# 4 Impact and risk predictions for assets and landscape classes

Page 4 of 13

## 4.1 Predictions at assessment units

The design choice that requires hydrological predictions to be possible at any location in the landscape (described in Section 2.3.5) outlines the importance of being able to make predictions at individual locations (nominally assessment units) as part of the impact and risk analysis. These predictions are central to the workflow illustrated in Figure 7.

### 4.1.1Hydrological response variables

Surface water and groundwater hydrological models make predictions of the hydrological response variables at model nodes or stream nodes. The range or distributions of these predictions are typically summarised by a series of percentiles – nominally the 5th through to the 95th in 5% increments.

For groundwater, the predictions of individual percentiles of maximum drawdown are interpolated to the assessment units to provide complete coverage across the assessment extent. Details of this allocation or interpolation are described in Section 4.1.1.1. This means that it is possible to represent the median (50th percentile), for instance, of maximum drawdown under the baseline and under the coal resource development pathway (CRDP), and the difference in drawdown that is attributable to additional coal resource development in the bioregion or subregion.

More generally, the 5th, 50th and 95th percentiles are used to represent the predictive uncertainty for drawdown and provides the ability for the reader to bound the potential drawdown. While the 50th percentile represents the centre of the distribution for maximum groundwater drawdown, the 5th and 95th percentiles provide the lower and upper bounds. For any given assessment unit in the modelled domain, it is very unlikely that drawdown will either be smaller than the 5th percentile or exceed the 95th percentile.

For surface water, a series of interpolation rules are created that map the predictions of percentiles at stream nodes to stream reaches or links. Assessment units that intersect with the reaches or links can then access the predictions from that reach or link. In some cases it is not possible or appropriate to interpolate between certain stream nodes or beyond some modelled stream nodes. For example, it is typically difficult to interpolate volumetric surface water hydrological response variables beyond modelled stream nodes to headwater streams given the changes in flow.

The allocation or interpolation for surface water is described in more detail in Section 4.1.1.1, with any subregion-specific differences documented in product 3-4 (impact and risk analysis).

The hydrological changes may be summarised by one or more zones of potential hydrological change as discussed in Section 3.2.4.

A meaningful change in drawdown is defined in all bioregional assessments (BAs) as the area with at least a 5% chance of exceeding 0.2 m drawdown in the relevant aquifer. Groundwater impacts of coal mines and CSG projects are regulated under state legislation and state regulatory and management frameworks. The 0.2 m drawdown threshold adopted in BAs is consistent with the most conservative minimal impact threshold under the NSW Aquifer Interference Policy (NSW Office of Water, 2012) and Queensland’s Underground water impact report for the Surat Cumulative Management Area (DNRM, 2016).

For surface water, the zone of potential hydrological change is defined across the nine hydrological response variables listed in Table 3. For the flux-based hydrological response variables (annual flow (AF), daily flow rate at the 99th percentile (P99), interquartile range (IQR) and daily flow rate at the 1st percentile (P01)), the threshold change at any location is if there is at least a 5% chance of there being at least a 1% change in the variable. That is, if 5% or more of model replicates show a maximum difference between CRDP and baseline projections of 1% or more (relative to the baseline value). For four of the frequency-based metrics (high-flow days (FD), low-flow days (LFD), length of low-flow spell (LLFS) and zero-flow days (ZFD)), the threshold change at any location is if there is a greater than 5% chance of there being a maximum change in the variable of at least 3 days in any year. For the final frequency-based metric (low-flow spells (LFS)), the threshold change at any location is if there is a greater than 5% chance of there being a change in the variable of at least two spells in any year. There are many surface water hydrological response variables to weigh up, and a consideration needs to be made as to whether the interest is in a subset of surface water hydrological response variables or changes across any of them.

