Groundwater Knowledge and Integration System (GKIS)

The Challenge 

The mapping of areas with potential for sustainable groundwater extraction is mostly qualitative, based on expert knowledge. Quantitative approaches are often statistical or data driven approaches, with limited system understanding. Accounting for uncertainty in groundwater potential estimates is not common. 

The intended use of groundwater is seldom made explicit; volume of water required, over what period, with which quality. Current groundwater prospectivity maps are only representing groundwater occurrence or yield. 

The groundwater group has a strong capability in modelling and analysis, but scripted, transparent and repeatable workflows are not yet standard practice, nor is digital, interactive delivery of our research. 

Our response (Opportunity) 

We further refined the Groundwater Knowledge Information System (GKIS) algorithm, developed in the Goyder Institute’s G-FLOWS-III project and Deep Earth Imaging Future Science Platform. This project allowed groundwater group researchers to further develop capability in coding and digital delivery. 

GKIS calculates the probability on a raster that groundwater can be extracted sustainably (in terms of volume, rate and quality), using analytic solutions to groundwater flow equations, using probabilistic estimates of hydrogeological properties, which can be based on international literature, local expert knowledge, geophysical surveys or measured data (Figure 1). 

Compiling datasets to inform probabilistic estimates of hydrogeological properties remains challenging. The project examined the availability of national datasets and how to access such information from webservices in scripts, the use of unsupervised spatio-temporal clustering and classification of large Synthetic Aperture Radar data to inform hydrogeological conceptualisation and a literature review of methods of regional scale assessment of surface water – groundwater connectivity that do not rely on hydraulic information. 

The equations underpinning the GKIS prospectivity were revised and updated, including functionality to assess streamflow depletion and radius of influence of pumping wells. The equations are incorporated in an object-oriented Python package. 

The GKIS algorithm is further extend with new functionality: a random borefield generator and stochastic water balance. Water resource assessment are generally not focussed on prospectivity for a single bore, but on the sustainable yield for a region. The borefield generator estimates how many bores need to be budgeted for to attain a given yield across a region, while the stochastic water balance estimator provides an internally consistent ensemble of groundwater balances, which provides context for the predicted yield. 

The probability for sustainable groundwater extraction depends on user-defined thresholds of salinity, pumping rate and duration. The Streamlit package was used to create an interactive, web-based application to interrogate GKIS results. 

Results (Impact) 

The results of the project are summarised based on the individual tasks.  

  • National datasets 

National hydrogeological datasets are outdated, mostly still based on geological maps from the 1980s. As part of the project, we engaged with Geoscience Australia. Their National Groundwater Systems project is on track to deliver relevant nation-wide groundwater datasets. The exception is recharge, the best national estimates are delivered by Russell Crosbie’s strategic work. 

The project developed Jupyter notebooks with examples of interrogating spatial datasets available from web services with Python scripts. 

  • Spatio-temporal clustering of SAR data 

Unsupervised spatio-temporal clustering of SAR data was tested on a dataset of the Burdekin, examining soil erosion. The size of the dataset was moderate for SAR applications, but available clustering and classification packages caused memory issues on this size of data. More CPU power is unlikely to resolve the issue. It will require further research into computationally efficient spatio-temporal clustering algorithms. 

This research activity did have a positive outcome as it increased collaboration between groundwater SAR researchers and geomorphology researchers. 

  • SW-GW connectivity & streamflow depletion 

The research activity on assessing surface water groundwater connectivity based on regional-scale, non-hydraulic approaches was hampered by researcher availability and COVID related cancellation of a crucial workshop. 

The project did however create a dedicated Python package that compiled a large number of analytic solutions for calculating streamflow depletion due to groundwater pumping, capturing over 2 decades of legacy CSIRO research and making this available to the international research community. 

  • GKIS Python package 

The GKIS algorithm has been extended and implemented in an object-oriented Python package. This operationalisation of the workflow allows for rapid deployment of the algorithm to new areas, provided the conceptualisation is similar and probabilistic estimates of hydrogeological properties are available.  

The Python package has been tested on two regions, the APY Lands in SA and the Beetaloo GBA region in the NT (Figure 2). The borefield generator demonstrated that the median total yield of drilling 100 bores in the Beetaloo region would be around 12 GL/yr (Figure 3). 

The stochastic water balance algorithm is in prototype stage and documented in a Jupyter notebook for further development. 

  • Interactive web application 

The Streamlit package was successfully used to create an interactive web-based application to interrogate the results for the APY Lands (Figure 4). The experience of applying this user-friendly package has spurred the development of other interactive web applications in projects other than GKIS, lead by Trevor Picket and Chris Turnadge. 

  • Capability development 

The project provided room for experimenting and a safe space for failure. This allowed many of the researchers involved to learn new skills in coding and digital delivery. 

Figure 1 GKIS flowchart: 1 – input information, 2 – establish probability distributions, 3 – generate ensemble of metrics, 4 – user defined threshold values, 5 – calculate probability from ensemble (after Peeters et al. 2022) 

Figure 2 Application of GKIS to water requirement assessment for unconventional gas development in the Beetaloo region (NT) 

Figure 3 Histogram of the number of successful bores (out of 100 drilled) and cumulative distribution of total borefield yield 

Figure 4 Screenshot of streamlit web app of APY GKIS results 

For more information, please contact luk.peeters@csiro.au.