- Home
- Assessments
- Bioregional Assessment Program
- Namoi subregion
- 2.6.1 Surface water numerical modelling for the Namoi subregion
- 2.6.1.4 Calibration
- 2.6.1.4.2 Australian Water Resources Assessment river model
Data
Input data to drive the AWRA river model (AWRA-R) calibration include climate, potential evaporation, catchment runoff (from AWRA-L), groundwater depth and town water supply diversions. Calibration datasets against which performance of AWRA-R and the various modules were evaluated are daily streamflow, dam storage volumes, water allocations, dam releases and irrigation diversions. The calibration period for the different AWRA-R components covers 1 January 1981 to 30 June 2006.
The only direct climate input into AWRA-R is daily precipitation, which is used to calculate precipitation directly onto the open water surfaces (the river channels and storages). Daily gridded precipitation data from the Bureau of Meteorology (Dataset 1) have been used in the calibration of AWRA-R. The gridded data were clipped and aggregated (spatially averaged) using reach subcatchment boundaries defined in the Namoi river system AWRA-R node-link network (Bioregional Assessment Programme, Dataset 4).
Daily estimates of potential evaporation and catchment runoff were obtained from the calibrated AWRA-L simulation (Bioregional Assessment Programme, Dataset 2) and aggregated to the reach scale defined by the Namoi river system node-link network (Bioregional Assessment Programme, Dataset 4) for input into AWRA-R.
Town water supply diversions are not calibrated and are used as inputs in AWRA-R (Bioregional Assessment Programme, Dataset 5).
The AWRA-R model used for calibration comprises 45 nodes and their contributing areas, shown in Figure 11. These include the 32 gauging stations with stage and streamflow data used in assessing the model’s performance in calibration (NSW Office of Water, Dataset 3), and one dummy node located close to the Pian Creek offtake. Twenty-three of the gauging stations are non-headwater gauges: they include ten on the Namoi River, three on the Peel River, three on Pian Creek, two on the Manilla River, and one each in Coxs Creek, Mooki River, Narrabri Creek, Cockburn River and Gunidgera Creek (red triangles in Figure 11). The other calibration model nodes (black dots in Figure 11) comprise nine gauging stations on headwater streams, plus 12 model nodes (see Section 2.6.1.3.2), located on ungauged headwater streams. Daily streamflows at the 12 ungauged nodes are simulated by AWRA-L (Bioregional Assessment Programme, Dataset 2) and stages obtained from idealised cross-sections at each location (see companion product 2.1-2.2 for the Namoi subregion (Aryal et al., 2018); Bioregional Assessment Programme, Dataset 4). Eight model nodes, defining an additional eight ungauged reaches, located downstream of other model nodes, are not used in the calibration but are needed for the AWRA-R model simulations.
Figure 11 The 45 streamflow gauging stations used for AWRA-R model calibration
AWRA-R = Australian Water Resources Assessment river model
Data: Bureau of Meteorology (Dataset 1), Bioregional Assessment Programme (Dataset 4)
Daily irrigation diversions, dam storage volumes, daily dam releases and allocations were sourced from the Namoi and Peel Integrated Quantity-Quality Model (IQQM) (Bioregional Assessment Programme, Dataset 5) for the period 1 January 1981 to 30 June 2006. These were used to calibrate the AWRA-R irrigation and dam storage and releases modules.
Model calibration results
Streamflow routing and reach water balance
AWRA-R was calibrated using 23 streamflow gauges defining a concurrent number of modelling reaches. Two variants of the model were calibrated using AWRA-L high-streamflow and low-streamflow calibration outputs, respectively (Bioregional Assessment Programme, Dataset 2).
The agreement of both high-streamflow and low-streamflow AWRA-R calibrations were assessed using the Nash–Sutcliffe efficiency (Ed) for daily streamflow (Ed(1.0)) and for daily streamflow transformed with a Box-Cox lambda value of 0.1 (Ed(0.1)) as well as both F1 and F2 values (see Section 2.6.1.4.1). The bias generally remains very low as each reach is individually calibrated and parameterised, as opposed to the regional calibration implemented for AWRA-L (see companion submethodology M06 (as listed in Table 1) for surface water modelling (Viney, 2016)).
