Why QA/QC Matters in Geotechnical Core Logging
Geotechnical drilling is expensive and time-consuming, so maximalising the value of data collected from each drill hole is essential. However, completing full geotechnical logs for all drilling may not always be feasible due to time constraints, budget limitations, or the availability of qualified geotechnical engineers and engineering geologists.
Datarock’s automated logging generates high-quality geotechnical datasets from core photographs, either using either general or site-specific models. To support confidence in these outputs, Datarock is committed to delivering transparent and auditable datasets.
This document presents the methods we use to validate Datarock’s predictions and provides guidance on how to perform quality assurance and quality control (QA/QC) on against ground-truthed logging, to ensure that the data is fit for downstream geotechnical analysis and design.
Common Data Quality Issues in Geotechnical Core Logging
Interpretation Variability Across Logging Teams
When a geotechnical drilling program is completed by a team, no matter how in sync the team is, inconsistencies in the logging will arise. These inconsistencies can arise from; individual interpretation, varying levels of training and experience between loggers, or changes in logging practices and standards over time, particularly when historical datasets are combined with modern logging programs.
Depth and Measurement Errors in Core Logging
Accurately measuring and recording the depth of core is fundamental to core logging. However, this seemingly straightforward task can quickly become complex. When core is broken or pieces are lost during drilling, it can be difficult to reconstruct how the material fits together within the tray. In highly broken ground, core spreading may occur, increasing the apparent length of core within the box. Conversely, core compaction can result in a shorter measured interval. These errors may compound progressively downhole, reducing confidence in the recorded depth and the interpreted locations of geological and geotechnical features.
How Datarock Improves Geotechnical Data Consistency and Reliability
Standardised Logging for Consistent, Deposit Wide Data
Geotechnical logging is inherently subjective, with interpretations varying between individuals due to differences in experience, training, judgement, and site familiarity. Datarock’s machine learning models provide a consistent single source of truth across all drilled core, applying the same logic and parameters to both historical and recent drilling. This enables legacy core photographs to be assessed against a more consistent standard that is closely aligned with the site’s current logging practices.
Depth Registration: Accurate Downhole Positioning of Core Data
Understanding the depths of features and measurements down a drill hole is critical to being able to use that data. Datarock provides this through its process of Depth Registration.
Datarock Core uses a methodical and repeatable depth registration process to assign accurate depths to core imagery. It identifies coherent and incoherent rock, core blocks, empty space and broken material, then combines provided box depths, OCR-detected depth marks, core loss data and compaction factors. These inputs are applied consistently across the dataset to interpolate depths between fixed tie points, producing reliable depth outputs for every pixel of core. Any depth registration anomalies outside predefined tolerances are automatically detected and flagged through Depth Health checks, providing an additional layer of quality assurance.
When core is highly fractured it can spread out or “Expand” or get clumped together or “Compact” when put into the core tray. By using the ground-truthed points marked on the core and core boxes, Datarock can calculate the expected length of core in a row. If there is too much rock or too little, an Expansion or Compaction factor is applied to the broken core to compensate for it.
With this method Datarock provides confidence in the depth measurements of all features and calculations throughout the hole.
Flexibility:
Datarock offers the flexibility to customize project settings to be in line with your site conditions.
The equivalent fracture counts can be configured by editing the values for Jigsaw, Broken and/or Rubble to better match your own equivalent fracture counts. Likewise, the JSA thresholds can be configured to any percentage.
Project Settings showing Fracture detection settings and Joint set analysis setting
This flexibility and control enables users to calibrate their results based on ground-truthed data, including logging and site observations.
For example, tolerances can be increased for historical core, where the mark-up could be poor compared to modern standards. Tolerances can be reduced for high precision work where the Depth Registration must be within a specific range.
To improve the performance of the Depth Registration models control is given to turn on or off the Optical Character Recognition and Core Block Recognition. This gives the options to calculate the core depths based on, all depth marking, core marking only, core blocks only, or to ignore all core box markings and base the depths wholly on the start and end box depths. This gives you control over the Depth Registration model to take into account the quality of the mark-up in your photos.
