The influence of model simplicity on uncertainty in the context of surface � groundwater modelling and integrated assessment

In the past, water resources in Australia have been over-allocated and this has led to the implementation of water reforms. An integrated assessment project is underway in the Namoi Catchment, located in New South Wales, Australia, that aims to assess the social, economic and ecological impacts of reductions in water allocations under current and future environmental conditions. Surface and groundwater resources in the Namoi are heavily utilised for irrigation agriculture and farming, are an essential source of drinking water and support ecologically significant riparian zones. The integrated assessment project considers the impacts of a range of climate scenarios and water policy options and investigates the benefits of adaptations in land management practises and innovations in water use efficiency. This paper focuses on the development of the hydrological component of the integrated assessment framework. The requirements of the hydrological model are that it should take climate and water extraction data and produce predictions of daily surface water flows and monthly groundwater levels. These hydrological impacts are transferred to an assessment of water availability and its social, economic and ecological outcomes. The paper discusses the trade-offs between model simplicity and complexity with respect to predictive uncertainty, equating model complexity with the number of parameters. A particular issue in hydrological modelling is that there is often insufficient data available to define the system, leading to underdetermined models and non-uniqueness of parameter values. The implication is that, while highly parameterised models generally contain enough flexibility to fit observation data well, they may have poor predictive performance due to parameter uncertainty. There are two main methods of dealing with the ‘complexity paradox’, the first being model parsimony and the second being the employment of a regularisation approach to stabilise the inversion of a highly parameterised model. While regularisation can be highly effective, there are costs associated with complexity that may include onerous data requirements and time-consuming development. The socioeconomic component of the integrated assessment project divides the Lower Namoi Catchment into a number of large zones that are considered homogenous with respect to the variables of interest. This paper argues that, for the purposes of the project, a simple model of surface and groundwater interactions is likely to perform as well as a more complex model when measured at the coarse spatial scale at which the results will be utilised. The proposed model utilises a catchment scale conceptualisation of the hydrological processes occurring, and is an extension of the IHACRES rainfall-runoff model. Finally, the paper presents a discussion of the uncertainty analysis approach that will be used to assess the performance of the model.