'Real' Geology and Computer Models.
This summer in England has been fairly typical - mostly cool and wet, with just a few hot days to remind us what summer can be like. However, I've had the opportunity to go out and about looking at rocks. Kyrill Pshenichny visited last week from St.Petersburg en route to do some volcanological fieldwork on Montserrat. We visited some of the classic geological locations around here. It struck me then, as it has done many times when I look at geology, how inadequate is computer modelling - any computer modelling - to capture the essence of real geological objects and relationships.
For many mining projects of course this is not too important a matter. A couple of months ago I visited Winsford Salt Mine in Cheshire, where they are extracting rock salt underground from a very thick and very uniform deposit, in the oldest underground mining operation in Britain (and probably one of the oldest in the world that is still being worked - it started in 1844 and is still going strong). Mine planning there is controlled purely by the bulk mechanical properties, and the geometry of their room and pillar operation. A few months ago I visited a number of chromite mines in the Bushveld, South Africa. There too, the detailed geology was of little importance to the mining operation as the reef deposits were very continuous in both quality and thickness.
However, in most mining operations the geology is of crucial importance. Whether quarrying aggregates or extracting copper or gold, the deposits vary in quality from place to place. Estimation of the spatial variations in both quantity and quality of materials to be extracted is an essential part of the mine planning process.
With development of computers in the past 50 years, has come the development of some very clever numerical techniques which purport to give good - or even the 'best' spatial models. Unfortunately, as many mining companies know only too well, the mineral values forecast by these methods often fail to materialise. Indeed, the discussion in the August issue of ESCA might well have been 'Are Computers Killing the Mining Industry?'
One of the big problems is the straitjacket imposed by the tractable mathematical techniques which have been applied. Assumptions of stationarity, additivity, distributions, and so on are made, often with little attempt at validating them. Of course real geology is not constrained by such assumptions, but by very variable and often very local physical and chemical controls. Where continuity holds over the scale of a mine or a mining field - as in the case of many salt, chromite, or coal deposits, for example - the geologist and the mining engineer can breathe a sigh of relief that all is well-behaved. They get nice variograms, and the estimates come out spot-on. However, where geological continuity can be measured in metres (or even in millimetres) life is not so simple. The words 'nugget effect' tend to be used to avoid doing a lot of geological interpretation. 'Stationarity' can be established by splitting up a deposit into sets of 'homogeneous' zones. Of course, in many cases the instant result of doing this is that some or all of the zones contain too little data to establish any geostatistical properties, let alone assure local stationarity.
What is actually needed is a more intelligent approach to the problem. A real geologist needs to do real geological mapping - at a scale of metres if necessary - and this will involve close-up examination of faces. It is then necessary to develop a geological model, and here I mean a mental model of how the geology evolved to produce the mineral deposit. This could be a model of the volcanic and hydrothermal history of a porphyry copper or a VMS deposit, a river sedimentation model for sands and gravels, a weathering history for a kaolin deposit, etc. Only then might it be possible to generate a computer model, when the geological constraints are known.
So why is this not standard practice ? A difficult question to answer. Maybe it's the attraction of powerful computer methods offering a quick fix - 'conditional simulation will solve all your problems'. Maybe it's just that there are too many geologists who prefer to look at a computer screen rather than look at rocks in the heat and dust of a mine. Or maybe it's the huge gap between what real geology is like and what 'industry-standard' computer methods will deliver. As an example of this last point, geology abounds with discontinuities - bedding planes, faults, joints, unconformities, .... These all represent the end results of a huge variety of geological processes. Yet in the usual computer modelling approaches, all are handled in one of two ways: not at all, or by splitting the region into different zones along the discontinuity surface. There is no recognition of the underlying process, or of its effects on the rock properties. A mine should not be considered as a research project of course - it is a commercial enterprise. However, a little understanding of the geology would help to maximise the commercial returns by improving the quality of spatial prediction. And if this means sometimes that conventional geostatistical methods might not be usable - well so be it - you need to find other numerical methods. "A rock is a true fact" - however inconvenient it might be !