Table 3 Thresholds for individual surface water hydrological response variables used to define the surface water zone of potential hydrological change

Hydrological response variable

Units

Description

Threshold

AF

GL/year

The volume of water that discharges past a specific point in a stream in a year. This is typically reported as the maximum change due to additional coal resource development over the 90-year period (from 2013 to 2102).

≥5% chance of ≥1% change in AF

P99

ML/day

Daily flow rate at the 99th percentile. This is typically reported as the maximum change due to additional coal resource development over the 90-year period (from 2013 to 2102).

≥5% chance of ≥1% change in P99

IQR

ML/day

Interquartile range in daily flow; that is, the difference between the daily flow rate at the 75th percentile and at the 25th percentile. This is typically reported as the maximum change due to additional coal resource development over the 90-year period (from 2013 to 2102).

≥5% chance of ≥1% change in IQR

FD

days

Number of high-flow days per year. This is typically reported as the maximum change due to additional coal resource development over the 90-year period (from 2013 to 2102). The threshold for high-flow days is the 90th percentile from the simulated 90-year period. In some early products, this was referred to as ‘flood days’.

≥5% chance of a change in FD ≥3 days in any year

P01

ML/day

Daily flow rate at the 1st percentile. This is typically reported as the maximum change due to additional coal resource development over the 90-year period (from 2013 to 2102).

≥5% chance of ≥1% change in P01 and change in runoff depth >0.0002 mm

ZFD

days

Number of zero-flow days per year. This is typically reported as the maximum change due to additional coal resource development over the 90-year period (from 2013 to 2102).

≥5% chance of a change in ZFD ≥3 days in any year

LFD

days

Number of low-flow days per year. This is typically reported as the maximum change due to additional coal resource development over the 90-year period (from 2013 to 2102). The threshold for low-flow days is the 10th percentile from the simulated 90-year period.

≥5% chance of a change in LFD ≥3 days in any year

LFS

number

Number of low-flow spells per year (perennial streams only). This is typically reported as the maximum change due to additional coal resource development over the 90-year period (from 2013 to 2102). A spell is defined as a period of contiguous days of flow below the 10th percentile threshold.

≥5% chance of a change in LFS ≥2 spells in any year

LLFS

days

Length (days) of the longest low-flow spell each year. This is typically reported as the maximum change due to additional coal resource development over the 90-year period (from 2013 to 2102).

≥5% chance of a change in LLFS ≥3 days in any year

The zone of potential hydrological change is the union of the groundwater zone of potential hydrological change (the area with a greater than 5% chance of exceeding 0.2 m of drawdown in the relevant aquifers) and the surface water zone of potential hydrological change (the area with a greater than 5% chance of exceeding changes in relevant surface water hydrological response variables).

While there is the intention is to be conservative in defining this zone of potential hydrological change in BA so that there is confidence in areas and water-dependent assets that are assessed as not impacted, it is possible in principle to repeat the process with different thresholds (and that speak to specific values that are important to key users) given the model data will be publicly available on data.gov.au.

Landscape classes or assets that lie outside of the zone of potential hydrological change are very unlikely to experience any hydrological change due to additional coal resource development. Where an asset or landscape class, either wholly or partially, intersects with the zone of potential hydrological change, there is the potential for impact. It is important to stress that this does not imply that there is impact – only that it cannot be ruled out on the basis of the hydrological change and that further investigation is required using qualitative mathematical modelling, receptor impact models and other lines of evidence. That further work also involves considering the nature of the water dependency of particular landscape classes within the zone of potential hydrological change. If a landscape class is not considered water dependent (e.g. ‘Production from dryland agriculture and plantations’), then potential impacts to that landscape class may be ruled out.

