Appraisal of geostatistical methods and geostatistical prediction of predominantly marine sand inclusions in the Frimmersdorf lignite seam in the Garzweiler open cast mine, Germany

Appraisal of geostatistical methods and geostatistical prediction of predominantly marine sand inclusions in the Frimmersdorf lignite seam in the Garzweiler open cast mine, Germany

Van Beuningen, F.W.G.M.

Benndorf, J. (mentor)
Thielemann, T. (mentor)

Civil Engineering and Geosciences

Geoscience & Engineering

Resource Engineering

2013-09-25

Mining is different from most businesses because knowledge of the product is essentially based on estimates, which by their very nature include a degree of uncertainty (Dominy, Noppé, & Annels, 2004). The risks associated with mining are varied and complex, where the dominant source of risk is the orebody itself (Snowden, Glacken, & Noppé, 2002). The modeling of orebody geology is the basis for all prediction of in-situ grades, mineral resources and recoverable ore reserves, as well as mine design and long term production forecasting. The Lower Rhine Basin is a rift basin located in northwest Germany which hosts several large lignite seams of considerable economical importance. One of these massive lignite deposits - the Garzweiler open cast mine - is containing marine and fluvial sand intrusions which are significantly affecting the reserve estimation and the operational processes. The current model based on a deterministic "best guess" approach fails to properly predict these sand partings within the lignite seams. Probabilistic or stochastic models can provide an alternative. The aim of this thesis is to elaborate a first appraisal of the possibilities for RWE Power AG to apply geostatistical modeling at its Garzweiler lignite operation. An analysis is made of the available datasets and of the applicable geostatistical methods, including their strengths and shortcomings. This exercise of applying the relevant geostatistical methods to the different datasets provides new insights for predictive modeling. On this basis, recommendations are formulated for geostatistical modeling methods applicable to the unmined area of the Garzweiler mine. Data collection is not an aim in itself. Purposeful data collection presupposes a theory or a model which gives meaning and significance to raw data by processing them as information that is useful and needed for a specified goal. In the context of this project, this goal is the accurate prediction of the composition of a lignite seam. The different data types are thoroughly described and their strengths and limitations are clearly indicated. From the point of view of geostatistical modeling, the current practice of collecting and processing surveying data has potential for improvement, because of the limited applicability of the existing data. While collecting surveying data demands great effort, production data - digitally stored and therefore easily automated - are readily available at no cost. A continuous, automated input of available production data creates a model with permanently updated geostatistical predictions and therefore a reduction of the empirical error. In addition, drill data provide information for a wider area about not only the coal / sand ratio, but also on the ash content of the lignite. Ideally, this geostatistical model should be complemented with the KOLA data - online analytics of coal composition. The KOLA data themselves can be used for predictive purposes, but above all they possess the unique advantage of being able to validate the predicted ash content. However, practical obstacles prevent an effective use of these data. The ordinary kriging method is widely in use and in general gives good predictions. These predictions are accompanied by an estimated error. In the context of the stationarity assumption - one of the fundaments of kriging - a significant limitation arises: ordinary kriging is poorly able to cope with sudden “structural” changes. Introducing an automation step will generate a model where continuous input and usage of the available data will result in permanently updated geostatistical predictions and a reduction of the empirical error, particularly in changing environments. The indicator approach, though not providing accurate predictions, does have the virtue of signaling areas with specific features. As opposed to kriging, simulation does not require a normally distributed dataset. The creation of multiple realizations gives a better feeling for possible scenarios of future mining. Simulation applied to ash content data presents promising results.

lignite mining
geostatistics

http://resolver.tudelft.nl/uuid:6d00e92a-989c-40e1-a34e-2e24a64e4a27

2015-09-25

Student theses

master thesis

(c) 2013 Van Beuningen, F.W.G.M.