The performance is reported for the 23 non-headwater gauges depicted as triangles in Figure 11 and where routing and reach water balance processes (including runoff from catchments between the two nodes, river water diversions, groundwater fluxes, overbank flow) take place.
Figure 12 shows boxplots summarising the performance of the AWRA-R high-streamflow and AWRA-R low-streamflow calibrations in the 23 gauges. Table 9 presents a summary of the goodness-of-fit metrics used to evaluate the two model variants. In terms of Ed(1.0) (emphasis on high flows), the AWRA-R high-streamflow calibration agrees reasonably well with observations, indicated by a median Ed(1.0) of 0.77 (interquartile range 0.40 to 0.84) and a median bias of 0.001. Fourteen of the 23 gauges have an Ed(1.0) greater than 0.6, while four gauges (419007, 419061, 419088 and 419089 corresponding to model nodes 47, 11, 4 and 3, respectively) have a negative Ed(1.0).
It is noted that the IQQM data used to calibrate irrigation diversions was based on a fixed irrigation development (year 2000), hence it would tend to over estimate irrigation diversion during the 1980s and 1990s when agricultural development was less. However, calibration of the irrigation module was performed prior to the reach-by-reach calibration in AWRA-R, therefore any undue impacts on streamflow arising from a fixed irrigation development were reduced through calibration.
AWRA-R = Australian Water Resources Assessment river model
In each boxplot, the bottom, middle and top of the box are the 25th, 50th and 75th percentiles, and the bottom and top whiskers are the 10th and 90th percentiles. F1 is the F value for high-streamflow calibration; F2 is the F value for the low-streamflow calibration; Ed(1.0) is the daily efficiency with a Box-Cox lambda value of 1.0; Ed(0.1) is the daily efficiency with a Box-Cox lambda value of 0.1.
Data: Bioregional Assessment Programme (Dataset 4)
Overall agreement in terms of Ed(0.1) (emphasis on medium to low flows) for the AWRA-R high-streamflow calibration is much poorer than Ed(1.0), with a median value of 0.16 (interquartile range –0.65 to 0.58). Seven of the 23 gauges have an Ed(0.1) greater than 0.5, while ten have negative Ed(0.1) values. The high-streamflow calibration yields a median F1 of 0.77 and the median bias is about 0.002.
Table 9 Summary of AWRA-R calibration for the 23 calibration gauges in the Namoi subregion
AWRA-R = Australian Water Resources Assessment river model; F1 = F value for high-streamflow calibration; F2 = F value for low-streamflow calibration (see Viney, 2016); Ed(1.0) = daily efficiency with a Box-Cox lambda value of 1.0; Ed(0.1) = daily efficiency with a Box-Cox lambda value of 0.1
Data: Bioregional Assessment Programme (Dataset 4)
The AWRA-R low-streamflow calibration has a median of Ed(1.0) of 0.78 (interquartile range 0.56 to 0.87). Sixteen out of the 23 gauges have an Ed(1.0) greater than 0.6, and the same four gauges have a negative Ed(1.0). In terms of Ed(0.1) for AWRA-R low-streamflow calibration, model performance is marginally better than for the AWRA-R high-streamflow calibration, with a median value of 0.21 (interquartile range –0.60 to 0.61). Only seven gauges have an Ed(0.1) greater than 0.5. The low-streamflow calibration has a median F2 value of 0.20 and a median bias of 0.001.
Overall, the low-streamflow calibration was deemed better in terms of Ed(1.0) and bias than the high-streamflow calibration, and marginally better in terms of Ed(0.1).
