Armed with artificial intelligence, scientists take on climate change
No time to read? You can listen to this article here:
Science needs to understand and predict how climate change—and the growing onslaught of hurricanes, fires, and floods it’s bringing—affects tropical forests. Will the forests respond to the assault with shorter trees? Will they store less carbon, or support less tree and plant diversity and fewer wildlife species?
To better understand the effects a changing climate will have on tropical forests, Maria Uriarte, Columbia University professor of ecology, evolution, and environmental biology, needs to analyze images of forests. These bird's-eye view images are the size of a postage stamp. And she’s working with enough postage stamp-size images to cover a 28,000-acre national park. Since it would take her years to analyze even a dozen of those images, there’s no way she could handle enough to get a real understanding of an entire forest.
Until recently, that is.
Analyzing images faster
Uriarte turned to artificial intelligence to advance her research—to better understand how forests are responding to a changing climate, and to predict how they’ll react in the coming years.
“With AI and deep learning, we can detect different species in the photographs by color and shape. It can do that very well and very quickly,” says Uriarte. “Our algorithm does better than the people who are best at finding these species. It would be impossible for me.… The AI system could analyze the images in days. And it allows us to ask questions at a scale we couldn’t hope to do without it. To ask these historical questions and questions specific to certain hurricanes, topography, and climate, it tells us what we can expect to see in the future. It completely changes the scale of what we can do.”
For example, using AI has enabled Uriarte’s team to identify that the Sierra palm is highly resistant to hurricanes. The tree species remains abundant in areas hit hard by two different storms.
Uriarte is just one of many scientists around the globe who are using AI to study climate change.
As scientists’ understanding of climate change grows, so does concern about the global phenomenon and the need to slow, stop, or even turn it around. With melting land ice and the expansion of warming seawater, sea levels are expected to rise one to four feet by 2100 because of the rapid, human-induced warming, according to a NASA report. Hurricanes are expected to hit more frequently and be more intense, according to a Climate Science Special Report released in 2017. The Arctic Ocean, scientists predict, could be essentially ice-free in summer by mid-century. Heat waves are expected to continue to rise, scorching agricultural fields and cities, while taxing our resources to counter the increasing temperatures.
Finding the needle in the haystack
Researchers are throwing their weight into better understanding the changing climate and finding ways to stall or even reverse it because of the speed at which the climate is changing and its effects on everything from rising sea levels to agriculture, wildlife, storm patterns, shrinking ice sheets, droughts, and human populations.
And with such a massive undertaking, scientists need the most powerful tools they can get their hands on. Massive databases, data analysis tools, and high-performance computing are all being used. But AI, often feared as a threat to humanity, is turning out to be one of the most powerful tools scientists are using to protect the planet.
“I don’t think scientists can take on climate change without AI,” says Dave Schubmehl, research director of cognitive and AI systems at IDC. “They’re going to need these advanced algorithms to handle what humans just can’t. As you get more and more volumes of data, it becomes less likely that humans can find the needle in the haystack themselves. They’ll need AI to do it and to make any progress with a job this massive.”
The amount of data is staggering. Vast amounts of information are coming in from global sensors, weather stations, satellites, radar, and buoys. Schubmehl estimates it goes well beyond trillions of data points and is expected to only continue to grow. Humans and traditional computer programs simply couldn’t sort through it all.
AI is made for the job because it can take on tasks that would be impossible for humans to do in multiples of the same amount of time. As more large-scale data becomes available to scientists, the role of AI grows increasingly important in enabling the discovery of patterns, recognizing images, assessing different strategies and policies, and developing predictions from billions of observations.
Tracking the success of crop changes
For researchers at Atlas AI, a start-up founded by a team of Stanford University professors, AI provides insight into climate-related issues like deforestation and crop yields by analyzing satellite imagery.
“We’d be overwhelmed by imagery with not enough people to process it,” says Temina Madon, director of business development at Atlas AI, which, without AI, would need 6 million highly trained people to analyze images from just three satellites. “You get massive geospatial datasets being beamed down to Earth, and they’re wonderful assets but no good to us if we don’t have the human resources to process that information and interpret what that means for climate change. And that information is a critical part of the overall picture.”
Calling AI the DNA of the company, Atlas AI CEO Victoria Coleman says AI is helping the company track the success of crop changes in East Africa by analyzing satellite images. Some farmers in Tanzania and Kenya have swapped out their normal crops for special varieties of maize and rice that are believed to be able to handle both drought and floods. Atlas AI is helping to figure out if that’s working.
“The [different crop varieties] work in a lab, but whether they’re more productive under drought or flood in the real world is still a question,” says Madon, adding that Atlas AI is using its own AI systems. “We get these massive sets of satellite images that show the African continent at relatively high resolution and frequency. AI can spot these small agricultural plots and chart their progress. [We] couldn’t do it with humans at that scale. If you multiply that across a whole country, we can generate a prediction of how much those countries will produce so the governments can use that information to decide if they have to import grain, and they can start to understand what the impact of climate change is in different parts of their countries.”
