Using citizen science and artificial intelligence to help conserve the Great Barrier Reef

Stretched over 344,000 square kilometres, the Great Barrier Reef in Australia is the biggest coral reef system in the world and an important marine biodiversity hotspot. But, the natural wonder that can be seen from space is in poor health. Now, researchers are looking to leverage citizen science crowdsourcing and artificial intelligence to help plan conservation efforts better.
According to a study published in 2012, the Great Barrier Reef has lost over half its coral since 1985. Added to that, climate change has caused the reef to suffer from various “bleaching events,” or when the coral expels the algae living in its tissue due to the water being too warm. When coral reefs bleach, they are not yet dead. They can still recover if conditions improve, but it is estimated that it takes them up to 12 years.
Don’t Miss: Apple Event Live update from Apple Park
While conservation efforts are ongoing for the Great Barrier Reef, there is another issue. The sheer size of the reef system and the land that it occupies means that these efforts can only be focused on small sections. According to Andy Ridley, CEO and founder of Citizens of Great Barrier Reef, which conducts the Great Reef Census, researchers have only been able to monitor around five to ten per cent of the 3,000 individual reefs in the system in the past. This makes informed conservation decisions difficult.

The Great Reef Census helps solve this issue using thousands of images captured volunteers on dive boats, tourism vessels, yachts, fishing charters, and tug boats to survey images from the Reef. It then uses these images to understand the health of the reefs and to identify the locations where conservation efforts must be focused. As part of the project, researchers from the University of Queensland, Dell Technologies, and other organisations have partnered to leverage artificial intelligence and crowdsourcing to help understand the reef better. Two Great Reef censuses have already been conducted.
In the second census, the researchers first crowdsourced over 40,000 images of coral in the Great Barrier Reef from thousands of volunteers. Then, the challenge was tagging and identifying the coral species within the images to see which ones needed to be studied marine biologs. Volunteers then tagged and identified various different coral species, which turned out to be a time-consuming process.
That is where the deep learning model developed Dell Technologies came in. Using this data from the volunteers, Dell trained a deep learning model to analyse every pixel in an image to classify the reel and coral infrastructure.

“The deep learning model first identifies which part of the image contains coral. And then, it is able to classify them if they belong to five selected species of coral. It is limited to five right now to maintain accuracy. This data, combined with the location data of the image can provide researchers with a snapshot of what kind of corals ex where. The researchers can then ascertain the health of the coral from selected images. They can then use this data to prioritise which reef must be conserved first,” said Aruna Kolluru, chief technolog, Emerging Technology at Dell, to over a video interaction.
This means that the deep learning model can take on the repetitive and tedious task of identifying and classifying the coral, freeing up volunteers for tasks that still require human input. These images can be analysed marine biology experts if necessary based on this identification and classification. Currently, the model takes one minute to analyse an image.
According to Dell’s Kolluru, the deep learning model can further be improved to not only analyse images more accurately but to also be able to identify more species of coral. But for that, it needs to be trained on much larger numbers of images and data points.

“For example, if you want to train a model to identify a car, you need to train it on many different kinds of images. It could be images of different parts of the car, like the tyres or the doors. It would also need to be trained on images of cars from different angles and with different lighting conditions. With all this data on how a car looks like in different images, it will better understand how to identify a car,” explained Kolluru.
As the model develops further, researchers envision using it to protect other reef systems globally. “We hope to expand the Census globally and build on the globally-available Allen coral atlas so that everyone can develop data on the most ecologically important reefs in their jurisdiction. The deep learning model will help standardise and support the ability of citizens to map the state of reefs worldwide,” said Peter Mum, a coral reef ecolog at the University of Queensland, to over email. Mum is part of the Great Reef Census team.
Ridley of Citizens of Great Barrier Reef told that the third Great Reef Census will kick off in October. As the researchers develop the census’s methods, they are also ensuring these methods and technology are scalable and can be used to conduct similar research with other reef systems across the globe.