Validating Model Outputs: Ground-Truthing Geotechnical Data
QA/QC Methods for Comparing Predicted vs Logged Data
Ground-truthing ties your direct observations and logged measurements to the data predictions provided from the platform.
There are a few ways within the platform to do this:
- Visualising the data in the Results tab
- Graphs and charts presented in the Results tab show the comparison of the predicted data and the ground-truthed data
- Displaying the data in graphs to check the relationship between predicted and ground-truthed logs
- Cross hole validation to look at the relationship between different holes
There are also some things you can do outside the platform:
- Displaying the data in graphs to check the relationship between predicted and ground-truthed logs
- Cross hole validation to look at the relationship between different holes
Visualising your results
One of the simplest ways to validate your results is to look at the pictures of the core and compare them with your geotechnical logs.
The Datarock Results page displays the full core images of a drillhole. Overlays show the polygons for each of the products, allowing you to see the decisions which have been made for each fracture measured, or which pieces of core were included in the RQD.
Depth markers are included in the image, making a comparison easier between the images and your logs. It is expected that you would be able to identify the same fractures identified in the image at the same depths as in the ground-truthed logs. Likewise the areas of incoherent rock and coherent rock should align with the areas measured for RQD in the ground-truthed logs, leading to a similar RQD result.
example: detection and classification of fractures showing Drilling Breaks, Measurable Fractures, Mechanical Breaks, Complex (jigsaw) Fractures, Core Blocks and Row Ends
example: RQD model categorisation showing polygons of Coherent rock greater than 10cm, Coherent rock less than 10cm, Incoherent rock, Rubble, and Brocken rock
Charts and Graphs
Datarock charts:
Datarock provides charts within the Results Tab, to display the geotechnical parameters at standard intervals and user defined intervals. The user defined intervals are the ground-truthed results uploaded as meta data.
These charts include:
-
RQD graph by 1m interval and user defined intervals
-
Fracture graph by 1m interval and user defined interval
-
RQD table by 1m interval
-
RQD table by user defined intervals
-
Fracture location table
-
Fracture frequency by 1m interval table
-
Fracture frequency by user defined interval
We would expect to see a good correlation between the Datarock predictions and the ground-truthed results. A good correlation would be the graphs showing similar trends. We do not expect to see the exact same numbers being displayed; however we expect the overall trend to be the same.
The numbers of fractures logged by the Datarock Predictions can be higher than the ground-truthed results, because the Predictions can be more sensitive and log more fractures in some cases, especially where the rock is overly fractured. However, in cases where fractures are not visible in the core photograph, Datarock can not log these fractures, so a lower number of fractures may be recorded.
Differences in RQD may be a result of incoherent rock which has the appearance of coherent rock in the images.
Example: RQD chart showing the Datarock predictions and the user defined ground-truthed measurements
Example: Fracture chart showing the Datarock predictions and the user defined ground-truthed measurements
Onboarding Reports
Before any Datarock model is deployed for customer use, it undergoes a rigorous validation process conducted by our technical team. Detailed onboarding reports are provided to document the testing, quality assurance, and performance assessments completed to verify that the model performs reliably based on the site's geological, geotechnical and operational conditions.
These onboarding reports use the charts and graphs with in the Datarock platform along with additional checks, including a comparison of predicted RQD results and ground-truthed RDQ results and a check of the internal consistency of the geotechnical results.
RQD Scatter Plot Comparison
This plot shows the relationship between the ground-truthed RQD measurements and RQD predictions provided by the Datarock model.
It is expected that that the trend of this data should lie along the diagonal line which represents a 1:1 relationship between predicted RQD and ground-truthed RQD.
