Astronomers Feel the Animal Heat

Astronomers help ecologists locate at-risk animals using imaging techniques honed from decades of hunting for stars
01 March 2020
By Mara Johnson-Groh
Thermal sensors make warm-blooded rhinos stand out from their surroundings
Photo Credit: Liverpool John Moores University

Hiding within the virgin rainforest of Malaysia, there lives a troop of Bornean orangutans. These fruit-eating great apes, who spend the majority of their lives in trees, are widely under threat from deforestation. Today, only around 100,000 remain in the wild. Last spring, they were tracked by a group of unlikely hunters: British astronomers equipped with drone-mounted infrared and visible wavelength cameras.

The astronomers, accompanied by a few ecologists, were taking a field trip far from their telescopes to help find new ways to save some of the world's most at-risk animals. Combining off-the-shelf equipment with data-processing techniques and machine-learning algorithms gleaned from astrophysics, the researchers are combining astronomy with ecology to pioneer a new synergistic field: astro-ecology.

Serendipity by Train

Conservation ecology research relies on finding and monitoring animals to understand the factors vital to their conservation—no easy task when the animals in question have had thousands of years to perfect their camouflage. Consequentially, conservational ecologists have readily adopted new technology over the years, such as camera traps and DNA sequencing.

In the last decade, conservation ecologists have also begun using drones equipped with optical cameras to survey parks and wilderness areas. Drones can quickly survey large areas, but deciphering all the data is time consuming. Small animals in particular can be hard to spot from a distant aerial viewpoint.

Infrared cameras, which pick up the animals' warm-blooded heat signature, can be much more reliable for identification. Until recently, infrared cameras were prohibitively expensive. But advances in technology reduced camera size and weight, and also decreased the price, opening up their usage in conservation.

By combining infrared cameras with drone technology, conservation ecologists have gained a huge advantage in surveying large areas.

Back in 2014, one such conservation ecologist, Serge Wich, a researcher based at Liverpool John Moores University (LJMU), was working with infrared cameras but had trouble analyzing the large volume of data. One evening, Wich mentioned his struggle in passing to his seatmate as they rode the train home after work. That person happened to be his neighbor, astrophysicist Steven Longmore, who is also a professor at LJMU.

For decades astronomers have been using infrared cameras to study the births of stars and planetary systems. They've perfected techniques to remove background noise and image artifacts, as well as developed systems to automatically identify sources.

Longmore himself has been working with infrared data for 15 years. He got started with a thermal camera mounted on the Gemini North 8-meter telescope in Hawaii where he observed star-forming regions to understand how gas clouds collapsed to form high-mass stars. He realized that looking for endangered animals wasn't that different from identifying young stars in gas clouds and offered Wich his help.

"[The animals'] glow is exactly the same as the kind of glow stars and galaxies have in astronomical images," said Claire Burke, an astro-ecologist at LJMU who was one of the first researchers to start on the project. "The idea was we could we use techniques from astronomy to find them."

Longmore and Wich began a collaboration that has since expanded into a multidisciplinary team of astronomers, ecologists, computer scientists, and engineers. Their single serendipitous conversation has since launched a flood of new research.

Milky Way galaxy. Credit: NASA/SOFIA/JPL-Caltech/ESA/Herschel

Composite infrared image of the center of the Milky Way galaxy. Photo Credit: NASA/SOFIA/JPL-Caltech/ESA/Herschel

From the Heavens to the Hilltops

In infrared astronomy, telescopes are typically equipped with high-resolution CCD cameras cooled to reduce internal noise. For conservation purposes, the researchers instead must use microbolometers, which are cheaper and light enough to fly on a drone. The microbolometer commonly used by the LJMU group—a FLIR Tau 2 640 long-wave thermal infrared camera-weighs only 72 grams, and lighter ones, like the FLIR Boson come in at just over 7 grams. The downside is that these lightweight cameras are low resolution, only 640 × 512 pixels.

"That's sort of industry standard for a top-of-the-range [infrared] camera, and that's not a lot of pixels," Burke said. "That's like a smartphone 10 years ago."

As a result, the cameras have to be flown close enough to their targets for identification. While this isn't prohibitive when searching for elephants on the African plains, it's challenging when searching for smaller animals in a forest environment.

Furthering the issue, pixels in microbolometers can bleed into one another. As a result, one data point might contain a heat signature from both an orangutan and the tree it's occupying. If the orangutan's shape doesn't cover enough pixels, its true temperature can't be distinguished—something known as the spot size effect. The astronomers calculated they'd have to fly no higher than 90 meters above an orangutan to resolve it—a height just above the trees' crowns, some of which reached 65 meters.

The lower flying heights required to combat the spot size effect limits the coverage area for smaller animals, but the data can still be valuable. In 2017, the group flew at a height of just 20 meters as they searched for riverine rabbits—one of the most endangered mammals in the world—in South Africa. They ended up with five sightings, which is remarkable considering their entire population is estimated at just 1,500 individuals.

In order to see fainter and smaller astrological objects, astronomers maximize their signal-to-noise ratio by extending observations over hours, sometimes even stacking exposures with the cumulative time of days. Unlike celestial bodies, animals don't tend to stay in one place, so improving the likelihood of finding an animal requires other techniques.

Instead of taking longer images, astro-ecologists increase their signal by choosing the time when they observe. At night, the ground environment is coolest, increasing the temperature difference between object and background. Some countries and many conservation areas have laws restricting drone usage at night, so the group often chooses to observe in the morning just after dawn, when the ground is still relatively cool.

