Territory Stories

Development of a Groundwater Model for the Western Davenport Plains



Development of a Groundwater Model for the Western Davenport Plains


Knapton, Anthony; CloudGMS Pty Ltd

Commissioned by

Northern Territory. Department of Environment, Parks and Water Security


E-Publications; E-Books; PublicationNT; WRD Technical Report 27/2017




Western Davenport Water Control District


CloudGMS has been commissioned by DENR to develop a numerical groundwater model of the aquifers within the central area of the WDWCD to improve confidence in the sustainability of the groundwater resources, as this is the area within the WCD with greatest potential for intensive development.


Made available by via Publications (Legal Deposit) Act 2004 (NT); Prepared for Dept Environment and Natural resources

Table of contents

Executive summary -- 1 Background -- 2 Physical -- 3 Available data -- 4 Conceptual model -- 5 Model design & construction -- 6 Parameter estimation -- 7 Water balances -- 8 Sensitivity analysis -- 9 Predictive scenarios -- 10 Conclusions -- 11 Reference -- 12 Document history and version control -- Appendix A - Groundwater level hydrographs - Appendix B - Alek range horticultural farm sub-regional modelling




Groundwater; Northern Territory; Western Davenport Water Control District; Conceptual mode

Publisher name

Northern Territory Governmnet

Place of publication



version 2.0


WRD Technical Report 27/2017


ix, 127 pages : colour illustration and maps ; 30 cm

File type





Attribution International 4.0 (CC BY 4.0)

Copyright owner

Northern Territory Government



Related links

https://hdl.handle.net/10070/842058 [LANT E-Publications: Development of a Groundwater Model for the Western Davenport Plains, version 1.1]

Parent handle


Citation address


Page content

Western Davenport WCD Groundwater Model (v2.0) Parameter Estimation CloudGMS 75 6 Parameter estimation 6.1. General parameter estimation (calibration) strategy Parameter estimation or calibration is a process, following model design and construction, by which parameters are adjusted until model predictions fit historical measurements or observations, so that the model can be accepted as a good representation of the physical system of interest (Barnett et al, 2010). In this study parameter estimation (calibration) was undertaken using a combination of the manual trial-and-error method and automated methods. Trial-and-error parameter estimation was used to assess the changes in the modelled response to changes in the model parameters as a kind of sensitivity analysis to inform the design of the automated parameter estimation process. In trial-and-error methods the model is run initially, using initial estimates of all model parameters, for the period during which historical data is available. Modelled heads are compared with observations both numerically and graphically to assess 'goodness of fit'. After each model run, with each new set of parameter values, the differences between runs are considered, and attempts to choose new parameter values that will in some sense bring the model predictions closer to all available measurements. For example: when hydraulic conductivities are increased, heads and gradients tend to decrease, and response times (lags) decrease. when storage coefficients are increased, the response to recharge or pumping is less, and response times increase. when recharge is increased, heads and gradients increase. Through the use of optimisation software, such as UCODE (Poeter and Hill, 1998) or PEST (Doherty, PEST Model-Indepedent Parameter Estimation User Manual Part I: PEST Software, 2016), the parameter estimation process is partially automated, with software doing much of the work on behalf of the modeller, and a rigorous mathematical methodology is applied that increases the reproducibility of the calibration process compared to trial-and-error calibration. Automatic parameter estimation is achieved by the optimisation software iteratively running the model, determining the objective function by statistically comparing the simulated values to observed values after each model simulation run and then updating adjustable model parameters to improve the fit between simulated and observed values. The iterative simulations continue until a 'best fit' between simulated values and observed values is attained based on minimising the objective function. Regardless of the technique employed all optimisation methods require: selection of a number of parameters to be estimated; an objective function, that is, a function of the measured values, defined such that its value is to be minimised; and constraints that somehow limit the possible choices for the values of the estimated parameters. 6.1.1. Optimisation code In this study automatic parameter estimation was undertaken using the PEST suite of modelindependent parameter estimation software (Doherty, 2010 and Doherty, 2013). The purpose of PEST, which is an acronym for Parameter ESTimation, is to assist in data interpretation, model calibration and predictive analysis (Doherty, 2010). The PEST code is open-source, public domain