Multiple zones of potential hydrological change can be considered and reported against. Given the number of near-surface assets, the most important zone relates to the hydrological changes in the uppermost geological layers using spatially explicit, probabilistic estimates of hydrological change from the regional groundwater models. For the purposes of BA, this is known as the regional watertable and is used to assess potential impacts to key surface ecosystems (landscape classes (except springs), ecological assets and sociocultural assets). In the case of groundwater bores and springs, it is important to determine the source aquifer of each individual bore or spring for the impact and risk analysis. The source aquifer for each bore or spring is identified from existing datasets. Where this information is not available, the assessment will typically assume that the bores or springs access the shallowest hydrogeological layer in that assessment unit (i.e. the regional watertable). It is, however, important that this is noted as it may have implications for the impact and risk analysis; for example, if it is not known which aquifer a spring or bore accesses, it is not possible to complete quantitative assessment for springs and bores.

#### 4.1.1.1Allocating modelling node results to assessment units

Surface water and groundwater modelling and uncertainty analyses are completed at specific points in the landscape called nodes. For the impact and risk analysis, it is necessary to interpolate those modelling results across the assessment extent so that inferences can be made about potential changes that may be experienced by particular ecosystems (landscape classes) and water-dependent assets. This is achieved by defining a zone of potential hydrological change, based on the union of a groundwater zone of potential hydrological change and a surface water zone of potential hydrological change, and allocating (where relevant) a groundwater modelling node and a surface water modelling node to every assessment unit within that zone.

The allocation of groundwater nodes to an assessment unit is achieved by selecting the node closest to the centroid of the assessment unit. This selection is manually checked to ensure the linking is hydrogeologically sound (e.g. to avoid selecting nodes that cross important geological boundaries).

Surface water modelling interpolation is achieved through a process of allocating node results to river reaches that extend upstream and/or downstream from the point of an individual modelling node. The first step is to select a spatial line network to represent the streams of the region. This stream network is broken into sections named reaches, as per the surface water conceptual model of the bioregion or subregion.

Initial assessment units are selected by way of their intersection with a buffered version of the stream reaches network. The size of the buffer is bioregion- or subregion-specific choice and is selected by expert judgement and informed by the specific landscape attributes of the bioregion or subregion. A further selection of assessment units is applied to include neighbours that are considered hydrologically connected by way of their intersection with water-dependent landscape class features, such as lowland streams, upland streams or floodplains.

At this stage, each reach is allocated one of four values: modelled impact via a modelled node, potentially impacted but not modelled, no impact, or an unknown impact as the reach was not part of the original conceptual model. These reach attributes determine their connection and status within the impact and risk analysis calculations.

To complete the process, all assessment units selected within the surface water subset of the zone of potential hydrological change are allocated a stream reach to determine the potential impact. The stream reach to assessment unit relationship is exclusively 1:1 and governed by the following hierarchical rule set. Units within the surface water zone of potential hydrological change but no intersecting reaches are allocated the nearest reach. Units containing a single intersecting reach are allocated that reach. For units with multiple intersecting reaches, a priority allocation is applied as: modelled change, assumed change (potential impact), modelled no change, or assumed no change.

The combination of assessment units that make up the surface water zone of potential hydrological change is reviewed by hydrology experts to ensure that assessment unit selections are hydrologically valid.

The selection of all assessment units included within both the surface water and groundwater modelled areas creates the zone of potential hydrological change upon which the impact and risk analysis is completed.

### 4.1.2Receptor impact variables

Receptor impact models make predictions about the response of receptor impact variables (ecosystem indicators) to one or more hydrological response variables. When the range or distribution of possible changes in those hydrological response variables is considered, and translated using a receptor impact model, it results in a distribution of receptor impact variable predictions. This distribution represents the range of possible outcomes for the receptor impact variable and incorporates the uncertainty in both the hydrological response variables and the uncertainty in the ecosystem response to that hydrological change as characterised by the uncertainty in the receptor impact model (Section 3.2.5).

Predictions can be made at an assessment unit based on the changes in those hydrological response variables at that assessment unit. It is important to note that those predictions are a predicted response across the landscape class for the local hydrological change in that assessment unit. They thus represent the predicted response in the receptor impact variable for all locations across the landscape class given that level of hydrological change.

