Opportunities to reduce predictive uncertainty

An important outcome from the Assessment is identifying the main sources of uncertainty and the opportunities for improving regional-scale groundwater modelling in the Hunter subregion. The sources of uncertainty broadly relate to data, model implementation and representation of the baseline and coal resource development pathway (CRDP). These can be inter-related: for example, the quantity and quality of data available will influence decisions about the geometry of the model and representation of processes.

The sensitivity analysis in Section identified the model parameters that had the largest influence on the groundwater model outputs of interest. Groundwater level predictions were found to be most sensitive to the drainage level of the river network (d_riv), hydraulic properties of the geological layers (Kh, K_lambda, KvKh) and the recharge multiplier (RCH). The drawdown, however, is most sensitive to the hydraulic properties (Kh, K_lambda, KvKh), porosity (ne) and variation in mine pumping rate (Q_mine). The drawdown predictions are much less sensitive to the drainage level or recharge. The predicted surface water – groundwater flux and change in this flux are most sensitive to riverbed conductance (C_riv) and hydraulic properties. Where feasible, opportunities to reduce predictive uncertainty of drawdowns are best directed to constraining the hydraulic properties.

Many of the assumptions and modelling choices, discussed in Section, were influenced by a lack of region-wide, good-quality data to better characterise the lithology and hydrogeology across the whole model domain and constrain model predictions: the available hydraulic conductivity data are not correlated with the geological layers and the lithology of the geological model; the drainage level (channel depth) along the stream network is generally poorly defined; acceptable historical groundwater level data to constrain the model are sparse; and the relative contribution of groundwater to streamflow is generally not known.

The hydraulic properties in the model do not vary with lithology (except that the alluvium has higher hydraulic conductivity and porosity than the underlying rock), therefore, there are no aquitards (at any scale) represented within the model layering. Hydraulic conductivities are assumed to decrease smoothly with depth (rate of decay is controlled by K_lambda). The model’s predictive uncertainty would be reduced through a more accurate specification of the hydraulic properties of the groundwater system. This includes both the in situ hydraulic conductivity and porosity, and an improved characterisation of the hydraulic enhancement process caused by mining-induced strata deformation (see Section More information about the hydraulic conductivity measurements (e.g. measurement method, lithology, depth, fracturing; presence of aquitards) would assist in classifying these data. Better resolution of the geological and lithological layers in the geological model might contribute to a better correlation of hydraulic conductivity measurements with lithology – that is, if better resolution permitted the observation data to be stratified more confidently by lithology.

Changes in the hydraulic properties used to represent the goaf in the model do not include changes to porosity (storage) and water head due to the volumetric expansion of the overburden. Inclusion of these changes will naturally decrease the initial slug of water (see Section, but will play no role in the long term behaviour of baseflow.

As discussed in Section, relatively few high-quality, long-term observations of groundwater levels are publically available. Many historical observations of groundwater level were not used for constraining the model as the observations did not have reliable coordinates, and/or the record was taken at the time of drilling and considered unlikely to represent the true groundwater level. A more detailed assessment of the quality of these data and field verification of bore locations are needed if more useful data are to be extracted from the existing dataset. Mining companies collect piezometer data, which could also help to constrain the groundwater levels locally.

The sensitivity analysis indicated that groundwater level and the surface water – groundwater flux are most sensitive to the riverbed elevation specified in the groundwater model through parameter d_riv. However, the predictions of drawdown (dmax) due to the additional coal resource development (i.e. maximum difference in drawdown between the coal resource development pathway (CRDP) and baseline) are not sensitive to this parameter. Unless the uncertainty in riverbed elevation is reduced independently – for instance, through surveyed river network information and higher spatial model resolution – additional or improved observation datasets will not greatly reduce the predictive uncertainty in dmax.

Selected percentiles of the total streamflow are used to constrain the surface water – groundwater flux in the model. There is a great opportunity to improve the regional-scale estimates of surface water – groundwater flux through a combination of baseflow separation, salt balance and environmental tracers.

Evapotranspiration in the groundwater model was represented by PET and an extinction depth which was assumed to be related to vegetation height. The extinction depth parameter was not included in the uncertainty analysis as it was deemed unlikely to affect dmax because it would have similar effects on the baseline and CRDP and therefore largely cancel out in estimation of dmax. The assumption was revisited using a one-at-a-time sensitivity analysis when model results showed differences in evapotranspiration between baseline and CRDP (see Section This analysis supported the initial assessment that model predictions were not sensitive to the extinction depth parameter. Nonetheless, better representation of local-scale influences on evapotranspiration would improve local-scale predictions of groundwater levels, and potentially yield a more tightly constrained set of model parameters when compared with historical water-level data.

The predictive uncertainty would also be reduced by a more accurate representation of mining progression over time. This includes expanding the mine footprint area over time, rather than using the maximum footprint area from the first day of mining, and hence phasing in the hydraulic enhancement over the life of the mine also. Some of these improvements are relatively easy to implement, and were not undertaken due to operational constraints of the BA. During the period when the model was under construction, the coal resource development pathway changed numerous times due to, for example, mining companies lodging, modifying or withdrawing environmental impact statements.

Last updated:
18 January 2019
Thumbnail of the Hunter subregion

Product Finalisation date