CALL FOR PROPOSALS:

ORGANIZERS

  • Harvey Thorleifson, Chair
    Minnesota Geological Survey
  • Carrie Jennings, Vice Chair
    Minnesota Geological Survey
  • David Bush, Technical Program Chair
    University of West Georgia
  • Jim Miller, Field Trip Chair
    University of Minnesota Duluth
  • Curtis M. Hudak, Sponsorship Chair
    Foth Infrastructure & Environment, LLC

 

Paper No. 5
Presentation Time: 3:00 PM

ASSESSING THE IMPACT OF PROGRAM SELECTION ON MODEL ACCURACY


MACCORMACK, Kelsey E., Alberta Geological Survey, Alberta Energy Regulator, 402 Twin Attria Building, 4999 - 98 Avenue, Edmonton, AB T6B 2X3, Canada, BRODEUR, Jason J., School of Geography and Earth Sciences, McMaster University, 1280 Main Street West, Hamilton, L8S 4K1, Canada, EYLES, Carolyn H., Integrated Science Program & School of Geography & Earth Sciences, McMaster University, 1280 Main Street West, Hamilton, ON L8S 4K1, Canada and PARKER, Beth L., G360 Centre for Applied Groundwater Research, University of Guelph, 50 Stone Rd E, Guelph, ON N1G 2W1, Canada, kelsey.maccormack@aer.ca

The use of 3D subsurface geological models has increased steadily in recent years, and so too has the number of software programs catering to these applications. Most of these programs offer very similar ensembles of algorithms that can be used for interpolating data. Several studies have evaluated the effectiveness of different algorithms for producing accurate models. However, little research has been done to assess the impact of program selection on model accuracy. It is commonly assumed that data modelled using an algorithm in one program would produce identical results if modelled using the same algorithm in another program, providing that both programs were supplied with identical input datasets. This assumption is tested by evaluating model outputs using algorithms from different programs and by assessing the impact of program selection on model accuracy.

To assess the impact of program selection, two commonly used algorithms (inverse distance weighting and ordinary kriging) from five different software programs (ArcGIS, ROCKWORKS 2006, VIEWLOG, MATLAB, and PETREL) were used to interpolate identical datasets extracted from 4 synthetic grids. The output models produced by each program were compared back to the synthetic grids to determine how much the interpolated results deviated from the original data points. The purpose of this study was not to show that any one program is ‘better’ than another, but to identify the degree and nature of the differences between the model outputs from each program. Additional analysis was done to determine the cause of the discrepancies between each program.

Initial results indicate that the model outputs from each software program are inconsistent and this variability can have a statistically significant impact on model accuracy. In most 3D subsurface investigations, a substantial amount of time and effort is spent collecting and analyzing data, as well as assessing data parameters to ensure the most accurate model possible is produced. The results of this study indicate that program selection should also be seriously considered as a possible source of model uncertainty, especially when modelling complex subsurface geological environments or interpolating with clustered data.

Meeting Home page GSA Home Page