GSTDTAP  > 资源环境科学
Research Highlight: Using Artificial Intelligence to Understand Deepwater Horizon’s Effects on Dolphins
admin
2018-06-27
发布年2018
语种英语
国家美国
领域资源环境
正文(英文)
Dolphins swim through oil-tainted waters in Gulf of Mexico in 2010 following Deepwater Horizon spill. Photo: NOAA

Dolphins swim through oil-tainted waters in Gulf of Mexico in 2010 following Deepwater Horizon spill. Photo: NOAA

A project born in response to the 2010 Deepwater Horizon oil spill has produced a new method that lets scientists sort through and classify millions of dolphin sounds more rapidly than ever before.

Newly developed algorithms by Kait Frasier, an oceanographer in the Whale Acoustics Lab at Scripps Institution of Oceanography at the University of California San Diego, may eventually provide scientists clues about how marine mammal populations  respond to oil spills and other major disruptions over longer time scales.

As part of a U.S. science effort in response to the April 2010 incident, Scripps oceanographer John Hildebrand and his team, including Frasier, began making recordings of dolphin sounds during the spill, using High-Frequency Acoustic Recording Packages (HARPs) developed at Scripps Oceanography. Over an eight-year period, the team made recordings of dolphin echolocation sounds, or ‘clicks’, at five locations around the Gulf of Mexico. Clicks are short, impulsive sounds used by various dolphin species as sonar to locate and capture prey.  Previous research by the group has found that different species of dolphins can display distinctive features in their clicks, when large numbers of samples are collected. The sounds can be used by humans to monitor geographical distribution and potentially population size.

Fraiser’s artificial intelligence algorithm uses ‘unsupervised’ learning, which means that the algorithm created its own categories for assigning different types of clicks, as opposed to ‘supervised learning’, in which large amounts of data must be ‘pre-labeled’ by human analysts before training the algorithm.  She applied the algorithm to 50 terabytes of data representing 52 million dolphin echolocation sounds, or ‘clicks’.

That was just the first two years’ worth of data.

“It took the algorithm about two weeks to process and sort clicks and assigning them to a type,” said Frasier. “It would have taken an expert human analyst about six months to do that, and it would have taken a year to train that person to do such a task consistently.”

The Deepwater Horizon disaster left 11 rig workers dead and is considered one of the largest oil spills in the history of the petroleum industry. The federal government estimated that 210 million gallons of oil spilled before the well was declared sealed five months after the initial explosion aboard the rig.

Dolphins are an indicator species, meaning that the health of dolphin populations often corresponds to the health of the ecosystem as a whole. In the Gulf of Mexico, which covers more than 1,550,000 square kilometers (600,000 square miles), most dolphin species range great distances. They also rely on the entire water column for survival, breathing air at the surface and feeding on marine organisms that can live up to several hundred meters below the surface.

Dolphins were among several marine creatures adversely affected by the spill. Researchers at other institutions identified what is termed an unusual mortality event among bottlenose dolphins in the wake of the spill and reported that a number of stranded dolphins found dead had developed lesions in their adrenal glands and lungs far out of proportion to dolphins outside the spill zone.

Frasier published an account of the development of the algorithm in the journal PLOS Computational Biology in December 2017. She said there are now several more steps to be taken before she can make inferences about how the spill affected dolphin activity or changes in the populations of various species in the gulf.  In total, her research team must work through half a petabyte (roughly 500,000 gigabytes) of data.

“Before we can get at the underlying ‘what has happened to them’ question,” Frasier said, “we have to figure out what species are we hearing in this massive dataset and how we identify the detected signals – billions of them at this point – consistently in the constantly changing soundscape.”

The next stage of the project will help correlate the categories of sounds identified by Frasier’s computer program to the 14 species of dolphins found in the Gulf of Mexico. Researchers will use a combination of visual identification and recordings from a towed surface array where a hydrophone similar to those in the HARP packages is towed behind a research vessel, recording vocalizations. Scientists onboard can then match the recordings to observations they made.

The distinct vocalizations of some species such as Risso’s dolphins are well-understood. Frasier said she is reasonably sure of some others, and is working to assign the rest.

“Whales and dolphins are hard to study because they're offshore and underwater, but at least they make sounds and come to the surface to breathe, so we can get some good information from them,” said Frasier. “The things they feed on, like squid and fish, and the things that their prey feed on, are incredibly hard to study, [being] deep and small, and they don't come to the surface. When times are good at depth, dolphins thrive. When something isn't right in an ecosystem, they reflect that.”

Dolphins live for 20 to 40 years. They reach sexual maturity between 5 and 10 years of age, meaning that the young born around the time of the Deepwater Horizon disaster are just old enough to breed now. Females of most species don’t breed every year, so affected populations are slow to recover. The most profound effects of the oil spill might only just be beginning to appear. The researchers said that this is what makes these long-term projects necessary and important.

“By rapidly deploying acoustic sensors in the Gulf of Mexico, we have been able to track changes in the marine mammal community following the Deepwater Horizon spill. We have also developed new tools for rapid analysis of these data, so that we are better prepared should another disaster strike,” said Hildebrand.

Now that Frasier has demonstrated that the algorithm approach speeds up the processing of large datasets, it will be applied to other projects from around the world. On top of the remaining years of data recorded in the Gulf of Mexico, projects in the Arctic, Antarctic, Hawaii, and along both coasts of the United States will be analyzed using the new algorithm. 

Interested in hearing the dolphin sounds discussed in this project? Josh Jones, a graduate student in the Whale Acoustics Lab, created the website Voices in the Sea. The sounds of marine mammals including whales, dolphins, and seals are included. The interactive features have also been developed into a kiosk at aquariums around the country.

 

– Melissa Miller

 

This story appears in explorations now, Scripps Institution of Oceanography's award-winning ocean and earth science magazine. Sign up to receive our free monthly story roundup.
URL查看原文
来源平台Scripps Institution of Oceanography
文献类型新闻
条目标识符http://119.78.100.173/C666/handle/2XK7JSWQ/107374
专题资源环境科学
推荐引用方式
GB/T 7714
admin. Research Highlight: Using Artificial Intelligence to Understand Deepwater Horizon’s Effects on Dolphins. 2018.
条目包含的文件
条目无相关文件。
个性服务
推荐该条目
保存到收藏夹
查看访问统计
导出为Endnote文件
谷歌学术
谷歌学术中相似的文章
[admin]的文章
百度学术
百度学术中相似的文章
[admin]的文章
必应学术
必应学术中相似的文章
[admin]的文章
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。