A model of imperfection

If you're planning to excavate Australia's biggest coal mine in the middle of an arid region in Central Queensland, you must obtain a mining lease (under the Mineral Resources Act), an environmental authority (under the Environmental Protection Act) and approval under the Environment Protection and Biodiversity Conservation Act. The approvals process includes a Groundwater Impact Assessment, which is part of an Environmental Impact Statement (EIS). This is because mining involves the dewatering of the rock strata in which coal seams occur, and the use of large volumes of water in the production of coal, for washing and dust suppression for instance.

This week in the Land Court, the Adani Mining Pty Ltd vs Land Services of Coast and Country (LSCC) case looked in detail at Adani's groundwater model assessing the impact of the Carmichael mine's drawdown on regional water resources, required by government regulators for a project of this size.

There are guidelines for modellers. There used to be three classes of models based on low, medium and high degrees of complexity. But this classification has been redefined in terms of levels of confidence in the objective, in particular in this case the assessment of the potential impacts of drawdown on nearby groups of springs. What this boils down to is whether or not the clay Rewan formation (see image at top of page), a largely impermeable stratum, offers sufficient protection for aquifers (Clematis sandstone and Dunda Beds) above the Rewan that supply the springs, wells and bores used by landholders, from the dewatering of the coal-bearing strata (Colinlea sandstone) beneath the Rewan.

Unsurprisingly, what you get out of a model is dependent on what you put in, and how the modeller tweaks the parameters.

A numerical model is based on a conceptual model. Physical attributes of the latter have to be formulated in mathematical terms and used to build a myriad three-dimensional cells. Certain simplifying assumptions have to be made in this process in order to enable values to be inputted; for example, that the density of groundwater is constant. The modeller adjusts the features of cells to replicate actual measurements where they exist. Key features in this model were recharge rates; hydraulic conductivity and hydraulic head (the resting level of groundwater, or the height to which water will rise in a bore); and the storage parameters of aquifers. The boundaries on the ground that confine the model were also significant.

LSCC's hydrogeologist Dr Webb had already explained to the Court in his conceptual hydrogeological model that faults might be present in the otherwise low-conductivity Rewan, through which water might flow between the Colinlea and the Clematis. No faults were included in the model, however. Adani's modelling expert argued that, since there is no evidence of faults, they should rightly not have been included because where would you put them? His opposite number suggested that faults should be in the model if only to disprove their effects. They both agreed that hydraulic conductivity values seemed to be too low, but disputed the head values. There was particular concern that the positioning of the western boundary would reduce the likelihood in the model's results of impact on the Great Artesian Basin (GAB).

I felt in the Alpha case in 2013, and again in this one, that the GAB is the elephant in the courtroom. It's overwhelming hydrogeological presence and its importance hydraulically for farming and settlement in vast swathes of central and eastern Australia should ensure that it cannot be ignored. All proposed development projects should, in my humble opinion, have to pay specific attention to this feature in approval applications.

But back to modelling. Even greater anxiety was expressed about the absence of springs in the model. Adani's expert explained that springs were in the 'too difficult' box; that they were too computationally intense, though not impossible to include. The survival or not of the unique Doongmabulla Spring complex – just 8 kilometres from the mine site – had occupied the court for days, so perhaps the modellers' brief should have been more specific.

The degree to which a groundwater model reflects a real situation depends on the accuracy of the input data and the parameters – the features or measurable factors that define the system. Once data has been inputted, you have a base model. Parameter Estimation (PEST) software can be combined with the intuition of an experienced modeller. Subsequently, inputters graduate a model's parameters to allow for irregularities. Model calibration is a process of adjustment of parameters within acceptable margins of uncertainty about the system to obtain the most likely potential outcomes that satisfy predetermined criteria.

The model produced in this case was a steady state model: that is, it presumed properties remained unchanged through time. The alternative, transient calibrations, would have allowed for variations in rainfall, for example, or mining activity. Both modelling experts agreed this would have been desirable.

The model outcomes were analysed in a report for the regulators who ultimately decide whether or not the Carmichael mine proceeds. That report was peer reviewed. The experts debated at length in Court the report's omissions, inaccuracies and misconceptions, not to mention flaws in the modelling process itself. There were underlying concerns about the lack of on-the-ground data; the absence of the 'linear superposition' of potential effects on the GAB; the range of the orders of magnitude used in sensitivity analysis; and the limitless number of particular values in an uncertainty analysis, which elicited the rather disturbing question from Senior Counsel Holt, 'Does this mean all models are wrong?'

There certainly seemed to be a pervading impression that no model can provide anything better than a best estimate.

Perhaps I am wary of an approvals methodology that does not err on the side of caution if there is sizeable doubt about the conservation of water resources.

Queensland's Co-ordinator General, who recommends – on the basis of the EIS and subsequent SEIS (Supplementary Environmental Impact Statement) – to the relevant state government ministers whether or not they should grant approvals to mining proponents, received advice from the Independent Expert Scientific Committee (IESC) on Coal Seam Gas and Large Coal Mining Development. This is what they said of Adani's initial groundwater model:

The Committee is not confident that the proponent's groundwater model will be able to accurately predict responses to perturbation of the groundwater system arising from the proposed mine. The Committee does not have confidence in the model's predictions for the potential groundwater impacts to the Doongmabulla and Mellaluka Spring Complexes and the Carmichael River.

I rest my case.

The image at the top is Dr John Webb's conceptual geological model used in the Groundwater Modelling Report for the Carmichael Mine