These predictions may be extended to areas of interest (e.g. stretches of river) by applying the receptor impact models at different assessment units and using the changes in hydrological response variables at each of those assessment units. Landscape class scale is the natural level of aggregation given that the elicitation for the receptor impact models is conducted at that scale. The aggregation to the landscape levels weights the contribution of each assessment unit by the amount of the landscape class contained in each assessment unit. This weight could be linear, as in the case of landscape classes defined by stream reaches, or by area, as in the case of some groundwater-dependent forested landscape classes.

The receptor impact variables and hydrological response variables used in the receptor impact modelling are selected using the qualitative mathematical modelling (see Section 2.3.10.2), and are summarised in Table 4 and Table 5, respectively. The hydrological response variables are based on the averages over the short term (2013 to 2042) and long term (2073 to 2102) rather than the maximum change over the 90-year simulation period as used for the standard hydrological response variables (Table 3).

Table 4 Summary of the hydrological response variables used in the receptor impact models

This is the entire suite of hydrological response variables used in bioregional assessments; each subregion uses only a subset of these hydrological response variables.

Hydrological response variable

Definition of hydrological response variable

dmaxRef

Maximum difference in drawdown under the baseline future or under the coal resource development pathway future relative to the reference period (1983 to 2012). This is typically reported as the maximum change due to additional coal resource development.

tmaxRef

The year that the maximum difference in drawdown relative to the reference period (1983 to 2012) (dmaxRef) occurs

EventsR0.3

The mean annual number of events with a peak daily flow exceeding the threshold (the peak daily flow in flood events with a return period of 0.3 years as defined from modelled baseline flow in the reference period (1983 to 2012)). This metric is designed to be approximately representative of the number of overbench flow events in future 30-year periods. This is typically reported as the maximum change due to additional coal resource development.

EventsR3.0

The mean annual number of events with a peak daily flow exceeding the threshold (the peak daily flow in flood events with a return period of 3.0 years as defined from modelled baseline flow in the reference period (1983 to 2012)). This metric is designed to be approximately representative of the number of overbank flow events in future 30-year periods. This is typically reported as the maximum change due to additional coal resource development.

EventsR0.2

The mean annual number of events with a peak daily flow exceeding the threshold (the peak daily flow in flood events with a return period of 0.2 years as defined from modelled baseline flow in the reference period (1983 to 2012)). This metric is designed to be approximately representative of the number of overbench flow events in future 30-year periods. This is typically reported as the maximum change due to additional coal resource development.

EventsR2.0

The mean annual number of events with a peak daily flow exceeding the threshold (the peak daily flow in flood events with a return period of 2.0 years as defined from modelled baseline flow in the reference period (1983 to 2012)). This metric is designed to be approximately representative of the number of overbank flow events in future 30-year periods. This is typically reported as the maximum change due to additional coal resource development.

LME

The maximum length of spells (in days per year) with low flow, averaged over a 30-year period. This is typically reported as the maximum change due to additional coal resource development.

LQD

The number of days per year with low flow (<10 ML/day), averaged over a 30-year period. This is typically reported as the maximum change due to additional coal resource development.

QBFI

Ratio of total baseflow generation to total streamflow generation, averaged over a 30-year period. This is typically reported as the maximum change due to additional coal resource development.

ZMA

​The maximum length of spells (in days per year) with zero flow over a 30-year period. This is typically reported as the maximum change due to additional coal resource development.

ZME

The maximum length of spells (in days per year) with zero flow, averaged over a 30-year period. This is typically reported as the maximum change due to additional coal resource development.

ZQD

The number of zero-flow days per year, averaged over a 30-year period. This is typically reported as the maximum change due to additional coal resource development.

Table 5 Summary of the receptor impact variables used in the receptor impact models

Receptor impact models may use all or a subset of the stated hydrological response variables for each receptor impact variable.