The impact of including eight additional ungauged nodes (where simulations are required) between gauges nodes (thus defining new reaches) marginally degrades agreement with observations compared to the original calibration of AWRA-R. For example, the median Ed(1.0) using the low-streamflow calibration with the additional gauges is 0.77 (range 0.53 to 0.86). The largest degradation is for gauge 419027 (Node 35 in the Mooki River), where Ed(1.0) decreases from 0.58 to 0.53, followed by gauge 419001 (Node 32 in the Namoi River) from 0.92 to 0.89. Changes in Ed(1.0) are marginal (<0.02) in the rest of the gauging stations. Again, median bias is less than 0.002.
Boxplots in Figure 13 show the bias of the two calibration schemes (AWRA-R high-streamflow and AWRA-R low-streamflow calibration) in predicting the nine hydrological response variables that characterise the impacts of coal resource development on water resources (see Section 2.6.1.4.1). Similar to the AWRA-L results, these boxplots show that generally the AWRA-R low-streamflow calibration has smaller model biases and similar interquartile ranges compared to the AWRA-R high-streamflow calibration for the low-streamflow metrics (LFD, LFS, LLFS and P01). ZFD and P01 remain poorly simulated. The median biases in the high-streamflow metrics are generally marginally smaller in the high-streamflow calibration, and the interquartile ranges for high‑streamflow metrics from both calibrations are narrower than their low-streamflow counterparts, highlighting less variability among the hydrological response variables in the calibrated Namoi reaches.
AWRA-R = Australian Water Resources Assessment river model
In each boxplot, the left, middle and right of the box are the 25th, 50th and 75th percentiles, and the left and right whiskers are the 10th and 90th percentiles for the 23 AWRA-R calibration reaches.
Shortened forms of hydrological response variables are defined in Table 8.
Data: Bioregional Assessment Programme (Dataset 4)
Irrigation module
The irrigation module in AWRA-R was calibrated in 14 reaches along the Namoi River and three reaches along the Peel River. In the absence of observed irrigation diversion and allocation data, calibration of AWRA-R was performed against monthly IQQM-simulated diversions and allocation data for the period 1981 to 2006 (Bioregional Assessment Programme, Dataset 5) using AWRA-R simulated streamflow (Bioregional Assessment Programme, Dataset 4). Crop types and crop proportions used in the calibration were also obtained from the Namoi and Peel IQQM models and adjusted during calibration if needed.
Table 10 presents a summary of goodness-of-fit metrics used to evaluate the performance of calibration. Simulated and observed irrigation diversions were compared using (i) the coefficient of determination (r2), which indicates the agreement in temporal patterns, (ii) the monthly Nash–Sutcliffe efficiency (Em), which indicates the calibration accuracy, and (iii) bias (B), which indicates the overall tendency of the model towards overestimation or underestimation. Overall, AWRA-R calibrated diversions are satisfactory, with a median r2 of 0.4 and interquartile range from 0.25 to 0.47. The median Nash–Sutcliffe efficiency is 0.35 (interquartile range from 0.14 to 0.47). Diversions are under estimated in ten reaches, with three reaches with nodes 49, 3 and 45 having a negative bias of more than 20%, whereas overestimation occurs in the remaining reaches and only one reach (node 15) has a positive bias greater than 20%. The median bias is 10%.
Table 10 Summary of AWRA-R irrigation calibration for 17 reaches in the Namoi river basin
r2 = coefficient of determination; Em = monthly Nash–Sutcliffe efficiency; B = bias; AWRA-R = Australian Water Resources Assessment river model
Data: Bioregional Assessment Programme (Dataset 4, Dataset 5)
Figure 14 shows examples of time series of observed and simulated diversions and mean monthly diversions for three reaches located on the Namoi River.
IQQM = Integrated Quantity-Quality Model
Data: Bioregional Assessment Programme (Dataset 4, Dataset 5)
Generally, diversion peaks during the summer months (December to February) are under estimated, particularly prior to 1996 though less so in the remainder of the calibration period. Mean monthly diversions are simulated reasonably well, with most of the diversions occurring during the summer months and little or no diversion occurring during the winter months, when irrigation demand is low. IQQM diversions increase in many reaches during September, perhaps for irrigation of winter pasture during the last growing stage or for filling of on-farm storages before the start of the cotton irrigation season. Large peaks and increase in diversions in September were difficult to replicate in AWRA-R, even with the modifications (off-allocation diversion) implemented in the model (see Section 2.6.1.3) since these occurrences do not happen in most years.
