Case Studies

Learn more about our current case studies in collaboration with other organisations

Flood Forecasting System

This project began after the severe flooding during the Queen’s Birthday weekend in 2020, where forecasting limitations impacted Civil Defence and local councils' ability to prepare effectively. Recent events, like Cyclone Gabrielle in early 2023, have further underscored the urgency for improved flood resilience and preparedness.

Building on our successful implementation of a flood forecasting system for three rivers in the Coromandel, we are now expanding our platform to cover regions across New Zealand. In partnership with MetService and regional councils, the platform provides real-time access to Doppler radar and environmental monitoring data, enabling accurate flood predictions. Designed for ease of use, it allows environmental scientists to access powerful forecasting tools without needing advanced machine learning expertise.

By utilising advanced machine learning algorithms, such as Transformers, we’ve significantly improved the accuracy and extended the forecast horizon from 3 to 6 hours. This extended prediction window enables Civil Defence and regional councils to better prepare and allocate resources during flood events, as demonstrated by its effectiveness during the Auckland floods in 2023.

Forest Flows

The AI Institute is collaborating with Scion on the Forest Flow Research Programme (FF), which focuses on real-time forest monitoring to support sustainable timber production in New Zealand. Using advanced data collection methods, the research aims to enhance forest hydrology monitoring and tree growth forecasting through machine learning (ML) and explainable AI (xAI) techniques.

We are currently improving the forecasting model and implementing xAI methods. Enhancements to the model include optimising data resolution and integrating new data to increase accuracy and adaptability to weather changes. xAI methods have been applied to understand the driving factors of tree growth, providing valuable insights that align with existing forestry knowledge and contribute to ongoing scientific discussions.

The improved models and methods will aid environmental scientists and forest managers in making more accurate forecasts of tree radial growth and understanding the underlying factors, contributing to more sustainable forest management practices.

DOC Habitat annotator and classifier

New Zealand's fragile reef ecosystems require careful protection, and analysing the growing volume of marine imaging data is critical for this effort. To address this, we are collaborating with the Department of Conservation (DOC) to develop an online image annotation platform, designed for efficient and collaborative analysis of deep-sea photographic and video data.

This platform simplifies the time-consuming annotation process, offering an intuitive and flexible tool for cataloguing and managing large video libraries. By incorporating DOC’s feedback, we’ve enhanced productivity and streamlined the annotation workflow.

We are currently refining the platform to improve the accuracy of classifications using interactive machine learning. Our goal is to provide a versatile solution for not just DOC, but any organisation in New Zealand working with extensive video and image datasets.

Annotation and classification system for Trail Cam images

 

New Zealand's unique bird species, including the iconic kiwi, have suffered greatly due to introduced predators. A proven solution for protecting these species is the creation of fenced sanctuaries, but these require constant monitoring to prevent predator breaches. Sanctuary Mountain Maungatautari, an ecological sanctuary with a predator-proof fence, plays a vital role in kiwi conservation by providing a safe environment for breeding and protection. However, monitoring the sanctuary's perimeter, using motion-activated cameras, is a significant challenge.

 

In collaboration with Sanctuary Mountain Maungatautari, we are developing an online image annotation platform that streamlines the review of footage from these cameras. This platform allows easy, collaborative access to images and videos, making the annotation process more efficient and less time-consuming.