Synthetic geotechnical investigation: AI soil prediction
Predict ground conditions between and beyond your boreholes, and decide where to drill next.
Physical boreholes and CPTs are sparse, slow and expensive offshore. Darkocean's AI/ML borehole prediction combines the geotechnical data you already hold with geophysical surveys to estimate soil profiles across the whole site. It supports planning and targeting of physical investigation. It does not replace it.
Existing site data is the starting point
With uncertainty, not a single guess
Targets and reduces physical sampling
Tested against ground truth you hold back
What is synthetic geotechnical investigation?
A conventional offshore ground investigation samples the seabed at a limited number of points. Everything between those points is interpolated by an engineer, usually by judgement and by reading geophysical data alongside the logs. Synthetic geotechnical investigation puts that step on a quantitative footing: a machine learning model learns the relationship between geophysical measurements and measured soil behaviour at the locations where both exist, then predicts likely soil profiles and CPT-like parameters, such as cone resistance and sleeve friction trends, where only geophysics is available.
The geophysical inputs are typically seismic reflection or sub-bottom profiler data, multibeam bathymetry and backscatter, and where available acoustic impedance or velocity models. The geotechnical inputs are borehole logs, CPT traces and laboratory results. Because soils are spatially correlated, the model can also use nearby measurements directly, in the same spirit as geostatistical methods such as kriging, but with the geophysical data constraining the geometry of layers.
Predictions carry uncertainty. Where the model has seen similar ground, the uncertainty band is narrow. Where the site contains conditions not represented in the training boreholes, such as a buried channel, boulders or cemented layers, the band widens and the result is flagged. That is useful in itself, because it shows where the next borehole or CPT will add the most information.
The result is decision support. It helps you place fewer, better-targeted investigation points, plan foundation or cable-route options earlier, and focus design effort. It does not remove the need for physical sampling, and design values for foundations should always rest on measured data and the judgement of a qualified geotechnical engineer.
How it works
Collate
Gather boreholes, CPTs, lab results and geophysical data, and tie them to a common coordinate and depth reference.
Train
Fit models that link geophysical response to measured soil behaviour at locations where both are known.
Validate
Hold out boreholes, predict them blind, and report how closely the model reproduces the measured profiles.
Predict
Generate soil profiles and CPT-like parameters across the site with uncertainty at each location.
Target
Rank the areas where extra boreholes or CPTs would most reduce uncertainty, and hand over for engineering review.
What you receive
Predictions
a. Predicted soil profiles at chosen locations
b. CPT-like parameter profiles with uncertainty
c. Depth slices and cross-sections of soil units
Targeting
a. Uncertainty maps across the site
b. Recommended locations for additional boreholes or CPTs
c. Comparison of predicted against measured held-out points
Reporting
a. Method and assumptions statement
b. Data-quality and coverage notes
c. GIS-ready layers for the engineering team
Where it is used
Offshore windPlan foundation investigation across a wide lease area and place CPTs where ground variability is greatest.
Export cable routesScreen long linear routes for changes in seabed and near-surface soils before committing to a full campaign.
Pipelines and flowlinesIdentify stretches of soft, coarse or cemented material that need closer sampling.
Ports and quay wallsFill the gaps between boreholes along waterfront and dredging footprints.
Jack-up sitingSupport early assessment of punch-through and leg penetration risk at candidate locations.
Desk studiesBuild a first view of ground conditions at a new site before any vessel is mobilised.
Frequently asked questions
Does this replace boreholes or CPTs?
No. The model is trained on physical measurements and relies on them. It estimates ground conditions between and ahead of samples so that physical investigation can be targeted and, in some cases, reduced. Design parameters should still be based on measured data.
What data do you need?
At minimum, some borehole or CPT data with accurate positions and depths, plus geophysical data covering the same area, such as sub-bottom profiler or seismic data. More measurements, and a closer match between geophysics and sampling locations, give better predictions.
How accurate are the predictions?
It depends on how much ground truth you have and how variable the site is. We test the model on boreholes that were held back from training and report the agreement, so you can judge reliability for your site before relying on any result.
Can it find hazards such as boulders?
It can flag areas where geophysical response is unusual or where the model is uncertain, which often coincide with hazards. It cannot confirm a boulder or a cemented layer. Those require dedicated geophysical interpretation or physical sampling.
Who should review the output?
A qualified geotechnical engineer. The predictions are an input to their judgement, alongside the physical data, codes and standards that govern the design.
Where should your next borehole go?
Send us your existing geotechnical and geophysical data and we will tell you what a synthetic investigation can add.