Atlas AI has shared its information with Kenya’s representative to the United Nations and the country’s minister of agriculture.
Adam Brandt, associate professor of energy resources engineering at Stanford University, is mostly using AI for computer vision approaches to detecting oil and gas air emissions, including methane, a potent greenhouse gas.
“AI is integral to that research stream and holds promise to greatly reduce the cost of detecting methane leakage,” Brandt says. “This will, we hope, make it more cost effective to drive emissions down to much lower levels than currently seen. Since the research is really focused on automation and avoiding the need for human operators, due to the expense of trained operators, we could not perform this work without AI.”
Making climate predictions that span decades
AI is proving to be a key tool for scientists working on climate change predictions, notes Dr. Eng Lim Goh, vice president and chief technology officer for high-performance computing and AI at Hewlett Packard Enterprise.
Unlike weather forecasting, which is more local and has a day-to-day time frame, climate modeling makes global predictions spanning decades and uses enormous amounts of corresponding data. With so much data and a myriad of factors to consider, it’s often difficult for us to ensure that all relevant causes are included. AI, according to Goh, can be a valuable tool to augment humans in taking care of this.
“The accuracy of traditional top-down rules-based or laws-based approaches is only as good as the scientists having accounted for all factors that contribute to the prediction,” Goh says. “This is not a simple endeavor. For example, after accounting for heat absorbed by our oceans, scientists also have to remember to account for heat reflected by our ice sheets. After accounting for emissions caused by human activity, they need to remember to include livestock emissions, and so on.”
Here machine learning can support traditional climate modeling. “Machine learning methods come at it from a bottom-up approach. Instead of trying to figure out all the factors, it learns from historically generated and recorded data,” says Goh. One way it does this is a sophisticated neural network with tunable dials for each connection. Historical data is fed into the neural network, and “a wild guess is made at first,” says Goh. “If it’s wrong, the dials are tweaked.” Eventually, after massive amounts of quality data is fed into the network and tweaks are made, a good prediction about the future may be possible.
Informing critical climate decisions, policy
Lloyd A. Treinish, a distinguished engineer and chief scientist at IBM’s Thomas J. Watson Research Center, says AI will be increasingly used in climate change research and in helping people make critical decisions.
“Ultimately, these are tools that can help people who are on the policy or business side understand the implications of the choices they are making,” says Treinish, who has worked with city officials and utility companies on long-term sustainability issues. “[AI’s] biggest potential role right now is in the decision-making aspects, trying to assess the implications of a changing climate and understand how choices affect it. It’s going to be a matter of policy to make the changes needed to reduce the effects of a changing climate.… We’d use AI to help them make decisions.”
In India, researchers are using AI to both reduce costs and carbon emissions related to electricity usage, using a grant from Microsoft’s AI for Earth project, according to Lucas Joppa, Microsoft’s chief environmental officer. AI for Earth has issued more than 200 grants for a total of more than $50 million in a five-year commitment to deploy AI to solve environmental challenges around the planet.
James Rising, an assistant professorial research fellow at the London School of Economics, has been focused on understanding the risks climate change poses to our food system, particularly how areas where coffee beans are grown will shift under a changing climate. He’s using machine learning, a subset of AI, to figure out what factors affect growing locations, crop yields, and farmers’ livelihoods, and what climate-related factors affect those things.
“The fact is, there are hundreds of thousands of variables that matter or that might matter,” says Rising. “We don’t really know. The only way to find out if those variables matter or not is to ask the data. When rainfall goes from being 400 milliliters to 450 milliliters a year, then does coffee stop growing well? Humans would struggle with that for a long, long time to understand those differences. Machine learning could do it in hours for me.”
The kind of global-scale questions that face climate change scientists calls for them to throw sophisticated technologies at the problems, says Rising. “We’re bringing together larger and more intricate datasets than ever before, and having the power and techniques to analyze them, we’re making new results that change the way people think about human mortality, labor productivity, agricultural productivity, and energy demand. We certainly would not have made this kind of progress without big data and big computing. And AI has supported that work. It’s an important tool in our toolbox.”
For Columbia’s Uriarte and her tropical forest research, it’s all about consuming and making sense of the firehose of data that is coming at her. And she can’t do that without AI.
AI and climate change: Lessons for leaders
- Scientists can't take on climate change without AI, say experts. AI handles what humans can't.
- A firehose of data—estimated at trillions of data points—is coming from global sensors, weather stations, satellites, radar, and buoys. Humans and traditional computer programs can't sort through it all without computer help.
- One of AI’s biggest potential roles is in the decision-making aspects, trying to assess the implications of a changing climate and understand how choices affect it.
This article/content was written by the individual writer identified and does not necessarily reflect the view of Hewlett Packard Enterprise Company.