Groundwater is an important source of drinking water in Aotearoa New Zealand but it is expensive and difficult to collect samples and data. Complex models paired with cutting edge statistical and mathematical tools and techniques are used to understand groundwater systems, but they must be tailored to the specific environment and questions being asked.
ESR Science Leader Theo Sarris and a team of hydrologists, modellers and data scientists, work on these complex models to help manage groundwater: studying contamination risks, mitigation options for reducing nutrient levels, management of groundwater resources and integration of AI tools into groundwater hydrology.
Although groundwater connects to our surface water systems and is a major source of drinking water drinking for Aotearoa New Zealand, it is unseen and often difficult to access. Sampling groundwater – drilling wells the surface, through the water-table and hundreds of meters below – is a costly and slow process. Researching groundwater systems through sophisticated models can leverage these expensive ‘data points’ of groundwater sampling and incorporate the vast gaps with statistical descriptions. For these models to be useful in understanding and managing our groundwater, they must be specific to the environment and to the questions being asked.
Model inversion, also known as inverse modelling, is a technique used in scientific research to estimate the unknown parameters or inputs of a mathematical model based on observed data or output. It involves running the model in reverse, starting with the observed output and working backward to determine the input parameters that would have produced that output.
Model inversion is used in many fields, including atmospheric science, oceanography, geophysics, and environmental engineering. For example, in atmospheric science, model inversion can be used to estimate the emissions of greenhouse gases from sources such as power plants, factories, and transportation, based on the concentrations of these gases measured in the atmosphere.
The process of model inversion involves using optimisation algorithms to minimize the difference between the model output and the observed data, by adjusting the input parameters of the model. This can be a challenging task, as many models have a large number of parameters that can interact in complex ways. However, model inversion is a powerful tool for understanding the underlying processes that govern a system, and for making predictions or designing interventions based on that understanding.
Dr Theo Sarris and the ESR team develop highly complex models to simulate aquifer structures and how these affect the movement of contaminants and pathogens. The goal is to provide insights and predictions that would be difficult or impossible to obtain through observation alone. They can help us understand the behaviour of water and contaminants in the subsurface, informing resource management decisions and risk assessments.
In one example, Theo and the team investigated diffuse agricultural nitrate leaching and nitrate removal within a catchment basin with only limited knowledge of the complexity of chemical heterogeneity - the variation in the complex chemical composition of soil due to differences in the mineral content, organic matter, and pH levels across different locations. The model used a hybrid catchment-scale flow and transport model paired with a Machine Learning model to evaluate the effectiveness of targeted area regulation. The study showed that when land use decisions are science based, outcomes (in this case better surface and groundwater quality without minimising economic activity) will always be improved. If management zones are delineated based on chemical heterogeneity and groundwater flow paths, land use regulation in discharge-sensitive zones was twice as efficient compared to other management options.
These models are an important tool for understanding complex groundwater systems and making informed decisions, but they should always be used with caution and with an awareness of their limitations.
Monte Carlo is a statistical method used to estimate an uncertain outcome by simulating many equally possible scenarios. It is commonly used in various fields, including finance, engineering, physics, and computer science.
The method involves generating a large number of random samples, each with its own set of parameters or inputs, based on a prior knowledge. These samples are then used to simulate the outcome of a system or model. By running the simulation many times, the Monte Carlo method can provide estimates of the likelihood and range of potential outcomes for a given scenario, along with associated levels of uncertainty.
Monte Carlo simulations can help to assess the risks and potential impacts of complex systems or models, by allowing researchers or practitioners to identify the most critical parameters or inputs and evaluate different scenarios. The method is often used in conjunction with sensitivity analysis, which helps to identify which parameters or inputs have the most significant impact on the outcome of the simulation.
As Monte Carlo analysis involves a large number of simulations, it is a computer-intensive process and can take considerable time and computational effort to conduct.