exposons la théorie de ce problème (suivant la méthode des répliques) et la comparons aux résultats d'une étude numérique approfondie.