Geostatistics as a discipline and Resource Modeling Solutions (RMS) as a division of GeologicAI do not provide solutions looking for problems. Critical problems in resource modeling have driven the development of geostatistics and motivated RMS from the beginning. This blog summarizes some of the problems we aim to tackle.
Rock properties are variable at all scales and limited samples are available. This leads to uncertainty. Uncertainty leads to risk of underperformance or risk that the true potential of a deposit is not realized. Rock properties are not independent of location stemming from complex geological processes. There are interesting and site-specific spatial features that are correlated over different length scales. Although this spatial dependence is unique among many statistical and machine learning techniques, it permits probabilistic prediction between drill holes and leads to improved resource estimates. In this context, let’s discuss a few of the problems that motivate geostatistics and RMS.
Mapping Geology. 2-D geological maps and 3-D models must be created from widely spaced data to appreciate or visualize rock properties over an area/volume of interest. The rock properties of interest must be interpolated in a geologically reasonable manner incorporating geological understanding while respecting the available data (drilling, remote samples, trends, geophysical observations). The problem is to create geological maps and models that are an accurate reflection of the true distribution of rock properties in the subsurface.
Integrating Different Data Types. Ideally, a statistical modeling project would work with a full valued set of measurements of all variables with negligible measurement error. Creating geological models of mineral deposits, however, faces different drilling (types, vintages, diameters), scanning at sub millimeter scale, geophysical responses measuring 10s to 100s of cubic meters, and production history responding to 10s of thousands of cubic meters of rock. The problem is to create geological models that reproduce all data sources within their volume support and error content.
Quantifying Uncertainty. What are the knowns, what are the known unknowns, and what are the unknown unknowns? Almost by definition there is nothing we can do with the unknown unknowns. Thankfully, given that we have evaluated hundreds of deposits, there are fewer of those. Most resource uncertainty falls within the known unknown category. This uncertainty must be quantified and managed. The transfer of uncertainty to risk must also be understood and managed. The problem is to quantify the uncertainty that must be managed.
Reflecting Heterogeneity. Many industrial processes are strongly dependent on natural variability. Flow processes are extremely sensitive to connected high permeability conduits and low permeability baffles, which dominate flow behaviour. Blending, stockpiling, stacker/reclaimer design, sampling, dilution at the practical scale of mining and many other challenges are also dependent on natural variability. Smooth interpolated estimates between widely spaced data are not fit for purpose. The problem is to create spatial models with the correct natural variability or heterogeneity.
Reporting Resources. Tonnes, grade, metal and value must be calculated and (often) disclosed to the public to support investment and internal decision making. There are reporting codes that must be adhered to, but they leave much in the hands of a qualified or competent person. All best practice tools must be considered for calculating the many rock properties that influence the valuation of a deposit. Considering that 1/10 deposits are drilled on and 1/100 of those drilled on are turned into a mine, the problem is to calculate defendable, comparable and consistent estimates across many deposits.
Recommending Drill Hole Spacing. The disclosure of resources must consider the level of geological confidence that the qualified person has in the resources. There are many ways of performing this task (a minor problem is to implement many of those criteria), but the amount of data available, summarized by the data spacing, is a key input for classification. Increasingly, we are considering multiple data types and sources that must be optimized simultaneously. The problem is to recommend the data spacing for different purposes considering production uncertainty at differing scales of relevance.
Valuing Information. There are many circumstances when the decision of data collection depends on the value that the data will bring. Clearly, more data will lead to a better result, but at a cost. Less data will lead to suboptimal decisions, but at a lower cost. This depends on the scale of geological variability and the ability of the engineering design to respond to the additional data. We must value a one-off data collection like geophysical data or a variable data collection like production data spacing. The problem is to consider the value of information and collect only the data that will pay for itself.
Mine Planning. Mining engineers need reliable resource block models as a foundation for mine planning and design. Mine engineers are concerned with pit staging, stope sequencing, development strategies, and equipment requirements to provide a steady feed of ore while ensuring removal of required waste and providing required access. Numerical geological block models must provide realistic and unbiased estimates of future production anticipating that more information will be available at the time of mining. The problem is to construct block models satisfying all of these requirements.
Ore Control. A high-resolution model of the rock properties is essential at the time of production where irrevocable decisions are being made. In surface mining, there is a requirement to optimize the routing of mined material. In selective underground mining, there is a need to determine the stope boundaries of what will be mined. These decisions may depend on multiple rock properties and other engineering considerations such as the asymmetric consequences of lost ore and dilution given fixed milling capacity. The problem is to create numerical models that support optimal short-term decision making maximizing expected economic return.
Geometallurgical and Geomechanical Property Models. Mining decisions are strongly influenced by the metallurgical processing characteristics and geomechanical properties of the rock. Geomet/geomechanical rock properties have unique characteristics such as limited sampling, strong non-linearity, and dependency on extrinsic factors. They are sensitive to intact rock properties and the discontinuities that further define the rock mass. The problem is to construct predictive models of geomet/geomechanical rock properties that are fit for purpose.
Repeatable and Rapid Decision Support. Many of these problems are addressed multiple times over the life of the mine (and in the case of ore control, often daily). To mine with confidence, the models generated must be repeatable and timely to provide rapid decision support. All appropriate steps of a workflow should be automated, so that limited practitioner time is dedicated to checking, interpretation, and refinement. The problem is to provide high quality models incorporating the latest data, in a timeframe that permits their effective use in decision making.
These problems are ongoing; they are not solved with one analysis, then forgotten. A thoroughly tested, repeatable, scripted, highly intelligent, and adaptive workflow must be applied on an ongoing basis. There are many other problems that are addressed by the tools encoded within our platform, but this provides a reminder of the core problems we tackle. Resource Modeling Solutions (RMS) has the latest techniques implemented for solving resource modeling problems.
