2.6.2.1.2 Groundwater numerical modelling

Page 3 of 60

In the Namoi subregion, the groundwater model has been developed using the MODFLOW-USG code (Section 2.6.2.3). To be fit for the purposes of a BA, the groundwater model needs to satisfy the criteria listed in Table 3. The remainder of this section discusses each of these criteria with regard to the numerical modelling approach undertaken in the Namoi subregion.

Table 3 Assessment of groundwater numerical modelling approach in the Namoi subregion

Fit-for-purpose assessment criteria

Components

1. Prediction of hydrological response variables

Probabilistic estimates of hydrological change at model nodes

Integration with receptor impact modelling

Integration with surface water numerical models

2. Design and construction

Modelling objectives stated

Model confidence level

Modelling approach

3. Integration with sensitivity and uncertainty analyses workflow

Parameterisation

Convergence

4. Water balance components

Conceptual model agreement

5. Transparent and reproducible model outputs

Model data repository

Model code and executables

Pre- and post-processing scripts

2.6.2.1.2.1 Prediction of hydrological response variables

The objective of the numerical modelling in BAs is to assess hydrological changes arising from coal resource development using a probabilistic approach. In the Namoi subregion, the CRDP includes existing open-cut and underground mining operations, proposals to expand existing open-cut and underground mines and proposals for new open-cut and underground mines and a coal seam gas (CSG) development (see Section 2.3.4 of companion product 2.3 for the Namoi subregion (Herr et al., 2018)).

The groundwater and surface water models predict changes for a set of hydrological response variables, chosen to represent important hydrological characteristics of the system or landscape class (e.g. flow volumes, flow frequencies). Some of the hydrological outputs become inputs to receptor impact models through which the potential impacts of coal resource development on water-dependent assets can be evaluated.

The hydrological response variables for groundwater are: (i) maximum difference in drawdown (dmax); and (ii) year of maximum change (tmax). Drawdown is the difference in groundwater level between the baseline and CRDP within a regional-scale, unconfined aquifer that spans the entire model domain. These variables are generated in the model at the model nodes shown in Figure 3. There are 13,629 model nodes in the regional watertable to enable an interpolated surface of drawdown to be created so that impacts on ecological, economic and sociocultural assets can be assessed and there are another 580 model nodes in the confined part of the Pilliga Sandstone that can be used to estimate impacts on economic assets. The regional watertable is the surface layer of the model which includes all geological units that outcrop. This is the uppermost geological unit, except where the alluvium is present. These model nodes are restricted in space to where there is drawdown due to coal resource development; there is no drawdown outside of the area with model nodes. Although the change in surface water – groundwater flux is an output of the groundwater model, it is an input into the river modelling and therefore encapsulated within the set of surface water hydrological response variables (see companion submethodology M06 (as listed in Table 1) for surface water modelling (Viney, 2016)). The surface water – groundwater nodes in Figure 3 show where changes in surface water – groundwater flux are generated in the groundwater model. Changes in the nine hydrological response variables for streamflow due to the coal resource development are reported in companion product 2.6.1 for the Namoi subregion (Aryal et al., 2018). Figure 3 shows the location of the surface water model nodes relative to the groundwater model nodes where surface water – groundwater fluxes are calculated.

The groundwater model is run 3500 times using a wide range of parameter values to generate an ensemble of predictions. From this set of runs, a probability distribution is defined for each groundwater hydrological response variable at each groundwater model node in the subregion. This distribution summarises uncertainty in the prediction (Section 2.6.2.8).

Figure 3 Model nodes for the Namoi groundwater model

The blue dots are the model nodes in the regional watertable (their density is too great to see individual model nodes in most of this area), and the yellow dots are where the surface water – groundwater fluxes are aggregated from upstream.

Data: Bioregional Assessment Programme (Dataset 1)

Pumping water that flows from the coal seam, interburden and weathered rock into the working area during mining produces a cone of drawdown and a drop in the watertable around the worked area; similarly, depressurisation from CSG developments may propagate drawdown into geological layers above. Where drawdowns expand into alluvial aquifers that intersect the river channel, the flux of water from the alluvial aquifer to the river will tend to decrease. To represent this surface water – groundwater interaction, the groundwater model represents alluvial aquifers and the river network in its model structure. A river model constructed to represent the same river network can receive these changes in surface water – groundwater flux at specified points along its network to represent the combined effect of changes to surface runoff and groundwater flux on streamflow. Since groundwater and surface water systems operate at different temporal scales, the models used to represent these processes run on different time steps. Streamflow is very responsive to individual rainfall events and is usually modelled at a daily time step or finer. Groundwater levels in shallow, unconsolidated alluvium are also responsive to changes in rainfall and river stage, but also to exchanges with deeper, intermediate- and regional-scale groundwater aquifers in more consolidated material (i.e. lower transmissivities), which respond relatively slowly to changes in rainfall recharge. To predict these intermediate- and regional-scale groundwater systems, a monthly or more infrequent time step can suffice. The length of the stress periods used for the model developed in this work is a month with a single time-step for most of the stress periods. However, the stress periods are divided into five time-steps for active long-wall mine development periods to implement time-varying properties of the interburden above long-wall mines.

While fully coupled surface water – groundwater model codes are available (e.g. HydroGeoSphere, Brunner and Simmons, 2012), their use is not feasible within BAs due to their high data requirements for parameterisation and operational constraints. The latter relates mainly to the general numerical instability of such models and long run times which would severely limit a probabilistic uncertainty analysis that requires the models to be evaluated thousands of times with vastly different parameter sets.

