Home / Our Science / Environment and Materials / Groundwater / Research Programmes / Smart Aquifer Models for Aquifer Management (SAM)

Smart Aquifer Models for Aquifer Management (SAM)

Figure1

Fit for purpose, smart models for aquifer management

The Smart Models for Aquifer Management (SAM) research programme is a GNS Science led collaboration with multiple organisations. The primary aim of the SAM programme is to identify optimal groundwater-surface water flow and transport models to address large scale, real-time, specific environmental management problems, that use data at hand, and can be used to inform targeted data collection to optimise them.  This work will provide a methodology to support judgements of what modelling strategy is most useful in any given data and decision making context, and to identify the gains and/or diminishing returns achieved with more data or more complex models.

Funder: New Zealand's Ministry of Business, Innovation & Employment

Co-funding: Waikato Regional Council, Greater Wellington Regional Council, Environment Southland

Primary collaborators: Victoria University of Wellington (VUW), National Institute of Water and Atmospheric Research (NIWA), Market Economics, Institute of Environmental and Scientific Research (ESR)

Additional collaborators: Beef and Lamb, CSIRO, Department of Conservation, Earth in Mind Ltd, Flinders University, Kitson Associates, Landwaterpeople Ltd, Ministry for the Environment, Ravensdown, Tubingen University, University of Waikato, Watermark Numerical Computing

Programme Duration: 2015-2018

Programme Leader: Catherine Moore

Programme Administrator: Abigail Lovett 

Background

Recent publication and white paper 

July 2017: “This short summary of the discussion paper titled "Simple and beautiful" was recently presented at the Australian Groundwater Conference workshop on Uncertainty Analysis. The workshop was sponsored by NCGRT Australia“

White paper -A discussion paper that outlines the background theory for the Smart Models for Aquifer Management project is currently in draft form ( SAM discussion paper) and will be updated over the next month.  Meanwhile it forms a theoretical and philosophical basis for the practical and numerical work being undertaken in this project in order to provide better tools for model based decision making throughout New Zealand.

The interactions between groundwater and surface water systems such as rivers, lakes, wetlands and estuaries are complex. Current models simulating these interactions are either too complex and slow to be practical or lack necessary integration, or are too simple to be accurate, yet the new National Policy Statement for Freshwater Management (NPS-FM, 2014) requires holistic freshwater management that satisfies community aspirations. This necessitates integrated groundwater-surface water modeling over larger areas and at finer spatial and temporal scales than ever before. The critical deficiencies of current modeling approaches therefore present a real risk to adaptive management of New Zealand's aquifers under the NPS-FM (2014).

Context

Modern-day environmental decision-making is largely based on numerical models.  It is recognised that the uncertainties associated with model predictions, “uncertainty analysis” should accompany model outputs.  Yet model-based decision-making still pays insufficient attention to the mathematical fact a model cannot tell us what will happen, but rather what will not happen, such that risk can be incorporated into the decision-making process.  This is because models are frequently far too complex for uncertainty analyses to be done. Their complexity promulgates numerical instability, drains modelling finances and time, and detracts rather than enhances, even intuitive assessment of model output uncertainty.

Figure1

Fit for purpose, smart models for aquifer management

Figure 2

Smart models need to consider all relevant model components in an impact assessment.

The SAM programme attempts to treat uncertainty as the fundamental context for environmental modelling rather than an afterthought. It recognises that it is often only one side of a predictive uncertainty distribution that is of interest, this being the side that assesses the possibility of unwanted events. Furthermore, it recognises that a model that is tuned to testing, and maybe rejecting, the hypothesis that such an event will occur, may not need to be complex, provided it is constructed specifically to explore that particular hypothesis. It also recognises that a simple model may contribute to predictive uncertainty through its very simplicity, and aims to take this into account.

By seeking to quantify uncertainty in contexts of data availability and integrity that prevail in New Zealand land-use management contexts, decision-makers can become better aware of what models can deliver through processing of that data. In particular, they can distinguish predictions that are accompanied by a large amount of irreducible (except perhaps at great cost) uncertainty, and those that are not. It is possible that in some instance, this information may prompt decision-makers to base policy and/or legislation on model outcomes that are of relatively high predictive integrity in comparison to those that are of low predictive integrity.

