The Ellisras Basin, with huge coal resources, is fault-bounded along its southern and northern margins and is a graben-type deposit. The study area is situated in the south-western part of the Limpopo Province of the Republic of South Africa and is geologically located in the Ellisras Basin. In this area the basin is influenced by three major fault zones, the Eenzaamheid Fault delineating its southern limit, the Zoetfontein Fault near its northern limit and the Daarby Fault, with a down-throw of approximately 350 m towards the north-east. Sedimentological facies changes also influence the continuity of the coal zones, with deterioration in coal development. The exploration project was a collaboration between two of the large role players in the South African coal mining industry Sasol and Exxaro, for the purpose of identifying whether the coal in the Ellisras Basin could be used for gasification purposes in the Sasol process, and that enough resources exist on the farms on which the two companies have the exploration rights.. The prospecting method used at the Project area, situated 50 kilometer west of the town of Lephalale in the Limpopo Province of South Africa, comprises the drilling of cored exploration boreholes on a random spacing of ± 1 000 m x 1 000 m, together with infill percussion drilling. The use of slimline geophysical methods to log lithologies is a technique which has been used extensively in the mining industry over a number of years. At the Project area the correlation between the measured densities derived from the traditional method of air and water measurement and those derived from Vectar processed derived densities from geophysical logging is better than 95%. As a method of “fingerprinting” the various coal zones and samples it was decided to calculate the distribution of relative densities in the chosen geological intersection. The data was then used to portray geophysically derived relative density cumulative distribution line diagrams (GDCDD) of the various lithotypes on either a sample-by-sample or zone-by-zone basis. Using the classification method proposed, the various coal seams and zones can be correlated to a high degree and discrepancies easily identified. The lateral correlation between lithologies can be accurately described and substantiated, and this would convince a Competent Person that the method proposed is invaluable in classifying coal resources in the coal basins. / Thesis (PhD)--University of Pretoria, 2014. / lk2014 / Geology / PhD / Unrestricted

[pt] No presente trabalho de tese é apresentado, analisado e avaliado um método inédito para prever perfis de poço (para um pseudo-poço), a partir de informações dos poços verticais vizinhos e dados de sísmica 3D. Este método é formado por quatro partes principais. Na primeira parte são processados os perfis de raio gama dos poços para produzir um conjunto de sinais estratificados. Estes resultados são utilizados na segunda parte para realizar a correlação lateral de poços, com um método baseado em Algoritmos Genéticos, cujos resultados são intervalos de estratos semelhantes que indicariam a existência de estruturas rochosas contínuas na subsuperfície. Na terceira parte são estimados os deslocamentos, em profundidade, entre alguns estratos do pseudo-poço e dos poços reais. Isto é realizado com um método, baseado em Algoritmos Genéticos, para correlacionar eventos entre traços sísmicos 3D nas posições próximas aos poços reais e ao pseudo-poço. Na quarta parte são utilizados os resultados das outras três partes para interpolar vários tipos de perfis do pseudo-poço, primeiramente, dos estratos semelhantes e, em seguida, dos estratos não semelhantes, ajustados em profundidade. Os resultados desta parte constituem as medições previstas. Os pesos para a interpolação, pelo método de Shepard, são calculados através de uma função que depende da distância entre o pseudo-poço e cada um dos poços utilizados. Um estudo de sete casos é apresentado para avaliar a eficácia do método de previsão. Esses casos utilizam arranjos de dois e três poços do Campo de Namorado na Bacia de Campos (RJ) e, adicionalmente, um poço próximo a cada arranjo serve de padrão para avaliar as medições previstas através do coeficiente de correlação e da média dos erros percentuais absolutos. Em termos globais os resultados das previsões apresentam correlação linear forte e os erros percentuais absolutos estão entre 5 por cento e 30 por centoem relação aos padrões. / [en] In this thesis is presented, analyzed and evaluated a new method to forecast well logs (for a pseudo-well), based on information of the neighboring vertical wells and the 3D seismic data. This method consists of four main parts. In the first part, the gamma ray logs from wellbores are processed to produce a stratified set of signals. In the second part, the stratified signals are used to perform a method of lateral correlation of wells with Genetic Algorithms; whose results are similar intervals strata that would indicate the existence of continuous rock structures in the subsurface. In the third part, a method, based on Genetic Algorithms, are used to estimate displacements in depth between some strata of the pseudo-well and real wells. That method correlates events between 3D seismic traces at positions near the real wells and the pseudo-well. In the fourth part, results from the other three parts are used to interpolate several types of logs of the pseudo-well, first of the similar strata and then of the not similar strata, adjusted in depth. The results of this part are the forecasted measurements. The weights for interpolation, by the method of Shepard, are calculated using a function that depends on the distance between the pseudo-well and each real well. A study of seven cases is presented to evaluate the effectiveness of the forecasting method. These cases use arrangements of two and three wells of the Namorado Field in the Campos Basin (RJ) and additionally a real well near each array serves as a standard for evaluating the forecasted measurements by the correlation coefficient and the mean absolute percentage error. Overall the results showed a strong linear correlation and the mean absolute percentage errors are between 5 per cent and 30 per cent in relation of standard data.