Longmore wondered about the efficacy of detecting animal heat signatures in the warm humid environment of a jungle. "We were a bit worried," he said. "But we are happy to see that even in these very hot, humid environments the orangutans are still detected by the thermal cameras."

Of course infrared data doesn't guarantee detection. Thick canopy cover can hide an animal's heat signature. While the group was able to see orangutans through some level of leaves, they're not yet sure how comprehensive their coverage is at different foliage densities.

The drones are also flown with optical cameras that can be used to confirm possible sources seen in the infrared data. With over twice the pixels as the infrared cameras, the higher resolution allows the researchers to zoom in and distinguish false sources from live animals.

thermal image of leopard in a tree. Photo Credit: FLIR

Photo Credit: FLIR. Taken with FLIR T1020.

Old Camera, New Tricks

Astronomers are used to working with infrared cameras specifically tailored to their needs. But commercially available infrared cameras are often designed for industry work, such as in glass and plastics manufacturing—not for the intent of tracking endangered animals. The microbolometer used by the LJMU group was specifically designed to measure the spectral band from 7.5 to 13.5 micrometers, with a scene temperature range spanning nearly 600 degrees Celsius. As a result, certain calibrations need to be made for use in astro-ecology.

"The cameras are optimized to work over a very large [temperature] range," Longmore said. "We're trying to find the ways to optimize the camera to work for the purposes that we're doing in conservation."

To identify different species by their thermal image, the data needs to resolve differences as small as 0.5 degree Celsius. Additionally, since wide-angle lenses are used for infrared imaging, the researchers also have to account for the drop in sensitivity of the microbolometer around the edges due to imperfect optics. The FLIR cameras have autocorrections and calibrations preprogrammed, but additional adjustments were needed to enhance the data for animal tracking.

The optimizations they developed were largely drawn from three concepts taken from astronomical imaging: flat fielding, stacking, and binning. Flat fielding allows the researchers to characterize the sensitivity of each element of the array. By additionally stacking multiple images of a uniform temperature source, the researchers can average the images and account for stable noise. Binning, or averaging neighbor pixels, can also help reduce spatial noise in the images. By adding these techniques to their data processing, the group found they could correct most of the large structure noise, improving image quality. However, issues of nonuniform noise are still challenging to correct.

So far one camera has been optimized, and the group is working to see if they can develop a generic calibration applicable to other instruments. The astro-ecology group has also adjusted how frequently calibration images should be taken to optimize their data for their aerial applications.

Ambient temperature, for example, can affect camera sensitivity. "If you're flying a drone, then you've got wind that's keep keeping your camera cool," Longmore said. "You might need to do fewer [calibrations] than if you had a different environment."

The cameras in the orangutan study were flown aboard a Tarot X4 drone, a type of drone with four rotors designed for aerial photography. These types of drones are useful in testing as they can hover and allow for controlled movement. The group has also used fixed-wing drones, which have much better battery life. Fixed-wing drones, which look like small airplanes, are advantageous when covering large areas, but also require constant forward motion and an open area for takeoff and landing.

FLIR T1020 thermal image of an elephant

Photo Credit: FLIR. Taken with FLIR T1020

Orangutans and Beyond

The LJMU group has taken their instruments around the world. The first proof-of-concept tests were done over an English cow pasture. Since then, they have flown in Mexico looking for spider monkeys, South Africa to find riverine rabbits, Tanzania to spot poachers, Madagascar in search of bamboo lemurs, and Brazil to find river dolphins, in addition to the work with orangutans in Malaysia.

"We've discovered that different species of animal have a unique thermal profile as well as being different shapes and sizes," Burke said. "And we're training a machine-learning algorithm to recognize different species of animal automatically."

Automatic identification and large data sets are both things astronomers are familiar with. The Pan-STARRS survey, which continually looks for moving and variable objects, has created over 1.6 quadrillion bytes of data, and the upcoming Square Kilometer Array is expected to produce even more—a quintillion bytes of data every day. With this data deluge, automated systems are necessary to reduce data and flag potential objects of interest.

Animals are found in environments with nonuniform background temperatures, making them hard to identify through thresholding—a common technique in astronomy that flags objects above a certain contrast level. Instead, the LJMU group has adapted machine-learning algorithms from more complex astronomical identification programs to better identify animals and distinguish them from other false sources, like a hot rock warming on a sunny day.

The programing for the calibrations and machine learning is done in Python and entirely open source. The hope is that game wardens and other conservationists will be able to use such setups to easily find and monitor animals. The drones and cameras are also all modified from off-the-shelf units, making them affordable and accessible.

The group is also exploring beyond conservation to issues related to climate change and safety. They're looking at applying the same technology to search and rescue planning, monitoring wildfires, and identifying underground peat fires.

"None of the stuff we're doing in and of itself is hugely groundbreaking," Longmore said. "It's the piecing together of lots of different bits from different areas."

Groundbreaking or not, the synergy of astronomy and ecology is steadily expanding the scope of environmental monitoring. And with an increasing number of possible applications, both in ecology and astronomy, the technology seems bound only by the imagination.

Mara Johnson-Groh is a freelance science writer and photographer who writes about everything under the Sun, and even things beyond it.

Enjoy this article?
Get similar news in your inbox
Get more stories from SPIE
Recent News