Why do solar farms kill birds? Call the AI birdwatcher

To review this article, select My Profile and then View Recorded Stories.

Daniel Oberhaus

To review this article, select My Profile and then View Recorded Stories.

American solar farms have a challenge for birds. Utilities have discovered bird carcasses scattered throughout their facilities for years, a strange and unforeseen result of the national solar boom. No one knew why this was happening, but it was obviously a challenge for a kind of energy that was advertised as environmentally friendly. For example, in 2013, an organization of utility companies, academics, and environmental organizations combined to shape the Aviar Solar Working Group to expand methods to mitigate avian mortality in solar facilities in the United States.

“Very few studies have been conducted on the effects of solar energy on birds,” says Misti Sporer, senior environmental scientist at Duke Energy, an application in North Carolina and a member of the working group. “What does it mean when you locate a dead bird? No one knew. But simply gaining knowledge about bird deaths in solar installations has proved difficult.

In 2016, a single study estimated that a lot of large-scale solar farms in the United States can kill approximately 140,000 birds a year. That’s less than one-tenth of one percent of the estimated number of birds killed through fossil fuel force plants (through collisions, electrocutions and poisonings), however, researchers expected that number to triple almost as planned solar farms are commissioned. The link between solar installations and bird death is not yet clear. One of the main theories suggests that birds confuse the glare of solar panels with the surface of a lake and rush to land, with fatal results. “But this speculation is from a human point of view,” Sporer says. “Do birds see people? We want to gather more knowledge to shape a complete picture.”

Earlier this year, the Department of Energy awarded a team of researchers from argonne National Laboratory in Illinois a $1.3 million contract to expand a synthetic intelligence platform committed to examining the avian habit in large-scale solar facilities in the United States. Researchers hope that the knowledge gathered through its formula will help birdwatchers get to the bottom of the mystery of the mass death of our feathered friends on solar farms. “The vital thing is to decrease the environmental effect on solar energy in all its forms,” says Yuki Hamada, a biophysicist from Argonne who runs the project. “These aviating disorders are a fear and anything that the renewable energy industry needs to perceive and mitigate.”

Only a few regions of the United States have regulations that require solar operators to report bird deaths at their facilities; Big giant American sun farms really care about this long, morbid calculation. Those who do have a limited ability to collect quality knowledge and can only send researchers to count bird carcasses on a solar farm once a month. While this is helping solar power plant operators perceive how many birds die, it provides a lot of data on the reasons for their death. To do this, they want real-time observations.

Counting dead birds is just the kind of repetitive and unpleasant task for which AI was designed. But in reality, the implementation of the formula in a solar installation presents many technical challenges. Perhaps the most complicated task is simply to teach the device a set of learning rules to reliably recognize birds in a complex environment. Birds come in other sizes, shapes and colors, meaning that the set of rules must have a smart enough understanding of the “bird” summary concept to be able to identify them, whether they fly over their heads or are perched on a solar panel.

Adam Szymanski is a software engineer in Argonne and leads the progression of the laboratory’s AI-powered bird observer. He says that commercial vision software was born from his paintings in some other task designed to automatically stumble upon small drones in the air. Amateur drones don’t have wings to flap or strut with their legs, so it’s simple to teach a set of rules about what a drone looks like. But reusing the bird tripping set will require Argonne’s team to meticulously label birds in thousands of photographs so that they can be used as educational knowledge for the rule set.

“The device learning studies we do are a little unique, because we don’t just need to classify an object into a single image,” Szymanski says. “You will have to classify a small object that moves in time. So, if the bird flies, in some photographs you will see a point and in others you will see its wings come out, and we have to stick to that object as it moves through the Camera. “

System hardware also presents some challenges. Solar installations tend to be in the middle of nowhere and sometimes don’t have the kind of infrastructure needed for complex device learning applications. There are no knowledge centers nearby, Internet bandwidth is limited and it can even be difficult to get electricity. “You think solar installations are forced because they produce electricity. But they don’t have power sockets connected to the panels,” Szymanski explains. This means that the hardware that will run Argonne’s bird-watching rule set will have to be incredibly resource efficient as it will run on batteries or use small solar panels, while you will need to analyze a lot of knowledge in real time. .

To achieve this, the Argonne team uses advertising devices developed through a company called Boulder AI to monitor pedestrian and vehicle traffic. Boulder’s small camera formula is designed for computing, the general term for on-site knowledge processing rather than a remote knowledge center. But instead of attaching it to a light pole, Argonne’s team will place it on a solar panel.

Today, Hamada and his team are gathering educational knowledge from cameras installed at two solar facilities in Illinois. The plan is to gradually expand the program to a few dozen government advertising and sun sites in the United States, but the pandemic has slowed that deployment. At first, Argonne’s AI will only seek to identify birds entering its vision box, but Szymanski says it will eventually be complicated enough to differentiate a handful of bird behaviors such as perching or colliding with a solar panel.

This knowledge will be essential for researchers who will eventually be tasked with locating responses to prevent the death of birds in solar installations. This will help them perceive how the local environment, such as time or time of day, affects the behavior of birds, or can identify other imaginable reasons for feather death. “Being able to see birds interacting with the site without the presence of a human observer is incredibly beneficial,” Sporer says. “This generation allows us to glimpse a world we don’t usually see, so that we can function in the least impactful way to wildlife.”

Updated 8-10-20 9:30 AM ET: Misti Sporer is a member of the Avian Solar Working Group, the coordinator.

WIRED is where it is done. It is the essential source of data and concepts that give meaning to a global and coherent transformation. The WIRED verbal exchange illustrates how generation is turning each and every facet of our lives: from culture to business, from science to design. The advances and inventions we notice lead to new thinking tactics, new connections and new industries.

Leave a Comment Cancel Reply