Dam storage volumes, releases and allocations
Two systems are considered to estimate allocations in the Namoi subregion: (i) the Namoi River system, which includes Split Rock Dam and Keepit Dam and the downstream reaches that order water from these two dams; and (ii) the Peel River system, which includes Chaffey Dam and the downstream reaches that order water from this dam. Storage volumes for Split Rock Dam and Keepit Dam were calibrated as a single dam with combined storage against IQQM-simulated combined daily storage volumes (Bioregional Assessment Programme, Dataset 5), referred hereafter as Namoi Dams. Only releases from Keepit Dam were calibrated, with Split Rock transferring water to Keepit Dam when downstream demand is higher than the volume of water stored in Keepit. Transfers do not exceed 1500 ML/day. Storage volumes for Chaffey Dam releases were calibrated against IQQM-simulated daily storage volumes, releases and allocations (Bioregional Assessment Programme, Dataset 5), rather than against true observed data as the latter are too patchy and incomplete. The general calibration methodology and justification is described in companion submethodology M06 (as listed in Table 1) for surface water modelling (Viney, 2016). Allocations in the Namoi River and Peel River systems are determined by the volume of water held in the systems’ dams.
Goodness-of-fit metrics used to evaluate the performance on calibrated storage volumes and dam releases include the monthly Nash–Sutcliffe efficiency (Em), the coefficient of determination (r2) and the root-mean-square error (RMSE) to assess the overall accuracy.
For illustrative purposes, time series of monthly storage volumes are shown in Figure 15a and Figure 15b. Temporal patterns agree well with observations, with r2 values of 0.84 and 0.89 for Namoi Dams and Chaffey Dam, respectively. In contrast, modelled volumes are satisfactory for Namoi Dams (Em = 0.57) and good for Chaffey Dam (Em = 0.71). The RMSE is 113 GL and 5 GL for Namoi Dams and Chaffey Dam, respectively. It can be seen in Figure 15 that the Namoi AWRA-R dam model generally under estimates storage volumes between 1981 and 1993, but over estimates them during the dry period between 2003 and 2006.
Time series of monthly dam releases are shown in Figure 15c and Figure 15d. Temporal patterns agree well with observations, with r2 values of 0.74 and 0.85 for Namoi Dams and Chaffey Dam, respectively. Simulated releases are good for both the Keepit Dam (Em = 0.53) and Chaffey Dam (Em = 0.71). The releases RMSE for the calibration period (1 January 1986 to 30 June 2006) are 156 GL and 24 GL for Keepit Dam and Chaffey Dam, respectively.
Figure 16 summarises the monthly percentage error in storage volume for Namoi Dams and Chaffey Dam, computed as the percentage of the difference between the volumes simulated and observed divided by the simulated volume. The error is generally between –20% and +20%. It is likely that the modelled storages will generally be within a similar margin of error.
AWRA-R = Australian Water Resources Assessment river model; IQQM = Integrated Quantity-Quality Model
Data: Bioregional Assessment Programme (Dataset 4, Dataset 5)
Figure 16 Summary of percentage error for monthly dam storage volumes for Namoi Dams and Chaffey Dam
In each boxplot, the bottom, middle and top of the box are the 25th, 50th and 75th percentiles, and the bottom and top end of whiskers are the 10th and 90th percentiles.
Data: Bioregional Assessment Programme (Dataset 4, Dataset 5)
The poor daily efficiencies for the simulated releases from Keepit Dam can be largely attributed to the under estimation of a number of peak releases, particularly during the period 1981 to 1993, but also a large spill release in September 1998. In the case of Chaffey Dam releases, the model conceptualisation captures well the downstream irrigation and town water supply demand.
