QUALITY CONTROL OF 3-D SUBSURFACE MODELS

TNO – Geological Survey of the Netherlands (GSN) develops and maintains four national-scale 3-D subsurface models. Three of these models are part of the ‘BRO’, a government database managed by GSN that holds all subsurface data acquired with public funds. The models in the BRO are: (1) GeoTOP, a voxel model with stratigraphical and lithological attributes up to a depth of 50 m below MSL, (2) DGM, the national stratigraphic framework model of the Neogene and Quaternary, and (3) REGIS II, our national hydrogeological model attributed with hydraulic conductivities.

Public bodies are legally required to consult both the data and the models in the BRO when making decisions related to the subsurface. This legal requirement is expected to lead to an increase in the use of the models. It also imposes more stringent demands on our models in terms of their quality.

In this presentation we will show how formal quality control (QC) was implemented in our modelling workflows, what lessons we have learned in the past 5-6 years, and what future developments we expect from modelling in the context of the BRO.

Our QC efforts up to now have been primarily focused on intrinsic quality: expert geologists (not involved in the modelling process) are systematically checking cross-sections, maps and 3-D views of the final model. They subsequently report all defects, artefacts and geologically implausible configurations as findings for the modelling team to address.

An important lesson learned is that QC should not be limited to the final product, but should start early in the modelling process by performing QC on all major modelling components. Another way to detect defects early in the modelling process is to work ahead, i.e. to construct a preliminary final model from intermediate model components.

Due to the increasing user community we foresee a shift from evaluating intrinsic quality towards assessing the usability of the models in applications: will the model deliver the right answer to the users’ needs? Another change imposed by the BRO is that we will more regularly update our models. QC will then focus more on the changes made with respect to the previous version: are these changes reflecting new data and knowledge acquired since the last release and do they correct for any errors reported by users?

We can conclude that QC should be an integral part of the modelling process, delivers the opportunity to improve our models in a systematic manner and is more effective when applied early in the modelling process.