For the Namoi subregion, the modelling suite includes the Australian Water Resources Assessment (AWRA) landscape water balance model (AWRA-L) (Viney et al., 2015) to calculate the surface runoff to streams; the MODFLOW-USG groundwater model to predict drawdown and change in surface water – groundwater flux (detailed in this product); and the AWRA river model (AWRA-R) (Dutta et al., 2015) via which surface runoff and change in surface water – groundwater flux are propagated downstream. The individual models have different spatial and temporal resolution which requires a set of customised processing steps to upscale or downscale model data to allow the models to be linked.

Figure 4 illustrates the model sequencing, parameters exchanged between models and the outputs generated at model nodes to inform the receptor impact modelling. The MODFLOW-USG, AWRA-L and AWRA-R baseline runs predict the hydrological changes of modelled coal mines that were commercially producing coal as at December 2012. The corresponding CRDP runs predict the combined hydrological changes of the baseline coal resource development (baseline) and those expected to begin commercial production after 2012 (see Section 2.3.4 of companion product 2.3 for the Namoi subregion (Herr et al., 2018)).

The maximum difference in predicted drawdown between baseline and CRDP runs, expressed in terms of dmax and tmax, yields the predicted hydrological changes due to additional coal resource development in the Namoi subregion. In the receptor impact modelling (companion product 2.7 for the Namoi subregion), the potential ecological consequences of the predicted changes in hydrological response variables in the fractured rock aquifers and alluvial aquifers are assessed.

Figure 4 Model sequence for the Namoi subregion

AWRA-L = landscape model; AWRA-R = river model; GW = groundwater; SW = surface water; No dev = no coal resource development model run; BL = baseline model run; CRDP = coal resource development pathway model run; HRV = hydrological response variable

2.6.2.1.2.2Design and construction

According to the Australian groundwater modelling guidelines (Barnett et al., 2012), the design and construction of a groundwater model should meet a clear set of objectives (see preceding section) and provide some measure of model confidence. The model confidence level is an a priori categorisation of a groundwater model to reflect its predictive capability and is a function of model complexity, prediction time frame and data availability. As explained in companion submethodology M07 (as listed in Table 1) for groundwater modelling (Crosbie et al., 2016), the groundwater models in the BAs are all Class 1 (lowest level) models because they are required to make predictions of unprecedented stresses over time frames longer than the periods with data available to constrain the model.

Further technical detail of the conceptualisation, parameterisation and implementation are provided in Section 2.6.2.3 for the MODFLOW-USG groundwater model and in companion product 2.6.1 for the Namoi subregion (Aryal et al., 2018) for the AWRA-L and AWRA-R models.

2.6.2.1.2.3Integration with sensitivity and uncertainty analyses workflow

Figure 5 Uncertainty analysis workflow

Blue boxes pertain to the design of experiment phase and green boxes pertain to the uncertainty analysis phase.

ABC = Approximate Bayesian Computation; HRV = hydrological response variable

Companion submethodology M09 (as listed in Table 1) for propagating uncertainty through models (Peeters et al., 2016) discusses in detail the propagation of uncertainty through numerical models in the BAs.

Figure 5 summarises the uncertainty propagation workflow which consists of three major steps:

1. Design of experiment: large number of model chain evaluations from the prior parameter values
2. Evaluate model runs and record:
1. each hydrological response variable at each model node
2. objective function tailored to each hydrological response variable at each model node
3. Sample model runs with an Approximate Bayesian Computation rejection sampler based on the objective function to generate a probability distribution for each hydrological response variable at each model node.

The first step is to carry out a large number of model chain evaluations, sampling extensively from the prior parameter distributions, the most likely range of the parameter values based on data and expert knowledge. For each evaluation, the corresponding predicted changes in hydrological response variables at the model nodes are stored, together with the predicted equivalents to the observations. The latter are summarised into objective functions, tailored to each hydrological response variable.

This information forms the basis for the subsequent uncertainty analysis. In the uncertainty analysis, the Approximate Bayesian Computation methodology is used to filter the predictions by only accepting those simulations that have an objective function below a predefined threshold (Vrugt and Sadegh, 2013). This results in a posterior prediction distribution, tailored to a specific hydrological response variable.

To incorporate the model chain into the uncertainty analysis it needs to be scripted so the parameter values can be changed in an automated fashion, be evaluated from a command line on high performance computers and, most importantly, be numerically stable so that the model converges for a wide range of parameter values.

The three models in the model chain for the Namoi subregion have text files as input files and can be executed from the command line. The robustness of each model is tested through a stress test in which a selection of extreme parameter combinations is evaluated. While this does not guarantee that all model evaluations will converge, it provides confidence that the majority of parameter combinations will.

Section 2.6.2.7 and Section 2.6.2.8 provide details of the implementation of this uncertainty propagation workflow for the Namoi groundwater model. The uncertainty analysis for the surface water model is in Section 2.6.1.5 and Section 2.6.1.6 of companion product 2.6.1 for the Namoi subregion (Aryal et al., 2018). These sections also have a qualitative uncertainty analysis that provides a structured discussion of the assumptions and model choices not included in the numerical uncertainty analysis and the perceived effect on the predictions.

2.6.2.1.2.4Water balance components

A secondary objective of the numerical models is to inform the water balance reporting in companion product 2.5 for the Namoi subregion (Crosbie et al., 2018). The groundwater model and AWRA models produce estimates of the water balances under baseline and CRDP.

2.6.2.1.2.5Transparent and reproducible model outputs

An overarching requirement of the BAs is for all model outputs to be transparent and reproducible.

Input data, model files (including the pre- and post-processing scripts and executables), and results are available at www.bioregionalassessments.gov.au with the specific URL listed in the dataset citation at the end of each section.

As the evaluation of the model chain is a highly automated and scripted process, it is possible to reproduce the results reported in this product using the scripts and executables, provided the computational resources are available.

Last updated:
6 December 2018