2.6.2.8 Uncertainty analysis

Summary

In the uncertainty analysis, the uncertainty in the model parameters is propagated through the analytic element model and the alluvial MODFLOW models and constrained with the available observations to obtain ensembles of predicted maximum difference in drawdown (dmax) between the coal resource development pathway (CRDP) and baseline, due to additional coal resource development and year of maximum change (tmax) at the receptors. In addition to that, maps of the probability of drawdown due to additional coal resource development (additional drawdown) exceeding 0.2 m are presented for the surface weathered and fractured rock layer and for the alluvia of the Avon and Karuah.

Normal prior distributions are specified for most parameters, either in native or log space, with a mean equal to the centre of the range sampled in the design of experiment. The variance is chosen such that 99% of the probability mass is within the sampled range. A covariance is specified between the parameters controlling hydraulic conductivity and storage and between parameters controlling horizontal and vertical fault hydraulic properties. For the parameters controlling the decrease with depth of hydraulic properties and the fault hydraulic conductivity, the mean is not chosen to be equal to the centre of the sampled range to ensure the hydrologic change is overestimated rather than underestimated.

The prior parameter distributions of the analytic element model are constrained with the maximum coal seam gas (CSG) water production rate which resulted in posterior probability distributions for dmax and tmax indicating that: (i) the effect on dmax is highly localised around the mine pits; (ii) it is unlikely to have drawdowns due to additional coal resource development in excess of 1 m, except in close proximity of the mine pits; (iii) the largest dmax values are attained within or shortly after the production life of the coal mines and CSG field. Smaller additional drawdowns, further away from the centre of the development activity, are realised at later times, where the smallest noticeable drawdowns due to additional coal resource development are not fully realised within the simulation timeframe; and (iv) in the modelled drawdowns, the effect of CSG production and the presence of faults and fractures cannot be distinguished from the effect of coal mining in the Gloucester subregion.

The prior parameter distributions for both alluvial MODFLOW models are constrained with historical estimates of the water balance. The resulting posterior ensembles of predictions indicate that: (i) hydrological change is limited to the immediate vicinity of coal mines and (ii) it is very unlikely to have drawdown due to additional coal resource development in excess of 1 m in the alluvial aquifers; and that (iii) it is very unlikely that the drawdown will cause the groundwater levels to drop below the drainage base of the stream network.

The qualitative uncertainty analysis lists the main model assumptions and choices and discusses their potential effect on the predictions. The model choices with the greatest perceived potential impact on the predictions are related to the implementation of the coal resource development pathway (CRDP). Other model assumptions, such as the hybrid modelling approach, the choice for drainage boundary to represent the stream network and the length of the simulation period are shown to be conservative choices (i.e. the hydrological change is overestimated rather than underestimated).

The uncertainty analysis is divided in two sections: quantitative uncertainty analysis and qualitative uncertainty analysis. As outlined in submethodology M09 (as listed in Table 1) for propagating uncertainty through models (Peeters et al., 2016), quantitative uncertainty analysis is the numerical propagation of the uncertainty in parameters defined in the parameterisation section through the model chain into an ensemble of predictions. Qualitative uncertainty analysis on the other hand is a structured discussion and scoring of the model choices and assumptions in function of their impact on predictions.

Last updated:
5 November 2018
Thumbnail of the Gloucester subregion

Product Finalisation date

2018
PRODUCT CONTENTS