Mining natural resources is a game of accuracy. Prospectors find the deposits. Miners extract the deposits. Selecting the richest basin, deposit and layer is the game. Operators are rewarded for finding the highest concentrations. This has not always been as tricky as it is today. Resources were much more widely available, accessible and abundant. Low-hanging fruit is no longer.
The industry has progressed from the age of abundance to the frontiers of scarcity. Resource scarcity has pushed miners ever deeper into remote, hostile and hazardous terrain. Corporate governance accountability and supply chain transparency throughout project lifecycles leave little margin-of-error.
The environmental footprint, including chemical volumes and diesel emissions, is a top concern in capital markets. The neck-and-neck competition of global markets squeezes commodity prices ever closer to their cost of production. Efficiency and productivity gains are crucial to maintaining a freedom-to-operate, low cost of capital and healthy returns to shareholders.
Accuracy alone is no longer sufficient. Maximising output, minimising input and mitigating waste must be done with ever greater precision. Markets, shareholders and customers demand it. As mining writes the playbook on its digital revolution, novel precision technologies are playing a pivotal role. Sensors, in particular, are helping to connect the physical and mechanical with the digital.
Read on for our interview with Friederike Koerting, founder and COO of rad.Data Spectral Analytics, on hyperspectral's benefits to the mining industry. She has a vision for the future of mining in which advanced optical sensing as a persistent eye-in-the-sky will be instrumental to mine operators who want to mitigate risk, reduce costs and unlock value. Proactive, continuous monitoring of open-pit mines in real-time can help make better, and crucially, fewer decisions.
It is essential to clarify some terms. Hyperspectral imaging or reflectance spectroscopy are terms that capture a simple concept. In principle, each material, in this case, an element like gold or quartz, reflects light differently. This is how vision systems, including the human eye, work. Our eyes register the light or the colour and texture that objects reflect.
Hyperspectral takes this concept to a new level of detail in two ways. In terms of the electromagnetic spectrum, it can see beyond what a human eye can see. In other words, there is the visible (e.g. red, green, blue) and the non-visible spectral range, which includes infrared, UV and X-ray. Some animals like bats or the mantis shrimp can see these so-called non-visible spectra, but not humans. Hyperspectral can see into the non-visible spectra to extract much more information about an object of interest.
Hyperspectral can see with much more detail. The resolution is such that it can extract information at the pixel level, building a visual map of mineral concentrations. In general, hyperspectral imaging has advanced to the point that the optical sensitivity is not the primary limiting factor. 
It is easy to get tripped up in the technical aspects. But the future of hyperspectral is exceptionally bright. Hyperspectral can, in principle, measure and map the chemical and molecular makeup of the surface of any object, crucially, non-invasively, non-destructively and rapidly. This is a revolutionary concept for anyone reliant on taking samples to a lab, which is the case in mining, to receive their material's chemical or analytical profile. Imagine knowing what is under your feet in real-time?
It is a matter of communication or education, plus the ruggedness and reliability of the equipment. The sensing technology is ready. Better communication between experts in optics and mining, respectively, will go a long way. We can now fix hyperspectral cameras to aerial systems such as UAVs to scan open-pit mines or bench-top systems to scan drill core samples.
Adoption has been slow, however, because the value proposition is not quite in alignment. Operators are not interested in more data, analytics or more decisions. Operators want fewer and faster decisions. Hyperspectral holds significant potential as a tool to guide rapid decision-making. Updated mineral concentration maps, for example, can guide extraction plans and chemical treatments. Higher precision will yield greater efficiency and productivity. We have been developing exactly this application at copper pit mines in Cyprus, for instance. At rad.Data we are packaging and developing something super-rugged that non-experts can use.
The sensors we have access to now are technically superb, but they are sometimes too delicate. The cameras can be over-engineered for applications that simply ask for a binary decision between "valuable" and "non-valuable" material. But to get to that point, you need advanced geoscience and optics skills paired with application expertise. Domain experts and application developers like rad.Data working directly with mine operators will produce the required solutions.
We aren't fixing the sensors to satellites or airplanes. They are mounted onto a drone or a tripod in close proximity to the pit front from a distance of 100m or less. This can make it easier to find the minerals of interest in the needed spatial resolution (pixel size) that is optimal for the complexity of the deposit. Reducing the sensor complexity to a level required for a specific deposit type will make these sensors more rugged to the harsh environment and easier to handle.
