Artificial Intelligence and Freshwater Modelling

An Integrated Framework Promoting Sustainable Water Management

Freshwater quality modelling is pivotal for resource management and decision-making. Its integration into the freshwater accounting and reporting system enhances transparency of freshwater management. However, despite its growing importance, current practices face challenges in effectiveness and risk losing trusts on model-based evidence.

Our programme will develop a pioneering AI-powered framework and platform tailored for process-based freshwater quality modelling. Not only will our platform streamline and integrate machine-learning capabilities, but by standardising both modelling practice and data flow, it will reduce the burden of modelling projects. This will enable more robust ensemble modelling to guide decisions; an approach informed by the widely accepted practices of the IPCC for climate modelling and local weather forecasting.  Our overarching goal is not just to enhance water quality modelling in Aotearoa New Zealand but to initiate a transformation towards continuous improvement, promoting a) stochastic results projection, b) objectively identifying knowledge gaps and priority investment areas, c) supporting participatory modelling practices, and d) building trust and confidence in the use of models.

Current approaches to process-based freshwater quality modelling revolve around isolated projects that are fit for their location-specific objective. The lack of standardisation practices makes the production of readily transferable knowledge and scalable evidence particularly hard. The models are often developed using short term observations, lacking adaptability to identify and quantify uncertainties from not previously observed conditions such as those predicted under climate change scenarios. Resource constraints often limit the use of best practices like ensemble modelling, confining decisions to the outcomes of a single model, making it difficult to assess decision risks. Added complexity of the process also frequently excludes the community, hindering participatory processes. Machine-learning techniques are commonly utilized in freshwater quality modelling, particularly for reporting. Yet, these data-driven models restricted by the data they are trained with. Given the currently available information, they primarily provide broad spatiotemporal forecasts, and the constraints become especially pronounced when predicting under climate change scenarios.

In response, our platform seeks to bridge these methodologies, addressing their individual shortcomings. Artificial intelligence (AI) can identify transferrable and generalizable patterns, concepts and knowledge. It also possesses an ability to learn continuously, adapting to changed conditions and improving over time. However, it faces challenges with limited, niche data, much like the current data-driven models face. The physical processes and expert knowledge encoded in the process-based models are an extremely powerful asset if incorporated into a hybrid AI system, helping it overcome the limitation. The platform, powered by AI, will enable more sophisticated post-modelling processes. This includes offering a simplified “metamodel” for the community to engage with, delivering visualized insights, and evaluating uncertainty along with identifying key knowledge/data gaps.

In this project, we will develop a novel standardisation framework and platform for freshwater quality modelling that integrates process-based and machine learning models. This will contribute to enhancing our understanding of environmental impacts on freshwater quality required for decision making, reducing burdens of and resources required for individual modelling projects, and driving the implementation of Te Mana o te Wai. The main objectives are:

  • Develop knowledge production practices for freshwater modelling to better represent and serve Māori. Through Indigenous Data Sovereignty, Digital Toi Māori, and Empowering Mana Motuhake, the practices will address community engagement, place-based designs and freshwater decision-making aligned with Māori values.
  • Create a product that aligns and facilitates implementation of the Freshwater Accounting requirements (National Policy Statement for the Freshwater Management) and Future Focused Freshwater Accounting (MfE).
  • Improve process-based models by integrating artificial intelligence (AI) and machine learning techniques, addressing data limitations, and developing suitable transfer learning algorithms to exchange knowledge between models.
  • Develop a distributed framework for unified access to freshwater models, allowing high and low-level access, compensating for software diversity in freshwater modelling. Due to the distributed nature of the framework, we can effectively address data sovereignty and intellectual property questions and provided tailored solutions while providing benefits to all users.
  • Demonstrate the practicality of the framework by addressing pressing environmental challenges, including transfer of discharge intensity across different sites, adaptation to changes such as in land use, due to climate change, or extreme events, and the identification of effective yet minimal mitigation strategies.
  • Focus on impact-specific projects, collaborate with regulators, and work with communities to validate the framework’s effectiveness across diverse national contexts and environmental scenarios.
  • Develop strategies to ensure the impact and implementation of the modelling platform, including fit-for-purpose evaluation, uptake strategy development, stakeholder engagement, and future-proofing.
  • Prioritise accessibility and usability by creating a comprehensive API, documentation, tutorials, and training workshops, translating materials into Te Reo, and enhancing the user interface and visualizations for wider usability.
  • Enable enhanced post-modelling processes, such as model accessibility through meta-modelling, improved visualisation modelling assumptions, uncertainty, and knowledge gaps (global sensitivity analysis).
  • Collaboratively investigate the feasibility of scaling up towards generating a crude national scale water quality model for the entire New Zealand using the framework and learning from the configuration at the pilot case studies, demonstrating the transferability of the knowledge.

Our transdisciplinary team includes leading experts from academia, Crown Research Institutes, Councils, and industry.  We have diverse expertise in Computer Science, Artificial Intelligence, Environmental and Freshwater Science, Agricultural and Catchment Systems Modelling, Māori Environmental Science, and Policy and Environmental Social Science. Our website lists our collaborators.

For more information, get in touch.