Algae occur almost anywhere there is water. Algae live in soil, snow, hot springs, on animals, coral and jellyfish and even in clouds.
They are an essential part of aquatic ecosystems, providing food for other organisms. But arguably their most important function is the production of oxygen. Algae are the source of more than 50 per cent of the oxygen we breathe. This is because they convert carbon dioxide into oxygen by photosynthesis.
Dr. Chris Jackett is a postdoctoral scientist with our Machine Learning and Artificial Intelligence Future Science Platform (MLAI FSP).
“Algae are fascinating. They occur in many forms, including free-floating cells, chains of cells, large amorphous blobs and larger structures such as seaweed,” Chris said.
“Every second breath we take is thanks to photosynthetic marine algae.”
Harmful algal blooms
In the presence of concentrated nutrient sources, some species of algae can grow exponentially, forming harmful algal blooms (HABs). These blooms can have harmful effects on people, birds, marine mammals, fish, shellfish and other aquatic organisms. In addition, some species that can form HABs also produce toxins.
Blooms occur in marine and freshwater environments throughout the world, with damaging ecological, social, and economic effects.
For example, on the east coast of Tasmania in 2012, a large harmful algal bloom of the dinoflagellate Alexandrium catenella caused the closure of the seafood industry.
In 2016, toxic blue-green algae bloomed along 1700 kilometres of the Murray-Darling River. As a result, the river water was not suitable for drinking, agriculture or recreation.
To predict and manage these events we need to be able to identify harmful algae.
Identifying algae using artificial intelligence
Identifying algae can be slow and labour-intensive. First, water samples are collected in the field and delivered to a lab. Next, an analyst views each sample under a microscope and identifies the species present. Each sample can take more than an hour to analyse and the results can be inconsistent between different analysts. This means it is difficult to get accurate results and to analyse samples from large areas as frequently as desired.
To solve these problems, we are using machine learning to train computers to automatically detect important types of harmful algae. This involves collecting a large image dataset. We then manually annotate the data by drawing bounding polygons around each algae cell. Next, we iteratively train deep neural networks to learn structures and patterns in the data.
“We are building smart machines capable of performing tasks that typically require human experts, such as being able to differentiate between species of algae that look extremely similar,” Chris said.
“We are taking thousands of photos of our living algae collection at ANACC and training AI systems to detect and classify different species of harmful algae. We’re also developing human-in-the-loop systems that can identify where the machine learning model is not performing well, allowing algae experts to improve the model.
“Using an efficient human-centered AI approach, coupled with powerful deep neural networks, we are combining the best capabilities of humans and computers to build effective detection systems,” he said.
Early warning systems for harmful algae
Our long-term goal is to create a mobile digital system that will allow us to identify what types of algae are in the environment. We will be able to use this both remotely and on location.
The system will be integrated with our other environmental monitoring approaches, including molecular identification tools and imaging flow cytometers.
Detecting the presence of harmful algae faster will potentially provide us with an early warning system of when and where blooms might occur.
This information can be used to keep people and livestock safe from harmful algae and help the seafood industry safeguard their stock.