Programme modelling

The modelling within the programme follows two pathways:

1. best ways to train simple models from complex groundwater models in each catchment; and
2. development of simple model designs that do not need training on a complex model, that can be built easily in any catchment - but for which we have an estimate of their simplification error through some complex/simple studies in selected catchments.

To balance between these two paths, the project is designed to answer questions such as:

Figure3

Examples of four groups of meta-models. SAM will develop and identify the subset of meta-models which avoid predictive bias and inflated uncertainty, ‘smart meta-models’.

‒ To what extent can parameters employed by a simplified model be informed by measurable characteristics of catchment geological and soil components?

‒ To what extent must they be informed by local calibration?

‒ Can parameters of a simplified model inferred through calibration in one catchment be “regionalised” for the use of simplified models in neighbouring catchments?

‒ What simplification strategies are appropriate for the type of model outcomes we are considering? Appropriateness must take account of:

  • Ability to quantify predictive uncertainty while increasing it as little as possible;
  • The ability of simplified parameters to be informed by expert knowledge at a variety of scales as much as this as possible;
  • Reduction of calibration induced bias incurred through inappropriate simplification.

‒ How can calibration induced predictive bias of a simple model can be reduced through adoption of a “simplification-smart” history matching strategy?

CASE STUDIES:

Three integrated groundwater-surface water catchment studies provide the focus for this research programme. The key management decisions in each catchment, defined by end-users, will identify questions that the meta-models must address. Testing of meta-models across these case studies and additional international catchments ensures that programme outputs are relevant, workable, and transferable to a range of New Zealand groundwater-surface water contexts. These three catchments have been chosen for the case studies because: they all involve substantial groundwater-surface water interaction and exchange; they all include freshwater sites of national and international significance, such as Ramsar wetlands; they represent a diverse range of climate, hydrologic and hydrogeologic environments broadly representative of New Zealand; they have all been the subject of groundwater or surface water modelling based on ‘traditional’ approaches; and they are currently or in the near future will be undergoing the community engagement process for limit-setting under the NPS-FM (2014). It is vital to cover this range of test catchments to ensure that the research outputs will be relevant, tractable and transferable to other catchments across the country.

Hauraki:

  • Key stakeholders: Environment Waikato, the Hauraki Collective which represents 12 iwi, Dairy NZ, landowners, and community. 
  • The Piako catchment in the Hauraki Plains currently has fully allocated water resources.
  • Optimisation of the relative distributions of groundwater and surface water abstractions to determine whether changes in the distribution of these abstractions could better mitigate the occurrence of critical surface water low flows is one focus of this work.
  • Optimisation of land and water management strategies to mitigate high nutrient concentrations in groundwater and surface water bodies is another focus of this work.

 Ruamahanga:

  • Key stakeholders: Greater Wellington Regional Council, Kahugnunu ki Wairarapa, Rangatane o Wairarapa runanga, South Wairarapa District Council, Department of Conservation, landowners and community.
  • The Wairarapa catchment includes Lake Wairarapa, a proposed Ramsar site. The impact of land and water management scenarios on lake and wetland habitat will be examined.  Note this work requires integration with lake models.
  • The project will also determine direct groundwater inputs for the University of Waikato project at five lakes in the Wairarapa, including Lake Wairarapa.

 Southland:

  • Key stakeholders: Environment Southland, Hokonui Runanga, Awarua Runanga, landowners and community. Focus areas are the Mid-Mataura/Waimea; Oreti; and Waimatuku catchments.
  • Environment Southland aim to provide robust, rapid and flexible simplified models of groundwater and surface water flow and transport, for a range of scales and timeframes. 
  • Determining optimal land and water management strategies that enable reduction of contaminant load to estuaries and coastal wetlands, in addition to managing for contaminant concentrations in specified river reaches, is one particular focus of this work.
  • Models must cater for the highly transient, climate event-driven fluxes of contaminants through the catchment.