Figure 17 IQQM and AWRA-R simulated allocations for the Namoi River and Peel River systems
AWRA-R = Australian Water Resources Assessment river model; IQQM = Integrated Quantity-Quality Model
Data: Bioregional Assessment Programme (Dataset 4, Dataset 5)
Time series of observed and estimated allocations are shown in Figure 17. Simulated allocations for the Namoi River (Figure 17a) reflect the amount of storage in both modelled dams and the continuous accounting system, which allows users to have more than 100% of their entitled licence. As is the case for dam storage volumes in the Namoi River system, AWRA-R generally under estimates allocations between 1981 and 1993, but over estimates them during the dry period between 2003 and 2006. AWRA-R simulated allocations in the Peel River generally show higher allocations in years when water volumes in the dam are high (greater than 50 GL) and do not show the inter-annual variability present in IQQM. This is because IQQM computes allocations on 1 July (defined as the beginning of the water year in the Peel River, Green et al., 2011) and allocation increments may occur during the year if available resources increase. Conversely, AWRA-R computes an annual allocation on 31 August (preceding the beginning of the irrigation season), when about 33% of the annual average inflow has reached the dam. Overall, AWRA-R shows reasonable allocation patterns both in dry and wet periods.
Implications for model predictions
Overall, AWRA-R streamflow is reasonably well simulated using both high-streamflow and low-streamflow calibrations. As expected, the low-streamflow calibration performs better when the evaluation is focused on the low flows, particularly in intermittent streams. The better performance of the low-streamflow calibration is also observed in terms of the hydrological response variables used to quantify the hydrological changes due to additional coal resource development, particularly for the variables characterising low streamflow, although the performance for the variables characterising high flows is typically only marginally worse. Generally, the hydrological response variables characterising low-flow conditions are under estimated, whereas those characterising high-flow conditions are over estimated. The number of zero-flow days remains poorly simulated with either model variant, though is slightly better for the low-streamflow calibration.
Similar to AWRA-L calibrations, the parameter sets obtained for AWRA-R high-streamflow and low-streamflow calibration show large uncertainty in predicting the nine hydrological response variables. The outputs from simulations that use the 3000 parameter sets for both AWRA-L and AWRA-R are expected to provide suitable uncertainty bounds for predicting the hydrological changes due to additional coal resource development on the nine hydrological response variables in each model node in the Namoi subregion.
A wide range of dam operating conditions are represented in the model as a result of choosing a 26-year calibration period that includes significant dry and wet periods. This ensures that the model can be used in some of the extreme conditions imposed by modelling of additional coal resource development.
The irrigation demand routine that computes releases for irrigation performs reasonably well. This is highlighted in the reasonably good representation of monthly patterns (r2 greater than 0.7 in both Keepit Dam and Chaffey Dam) between simulated and IQQM-modelled releases. Releases are well simulated for Chaffey Dam (Em = 0.71). Some peak releases from Keepit Dam are not well matched by AWRA-R, resulting in poor overall agreement with an Em equal to 0.08.
AWRA-R simulated dam storage volumes, releases and allocations are reasonable and comparable to studies that simulated dam volumes and releases in multi-purpose reservoirs for scenario modelling (see Wu and Chen, 2012).
Overestimation of releases during the dry period between 2003 and 2006 can be partly explained by AWRA-L inflows to both dams, as AWRA-L tends to over estimate streamflow in the Namoi River during this drought period. Moreover, the calibration scheme uses one parameter per dam that linearly scales these inflows, thus it is difficult to solve this issue through calibration of the AWRA-R dam module only. Improvement of model performance can be achieved through better model conceptualisation and calibration strategies.
Finally, it should be noted that the hydrological modelling (in Section 2.6.1.6) presents the difference between the baseline and CRDP futures. As both futures use the same set of model parameters for simulation, any shortcomings in model performance will be common to both runs and would be largely eliminated on taking the difference.