**5. Discussion**

#### *5.1. Benefits of The Integrated Modelling Approach*

For the irrigation areas with deep vadose zones in which perching can occur, the only viable approach before the development of the TF models was an inverse approach in which the aquifer conductivity values and then recharge values were inferred (with some limited adjustments of conductivity) on the basis of observed groundwater responses [21]. The integrated modelling approach clearly addresses a deficiency of the inverse approach, namely, not being able to transparently link actions to salt load impacts. In addition, the inverse method requires a complex process for calibrating recharge over many zones and stress periods, and may have few other datasets to constrain what is a non-unique problem. A careful logic and many groundwater model runs are required to provide sensible values. The integrated modelling approach is simpler than this because it is objectively formulated.

The results demonstrate the impact that recharge input values in a groundwater model has on the salt flux to river estimates used for register entries. Given a scenario where there is limited data to constrain recharge values, use of the integrated method provides some means to quantify the reduction in uncertainty of recharge estimates via automated calibration software. There is presently no method to fully quantify reductions to uncertainty using manual calibration. Adopting the transfer function for future assessment promotes transparency and repeatability of the modelling approach. Moreover, it will promote consistency between groundwater models used for other irrigation areas by reducing subjectivity of the calibration process.

#### *5.2. Limitations of the Integrated Modelling Approach*

The trial demonstrated that the integrated model may require an increase in model development time. The extent of the increase will be dependent on the modeller's familiarity with a scripting language such as Python and/or an automated calibration utility such as PEST.

Initially, the method will require time and effort to upskill users and to incorporate routinely into current workflow. However, this additional cost will reduce over time once the approach and the linking script are developed for a particular model.

A lack of data may still affect the success of applying the integrated approach. Drainage data for the Loxton district provided some constraint on the outputs of both the AWB and transfer function in this trial. In addition, attempts to minimize the difference between simulated drainage in the AWB and transfer function provided a further constraint on the calibration of irrigation application efficiencies and clay conductance.

Access to high performance computing is also strongly recommended if the integrated approach is to be used with automated calibration.
