Using model-data fusion techniques delivers more credible predictions around potential groundwater impacts. This strengthens Australia’s capacity to make decisions in relation to investment, resource development and management across the mining, energy, and agricultural sectors.
Increasing the reliability of model predictions is best achieved if different types of observations are used to develop the models
Data fusion is the process of integrating multiple data sources to produce more consistent, accurate, and useful information than that provided by any individual data source.
As the complexity of water-related problems grows, the need for tools that are able to capture those complexities while still being useful and practical in a management context also increases. Groundwater models and decision support systems that make use of such models need to produce a demonstrable level of accuracy (when estimating past conditions) or confidence (when predicting future conditions). Developing models that instil trust, whether conceptual or numerical, requires many different types of data that each constrain a different part of a model. The challenge is in identifying, for a given management question, what the best combination of data is that will deliver a model that behaves appropriately.
We apply model-data fusion techniques to develop better conceptual and numerical models
Model development starts with the conceptualisation of the system that needs to be studied: our conceptual models are robust for they integrate traditional hydrogeological information, airborne and surface-based geophysics, and environmental tracers, as a minimum. Geophysical information is typically used to generate the three-dimensional skeleton (geometry of layers) of an aquifer while boundary conditions such as recharge are derived from environmental tracers.
We use remote-sensing data to estimate evapotranspiration from vegetation. Hydrogeological properties are generated from traditional field-based or lab-based testing, or may be obtained from borehole geophysical observations or certain environmental tracers. Numerical models derived from robust conceptual models are more likely to be more reliable. Furthermore, to capture conceptual model uncertainty, multiple models are considered and tested using a variety of data.
More robust conceptual and numerical models
The value of integrating various types of data for improved system conceptualisation was demonstrated in the Peel Integrated Water Initiative (PIWI) in Western Australia. The project identified recharge rates and connectivity and fluxes between different aquifers. This involved interpretation of data from an airborne electromagnetic survey (AEM), a seismic survey, environmental tracers, lithological data and climate data.
The South Australian Goyder Institute’s G-FLOWS Projects lead by CSIRO extended our understanding of the use of regional geophysical data supported by traditional hydrogeological information and tracer data in large scale aquifer characterisation and conceptualisation across the Musgrave province. The focus was on fine-scale investigations commensurate with the scale of palaeovalley aquifers in order to better constrain the available groundwater resources to support remote communities, agriculture and industry development. The methods and approaches developed by CSIRO for water resources assessments underpinned by robust conceptual models are readily transferable and scalable to other basins and conditions.
The benefit of using of non-traditional data types for numerical models testing was recently demonstrated in a modelling study that combined remote sensing and geophysical data (seismic refraction) to evaluate if sinkhole-like depressions acted as recharge features. The approach allowed to eliminate models that showed inconsistencies with the geophysical and/or remote sensing data.