Oral Presentation Australasian Groundwater Conference 2017

Permeability modelling in the Surat Basin using artificial neural networks (#168)

Gerhard Schoning 1 , Daan Herckenrath 1
  1. Department of Natural Resources and Mines, Brisbane, QUEENSLAND, Australia

The Office of Groundwater Impact Assessment (OGIA) is responsible for assessing the cumulative impacts of groundwater extraction by large-scale Coal Seam Gas (CSG) activities in Surat Cumulative Management Area (CMA). For this area a regional groundwater flow model has been developed and is being periodically updated and revised to predict the cumulative impacts of groundwater extraction for CSG production purposes.  Initial parameterisation of this model is undertaken using stochastic lithology permeability models and numerical permeameters which translate the available small-scale permeability and wireline log derived lithology data into stochastic distributions of regional-scale permeability. In order to refine and improve the existing probabilistic lithology permeability models, OGIA has developed a number of Artificial Neural Networks (ANN’s), an increasingly popular technique for extracting information from large datasets.

Using wireline logs and permeability data for over 4,000 CSG wells OGIA has developed several ANN’s to extract complex data structures and predict permeability for coal and non-coal lithologies for various stratigraphic units in the Surat Basin. OGIA has further evaluated both the neural network architecture and different regularisation settings to optimise the ANN’s performance.

The optimised ANN’s yielded better results than permeability estimates obtained with conventional petrophysical methods such as the K-Timur-method and porosity-permeability relationships and outperformed multivariate linear and polynomial regression. By using gamma ray, density, depth-of-burial and location as input features, the models could on average explain 70% of the variance in core and DST permeability datasets not used to train the model. The current ANN's have subsequently been used to generate high resolution permeability profiles at each well which will be used to parameterise OGIA’s next generation local and regional groundwater flow models. These results suggest that ANN’s are a powerful tool to predict permeability where large data sets of borehole permeability measurements and geophysical logs are available.

  • We are offering awards for Career and Early Career presentations and posters. Please indicate length of time since highest degree completed.: 0 to 5 Years or currently studying