The coloured heat map shows that this is the case for this dataset. Outliers can be examined to determine if they are the results of logging or transcription errors, or an error in the Datarock model.
example: RQD scatter plot showing good correlation between the Datarock predicted RQD and the user defined ground-truthed measurements
Internal Consistency
Comparing geotechnical parameters to each other is another way to assess the results. By looking at the spacing between fractures (Fracture Spacing) you can get an idea of the lengths of intact rock. Lengths or rock greater than 100mm are used to calculate RQD, therefore it follows that that as the fracture spacing increases the RQD values would show an increase. Low fracture spacing values should correspond to low and highly scattered RQD values, reflecting heavily fractured, poor-quality rock.
By assessing this internal consistency, we can validate the parameters against each other to check that the models, while being run individually, are producing results which are working together to describe the rock mass.
Datarock assesses this using a graph based on the work of Bieniawski (1989), the mean discontinuity spacing is shown on the horizontal axis with a logarithmic scale, and RQD is shown on the vertical axis from 0 to 100% . It is expected that the data points should be following the trend line shown in yellow.
example: graph based on Bieniawski (1989) showing the good correlation between the mean discontinuity spacing and RQD
Other Methods of Validation
Cross hole validation
Boreholes drilled in close proximity would generally be expected to show similar geotechnical characteristics at comparable depths. While local geological variability means this will not always be the case, some degree of correlation should typically be observable between nearby holes.
Plotting the borehole data in 3D space is an effective way of visualizing these relationships. At present we do not have a 3D modelling function within the Datarock platform, however, the results can easily be exported from the platform and imported into an external 3D modelling program.
By applying colour coding to the geotechnical parameters, it is possible to visualise trends in the data, identify correlations between nearby holes at similar depths, and spot outliers in the data.
Probability Columns
Most Datarock products calculate a probability score for each prediction. This score represents the model's confidence in the assigned result.
The probability score can be used as part of the validation process. When reviewing individual samples, it provides an indication of how confident the model is in each prediction.
When visualising data in 3D modelling software, probability thresholds can be used to filter prediction results. Several of our models, including Fracture, Geotechnical Weathering Class, Unique Classification and Discing (referred to as Confidence in Discing results) assign each prediction a probability score that represents the model's confidence in the prediction. These scores can be used to control which predictions are displayed, and therefore help to query the data. For example, you may choose to display only predictions with probability scores greater than 0.2, 0.5, or 0.75, depending on the level of certainty required for your application. Increasing the probability threshold, in general, reduces the number of displayed predictions but generally increases the reliability of the results. Conversely, lowering the threshold retains more predictions, improving coverage at the expense of including more uncertain predictions.
example: of 3D visualisation of boreholes showing RQD% (left) and Fracture Spacing (right)
Continuous Validation
It is important to note that QA/QC of geotechnical results is not a one-off exercise completed at model delivery. It is an ongoing process of verification that should be maintained throughout the life cycle of a drilling or mining program.
This ongoing validation helps confirm that the model continues to perform effectively and can also highlight changes in geotechnical conditions across the site.
Periodic manual logging of selected holes to obtain ground-truthed data remains the most effective way to validate Datarock predictions. The frequency of this validation is site-specific and depends on geological conditions, drilling schedules, and the availability of geotechnical personnel. However, a starting benchmark of approximately 1 in 10 holes is often appropriate.
Regularly uploading ground-truthed data as metadata within the platform enables efficient use of built-in validation tools. If a shift in model reliability is observed, users are encouraged to contact their Datarock Customer Success Manager or support team so that potential causes can be investigated and model updates considered if required.
Conclusion
Geotechnical models - and the design decisions that depend on them - are only as reliable as the data they are built from. Datarock provides geotechnical data that supports the development of these models, with a strong emphasis on transparency, traceability, and auditability to ensure confidence in the results.
To maintain data quality, validation between Datarock predictions and ground-truthed logging is essential. The methods outlined in this paper provide a foundation for establishing robust QA/QC procedures and assessing the reliability of digital geotechnical datasets.
By delivering results more quickly than traditional manual logging, Datarock enables geotechnical engineers to spend less time collecting data and more time interpreting results, developing models, and optimising designs.
As the industry continues to adopt data-driven workflows, rigorous validation and QA/QC processes will play a critical role in building confidence in digital geotechnical data and unlocking its full potential.