Receptor impact variable (with associated sample units)

Hydrological response variables

Annual mean percent canopy cover of woody riparian vegetation (predominately Casuarina cunninghamiana, Melia azedarach, Eucalyptus amplifolia, E. tereticornis and Angophora subvelutina) in a transect 20 m wide and 100 m long covering the bottom of the stream bench to the high bank

dmaxRef

tmaxRef

EventsR0.3

EventsR3.0

Annual mean projected foliage cover (m2/m2) of sclerophyll forest (predominately Angophora costata, Corymbia gummifera, Eucalyptus capitellata, Banksia spinulosa) in a 0.25 ha plot

dmaxRef

tmaxRef

Mean annual projected foliage cover (m2/m2) of woody riparian vegetation (predominately Eucalyptus tereticornis, Casuarina cunninghamiana and Eucalyptus camaldulensis) in a 0.25 ha transect extending from the channel to the top of the bank (including floodplain overbank)

dmaxRef

tmaxRef

EventsR0.3

EventsR3.0

Annual mean percent foliage cover of woody riparian vegetation (target species include Eucalyptus camaldulensis and Melaleuca spp.) in a transect 10 to 15 m wide and 100 m long covering the stream channel to the top of the stream bank

dmaxRef

LQD

EventsR2.0

Annual mean projected foliage cover of forests dominated by river red gum (E. camaldulensis)

EventsR3.0

dmaxRef

tmaxRef

Annual mean projected foliage cover of species group that includes: Casuarina, yellow box, Blakely's red gum, Acacia salicina, Angophora floribunda, grey box. Transect of 50 m length and 20 m width that extends from first bench (‘toe’) on both sides of stream

EventsR3.0

dmaxRef

tmaxRef

Annual mean projected foliage cover of species group that includes: yellow box, white cypress pine, Eucalyptus crebra, dirty gum, Blakely's red gum, Angophora floribunda, Eucalyptus fibrosa, fuzzy box. Transect of 50 m length and 20 m width that extends from first bench (‘toe’) on both sides of stream

ZQD

dmaxRef

tmaxRef

Mean abundance of larvae of the Hydropsychidae family (net-spinning caddisflies) in a 1 m2 sample of riffle habitat

ZQD

ZMA

Annual mean abundance (30 years, >33 sites/year) of the mayfly Offadens (family Baetidae), 3 months after the wet season in a 2 m × 0.5 m (1 m2) area of riffle habitat

LQD

LME

Mean abundance of the eel-tailed catfish (Tandanus tandanus) in a 600 m2 transect (100 m by 6 m) whose long axis lies along the mid-point of the stream

ZQD

QBFI

Mean probability of presence of the riffle-breeding frog (Mixophyes balbus) in a 100 m transect

ZQD

ZMA

Probability of presence of tadpoles from Limnodynastes genus (species dumerilii, salmini, interioris and terraereginae), sampled using standard 30 cm dip net

EventsR3.0

ZQD

ZME

Average number of families of aquatic macroinvertebrates in riffle habitat sampled using the NSW AUSRIVAS method for riffles

ZQD

ZME

Average number of families of aquatic macroinvertebrates in instream pool habitat sampled using the NSW AUSRIVAS method for pools

ZQD

dmaxRef

tmaxRef

Mean richness of hyporheic invertebrate taxa in 6 L of water pumped from a depth of 40 cm below the streambed (riffle and gravel bars; Hancock, 2004)

ZQD

ZMA

The hydrological response variables that are used in the receptor impact models are also based on the suite of runs of the surface water and groundwater hydrological models rather than the percentile summaries described in Section 4.1.1. The reason for this is that the runs preserve the correlation (dependence) between the individual hydrological response variables. This means when receptor impact models use two or more hydrological response variables, realistic combinations of hydrological response variables occur with the correct frequency. For instance, if the number of overbank and overbench events are positively correlated (or more generally positively dependent), it would be more likely that if one event measure is high (or low) then the other is also likely to be high (or low). If each hydrological response variable is treated independently, that correlation constraint is not considered and the enhanced frequency of high (or low) overbank flows and high (or low) overbench flows is lost in the simulations