Another challenge is in the amount of data collected. More imagery requires more power and data processing. A classic hyperspectral camera can have more than 100 bands, and those bands are narrow and collect the incident light in consecutive overlapping bands. Some readers may be familiar with multispectral, which works in the same spectral range (400 – 2500nm) but has a lower number of channels collecting in between 3-15 distinct bands, depending on the application. The full spectral range is not strictly required, depending on the application and deposit.
Copper alteration mineralogy, for example, is most responsive in the SWIR region to distinguish between the potassic and phyllic alteration zone. Elemental copper, however, is not spectrally active. We can, however, improve the performance of the camera by trimming down the number of channels. Ideally, we would tune the cameras like an old television, reducing the camera to fewer and the right set of channels for the deposit at hand. This is another aspect of the art of extracting the signal from the noise.
I am optimistic about what I call superspectral, between 15-100 bands, for mining, where we have the appropriate balance of performance, sensitivity and specificity. The more bands, the higher the signal-to-noise ratio, but that comes with a few downsides in the field. By looking at larger portions of the spectrum within fewer spectral bands (wider bands) and changing this setting in the sensor's hardware, it is possible to make the sensor more rugged and facilitate a more straightforward experience for the operator. This also enables easier data management and fast processing and analysis. Of course, this reduces the level of spectral detail that the sensor can detect, and narrow absorption features cannot be detected. It is a balance that we are in the process of working out.
Mining operations often take place in extreme environments and far-flung locations. There are 4500 mines worldwide, half of which are open pit, with total annual operating expenses of about $500bn. That's $100m per mine. Mining remains volume-driven in the sense that opportunity costs are high, so operators are trying to extract as much ore as possible. There can be a lack of precision. Remedying this is especially important as mines gradually become more automated. Autonomous equipment has to navigate variable terrain deftly. This is just not possible with blunt instruments. Sensitivity to the terrain and environment is critical.
Hyperspectral imaging is an underrated and powerful real-time decision-support tool. As layers are removed from the mine face or added to a stockpile, you can generate material profiles at the pixel level. At the moment, much of the decision-making relies on low-resolution geomagnetic, geophysical and spot samples or point-in-time data. This could be analogous to a farmer continuing with uniform seeding, knowing that natural density varies by the square meter, not the hectare. Ignoring the distribution of mineral concentrations in real-time is a missed opportunity to save time, money and operate a safer, more consistent operation.
Friederike Koerting wrote her doctoral thesis on superspectral copper ore detection and the implementation of hyperspectral data and instruments in the open-pit mining environment. Her experience in the VNIR-SWIR spectral range and the mapping of outcrops, mine faces and drill cores was gained through multiple field campaigns around the globe and countless hours in the hyperspectral lab where she started her research as a student in 2012. She is the co-founder of rad.Data, where she is applying hyperspectral imaging to help miners mitigate risks, unlock value and save time.
 The challenge is not in the sensor or the data processing but in how we integrate these insights into established and evolving decision-making systems. Hyperspectral scans of the topsoil and pit face provide a spatial understanding. You can map the distribution of minerals as the mine evolves. Plus, it can be used to scan dill cores without waiting for results which can arrive the next day. This saves a lot of time in an operation that has high per-hour opportunity costs. We can also see significant potential as a safety and environmental protection tool. And as a way to reduce the environmental footprint.
For example, the level of liberation depends on the deposit and mineral. Hyperspectral cannot predict the level of liberation. But it can be used for bulk sorting when the metal is sufficiently liberated in bulk form or when the deposit itself is heterogeneous and we can distinguish between different ore types by hyperspectral. If the metal is not liberated and disseminated within the bulk ore, different liberation methods or steps may have to be taken before hyperspectral can be used for sorting efforts. In other words, hyperspectral is a useful tool for increasing efficiency (e.g. reduced chemical use).
 A phyllic alteration zone is rock that has been affected by water. It is one of several types of alteration, which can indicate the character of the minerals or the kinds of elements in the area.
 We are getting better at generating proxies for mineral concentration, which is hard to pinpoint. There is a way of semi-quantitatively analysing the pixels, but a material distinction is a lot easier, and qualitative analysis is usually happening. But, if the spectral library already presents the different levels of ore grade, it is a proxy for quantity and concentration.