The results are significant both in the fields of geotechnical engineering and geophysics as they demonstrate how to unlock the value that airborne geophysical data offer for efficiently mapping quick clay on a large scale.
Helicopter-based airborne electromagnetics, also known as AEM, is something EMerald Geomodelling has been using for a long time. It’s an efficient method to map the ground conditions, as it measures the electrical resistivity of different materials in the ground.
One would think that this could be easily applied to identify quick clay as well, given that quick clays have a lower salt content compared to stable clay, and therefore have a different resistivity signature.
Unfortunately, that hasn’t been the case.
Machine learning - the key ingredient
In the past, many groups have tried using both frequency- and time-domain EM to map quick clay. However, their success has been limited, partly because certain types of soil can have resistivity similar to that of quick clay.
In other words, quick clay has a non-unique resistivity signature, which makes it difficult to identify solely with the use of AEM. Simple interpretation techniques, such as choosing a resistivity threshold, lead to poor models of quick clay occurrence - with many false positives.
Emerald Geomodelling’s breakthrough was to implement machine learning algorithms. By combining the geophysical and geotechnical parameters together in nuanced way, we managed to resolve the ambiguous resistivity signature of quick clay.
Our study found that using spatial attributes helped us keep track of contiguous bodies of the same type of material and to closely match the geotechnical data. At the same time, we found that the geophysical data added value by ensuring that the output was more geologically plausible between boreholes. This is significant, as it enables us to estimate the uncertainty in our predictions, which again helps us pinpoint where further ground investigations are most needed.
What is the impact?
The new method is a very important innovation, as traditionally it has been very difficult and time consuming to do large-scale quick clay hazard mapping. Whereas traditional site investigation methods give information only at a single point location, our integrated method gives continuous data coverage over a large area.
This results in both time and cost savings in infrastructure projects, as it allows the contractor to focus detailed site investigation only where they are truly needed. It also has a positive environmental impact, as reducing the number of boreholes drilled and reducing the amount of cement stabilization required during construction minimizes the overall CO2 emission.
Still, there are still some limitations and weaknesses that will be addressed going forward. For one, helicopter-based AEM has a limited resolution. Our best-performing machine learning algorithm compensated for this by using the information contained in spatial attributes. However, using more physics-guided modelling would likely improve results.
This could be achieved by utilizing additional data sets, by including higher-resolution resistivity models from ground- or drone-based system, or by accounting for lateral variations in resistivities. EMerald Geomodelling will continue exploring all these alternatives going forward.
Interested in hearing more?
If you want to learn more about the research, the full article - “A machine learning–based approach to regional-scale mapping of sensitive glaciomarine clay combining airborne electromagnetics and geotechnical data” - is available online for free at Wiley. Read it here!
You can also contact the EMerald team directly, at firstname.lastname@example.org or +47 481 32 299 (NO)/+1 650 656-0174 (US)