IBM using AI to help prevent Australia's beaches from washing away

Australia is home to more than 10,000 beaches, ranging from a few dozen metres to hundreds of kilometres long. But increasingly, these beaches are slowly disappearing before our eyes.

"Beaches across Australia are eroding, simply because waves come in pull sand away -- and big storm surges pull more sand away," IBM Systems Data Scientist Dr Adam Makarucha told the Gartner Application Architecture, Development, and Integration Summit in Sydney.

While the likes of Gold Coast Council have invested AU$14 million into rehabilitation projects -- such as one for a 12km stretch of beach, equating to more than a million dollars per kilometre -- Makarucha said prevention is more viable than rehabilitation.

Prevention, however, is challenging.

Makarucha said the best way to prevent beach erosion is to look to a natural defence, such as seagrass.

According to Makarucha, these underwater plants form large meadows that help to stabilise the sea floor and reduce wave energy, significantly reducing damage to coastlines.

But changes in the environment caused by increased wastewater entering the ocean has caused the degradation and loss of many seagrass meadows -- and they can take 50 years to re-establish.

See also: How AI and drones are trying to save the Great Barrier Reef (TechRepublic)

To ensure their survival, intervention plans require the ongoing monitoring of these meadows via underwater video footage and manual assessments by marine scientists.

This, he detailed, is very time-consuming.

"The first question you might have is, 'How do you monitor seagrass?', you can't fly a plane over and take photos, you can't see under the water, so what they do is they jump out in a boat, get in some scuba gear, and basically attach themselves to that boat, and it drags them along and they have a camera, and they record it all," Makarucha said.

"Once you record all this data, to actually get some analytics out of it, you have to work out how healthy the seagrass is, what type of seagrass it is, so they go through a manual process of watching the videos and assessing what type and what coverage there is.

"They do this over and over and over -- and you can imagine that with hours of footage, this could take quite a long time."

Working with a group of scientists that are conducting this work near Adelaide in South Australia, IBM and Adelaide-based digital agency KWP are helping them use AI for image segmentation so they no longer have to spend hours manually labelling video footage.  

Makarucha said the scientists built an application using Microsoft Access and Excel to store the data. He explained that as they played a video, they were clicking what type of seagrass they found, and what the density was.

"Really just a processing platform, they have no visualisation, they have no data storage ... and they effectively just store everything by sending each other emails," he explained. "Not a very scalable or usable platform. And It also takes them an enormous amount of effort to do this labelling -- five hours for a 10 hour video."

The scientists have over 500 hours of existing unprocessed footage.

IBM stepped in to help them build a new cloud-native application for this workload.

The scientists are now using AI to automate the process of labelling and identifying the type of seagrass and its density and coverage, taking the labelling time from eight hours to 20 minutes, with the AI model reaching an accuracy of 91%.

With additional data and training, the model accuracy is expected to increase.

"The over 500 hours of existing footage they have, well this system could process it in less than a day, in comparison to 10 person days. And If you extrapolate that 10 person days into actual work days, that's six and a half weeks of somebody constantly labelling seagrass," Makarucha said.

IBM took a hybrid cloud approach to AI on the IBM Power Systems AC922 using IBM Watson Machine Learning Accelerator Community Edition.

See also: How IBM Watson is revolutionizing 10 industries (TechRepublic)

"The benefits of a private cloud infrastructure is that we can get very specific compute very fast and high performance for accelerating these training jobs," Makarucha explained. "I think the most important thing is that you want to do your training where your data lives, and in this case, our data lives in the private cloud infrastructure."

"But then you want to deploy into the public cloud -- this is a really bursty workload, they don't go out in on the boats and scuba dive every day, they do it maybe once or twice a month, depending if there are sharks around or not," he continued.

"So because this work is bursty, they get maybe 10 hours of footage a month, we need flexibility in scale."

After going through this process, Makarucha said the scientists are looking at how they can use the data they've extracted to start predicting the health of seagrass and how it's going to change over time.

"And if they know where it's going to die, well, then they can start taking intervention methods more quickly and they have a process to implement this new idea very efficiently and very quickly," he said.

It took IBM and KWP two weeks to put the AI model in place.

"Being fast when talking about saving the environment is crucial," Makarucha said. "We don't have time if we lose the seagrass meadow -- that's 50 years we have to wait to get it back, that's too long. We need to act right now."

RELATED COVERAGE

• Does AI have the solution to climate change and human trafficking?
• Fujitsu using drones and analytics to protect native endangered species
• Australian universities use artificial intelligence to map global foreshore
• Planet analytics: big data, sustainability, and environmental impact
• Microsoft backs six Aussie environment-focused AI projects
• Australian universities use artificial intelligence to map global foreshore 
• Microsoft commits $50 million to apply AI to the problem of global climate change (TechRepublic)