While the surface water and groundwater models are loosely coupled, the runs operate on different time steps and the dependence between individual surface water and groundwater runs is maintained. For the surface water hydrological response variables used in receptor impact modelling the correlation is maintained within high-flow hydrological response variables and low-flow hydrological response variables but not between them. In practice this is of no consequence as individual receptor impact models are almost always constructed using either low-flow or high-flow hydrological response variables but not both.

## 4.2Predictions for landscape classes and assets

The overarching purpose of BAs is to quantify potential impacts and risks to water resources and water-dependent assets due to coal resource development. This requires predictions of potential hydrological changes (through hydrological response variables) and potential ecosystem change (through receptor impact variables) to be made at locations (assessment units) that are relevant to that water resource or asset and at key points in time.

Predictions across the extent of an individual water resource or a water-dependent asset can then be aggregated or presented in various ways to create a summary of impact or risk. If those locations are representative of the water resource or asset then a simple unweighted summary across those locations is representative.

Landscape classes have been introduced as a classification of biophysical ecosystems in response to the large numbers of assets. Within a landscape class the ecosystem is expected to be relatively homogenous in the key hydrological drivers and how it responds to them – relative to the differences between landscape classes. To a large degree individual landscape classes can be considered as an ‘ecosystem asset’ and the prediction and summary challenges relevant to landscape classes are also relevant to water-dependent assets. To assess potential impacts on and risks to a landscape class requires predictions to be made at locations that are relevant to that landscape and at key points in time. If predictions are made across a set of locations that are representative, they may be aggregated and summarised to emphasise the potential impact and risk.

While water-dependent assets and landscape classes may be polygonal (e.g. groundwater-dependent vegetation ecosystems), linear (e.g. parts of a stream ecosystem) or points (e.g. individual springs), the concept of aggregating or summarising across those assessment units that pertain to that asset or landscape class persists. Chapter 6 outlines some of the specific choices for reporting and communicating the predicted impacts and risks for landscape classes and assets.

## 4.3Systematically processing the data

There are a very large number of multi-dimensional and multi-scaled datasets that are used in the impact and risk predictions, and the analysis more generally, for each BA. These include model outputs, and ecological, economic and sociocultural data from a wide range of sources. Part of the approach used to manage these multiple dimensions and produce meaningful results is to adopt a clear spatial framework as an organising principle. While the inherently spatial character of every BA is important and must be addressed, it is also essential that the temporal and other dimensions of the analysis do not lose resolution during data processing. For example, knowing where a potential impact may take place is obviously important, but so is knowing what kind and level of impact and which assets may be affected.

The design of the system for ingesting, managing and producing data useful for analysis purposes is based on a spatially-enabled open source relational database (PostGRES) with strong provenance tracking capability. The data are organised into impact and risk analysis databases to enable efficient management. Only data that are registered as datasets at data.gov.au are ingested and used. There are multiple stages of processing the data to ensure compliance with the information model and database normalisation requirements.

The purpose of the databases is to produce result datasets that integrate the available modelling and other evidence across the assessment extent of the BAs. The resulting datasets are required to support the BA analyses. These data are delivered to the Assessment teams as a series of queries the teams have developed in collaboration with the database management team. These queries can be loaded by the Assessment teams into data analysis software, such as ArcGIS, QGIS, and statistical analysis environments, such as R and Python.

Appendix A provides further detail about the approach undertaken to manage this data and the queries required from it. The following sections explain the analyses the Assessment teams need to produce, with detail on how to communicate and report the results.

Last updated:
7 December 2018