- A six-year study of leopards in the wildlife-rich southern Indian state of Karnataka, using grids of motion-sensor camera traps across the state, suggests the big cats are thriving in a variety of habitats and land uses.
- The researchers’ use of machine-learning algorithms significantly reduced the workload needed to identify 363 individual leopards from the sample’s 1.5 million camera-trap images. The figure indicates there are an estimated 2,500 leopards living in Karnataka.
- Although a forest department official said the state was unlikely to expand its protected forests in the foreseeable future, the researchers said such a policy was necessary for leopard conservation, stressing that the proximity of natural landscapes to agricultural fields allows leopards to use those unprotected areas.
An extensive study of the leopard population in the wildlife-rich southern Indian state of Karnataka has indicated that these big cats are thriving there, buoying hopes the species’ genetic pool is stable in the region.
Researchers from the Karnataka Forest Department (KFD) and the independent research organization Nature Conservation Foundation (NCF) jointly surveyed the cats in protected forests, private lands, rocky outcrops and non-protected natural areas spanning a diverse landscape covering roughly 27,000 square kilometers (10,400 square miles). They used grids of motion-sensor camera traps to better understand leopard distribution and habits in a variety of landscapes.
Karnataka is the first Indian state to scientifically estimate the population of leopards (Panthera pardus) over such a large area, following similar studies for tigers and elephants. The project took six years to complete and identified 363 individual leopards based on the rosette patterns on their bodies, which, like human fingerprints, are unique to each individual. The cameras captured more than 1.5 million images at sites across the state. Using statistical extrapolation from this sample, the researchers estimate that Karnataka is home to 2,500 leopards.
Algorithms speed image analysis
To aid such a large-scale survey, the researchers applied artificial intelligence to examine the collection of camera-trap images. This significantly reduced the time and human resources needed to complete the analysis.
Using a convolutional neural network (CNN), a framework enabling machine-learning algorithms to work together to analyze visual imagery, researchers reduced the workload from weeks to hours in some cases.
Led by AI expert Ramprasad K.R., the technology team comprising staff from Ramprasad’s Camera Commerce startup and NCF members helped the wildlife researchers identify leopards in their images. “This kind of AI is called ‘computer vision’ which enables machines to see the visual world and interpret it the way a human would,” Ramprasad, the former chief technologist for AI at Indian tech giant Wipro, told Mongabay.
The team fed the CNN with thousands of images of leopards. These “training” images allowed the network to recognize the features, colors, shapes, sizes and unique patterns associated with leopards. Ramprasad said it usually takes between 1,000 and 2,000 images to train a CNN for a particular species.
In the initial stage of analysis, the CNN removed “noise”: those images without animals or with humans or livestock. During the second phase, the network identified from the remaining hundreds of thousands of images containing some type of animal those images with the species targeted by the researchers.
The trained algorithm can process up to 60,000 images in nearly half the time required by three researchers working full-time for three weeks. Additionally, its accuracy rate increases as the network processes more pictures of a particular species by learning more characteristics of the animal.
“When given a new picture, it can quickly infer within half a second to label it with the right species’ name,” Ramprasad said.
The deep-learning network is able to identify animals from different angles, captured in daylight or at nighttime, as well as from pictures containing only some body parts. “Many camera trap pictures may just have parts of an elephant or tiger, etc.,” Ramprasad said. “The network identifies even these partial pictures to a high degree of accuracy.”
Ramprasad said the CNN is easily customizable for other species, and the accuracy rate for wild cats is roughly 90 percent. “Animals and forests in each region are different,” he added, “and hence it is necessary to keep the network’s model open, offering flexibility, scalability and replicability to wildlife researchers across the world.”
Field challenges remain
The sheer vastness of the area the field team needed to cover to collect images of wild leopards across an entire state required the researchers to install the camera traps in stages. “We do it one site after another. And within a site, if the area is large, we deploy camera traps one block after another,” field team leader Sanjay Gubbi, a conservation biologist with the NCF, told Mongabay.
Gubbi and his colleagues lost some valuable data to thieves who stole cameras, and had to contend with wild elephants damaging other cameras. They adapted by rigging the cameras with more protective gear and camouflage to save their equipment.
A third challenge was to keep the batteries inside the cameras functioning, as they tended to die out frequently. “The cost of camera traps and importing them is a tedious process,” Gubbi said. “They consume a lot of batteries, but we try and recycle used batteries.”
A role for unprotected natural landscapes
India’s first ever nationwide leopard census was carried out in 2014, which put the species’ population at an estimated 12,000 to 14,000 individuals. Leopards are one of five big cat species found in the Indian subcontinent, alongside Asiatic lions, Bengal tigers, snow leopards and clouded leopards.
Gubbi said the study’s findings were crucial to the wild cats’ conservation across India. The research team monitored leopard abundance and density inside and outside protected areas, and the preliminary data suggest leopards maintain healthy populations even outside protected areas.
“It is a popular notion that leopards do extremely well in sub-optimal, dynamic agricultural fields such as maize and sugarcane fields,” Gubbi said. “But the fact that these agricultural fields, where there are leopards in higher densities, are abutted by natural landscapes is not accounted for.”
“What role these natural landscapes play in leopard density dynamics is very critical. Unfortunately, using this popular theory, economic interests are pushing for leopard habitats to be used for extractive activities that leads to total loss of leopard habitats,” he added.
More protected areas
Based on the results, researchers recommend that the government designate more protected areas across Karnataka for leopard conservation, as the species’ population is in decline in much of its global range.
“It is critical to set aside some areas in the name of leopards. Apart from supporting viable leopard populations, such areas also support other unique wildlife species, so leopard becomes an umbrella species for other lesser-known species such as the chinkara [Indian gazelle], four-horned antelope, striped hyena, yellow-throated bulbul, etc.,” Gubbi said.
But the Karnataka government said it wasn’t possible to expand protected wildlife areas, as the state already has a vast region marked for that purpose. India’s iconic Western Ghats, a UNESCO World Heritage Site and global biodiversity hotspot, accounts for about 60 percent of Karnataka’s forest area.
C. Jayaraman, the KFD’s principal chief conservator of forests, told Mongabay that the state already had the largest protected forest area in India and was unlikely to expand it substantially in the foreseeable future.
“One should know leopards are used to stay very close to human habitats, it could be close to edges of cities like Bangalore [Bengaluru] or townships,” Jayaraman said. Still, human-animal conflicts involving leopards are significantly rarer than perceived. “For example,” Jayaraman said, “the number of such incidents involving elephants or sloth bears is comparatively more than the ones involving leopards.”
He added that leopards, which often feed on stray dogs in this region, do not usually attack livestock and are reluctant to come into close contact with humans. Therefore, he said, there is no immediate necessity to establish more protected areas.
Jayaraman welcomed the latest study’s findings that the leopard numbers are stable despite shrinking wildlife space and increasing human population. “The study is a good indicator whether we have a population that [has a] genetically viable future,” he said.
In addition to its findings on leopards, the camera-trap survey had some incidental benefits, such as the spotting of poachers in some areas, resulting in law-enforcement action. The camera traps also captured several species previously thought to be absent or extremely rare in the region, including the honey badger, or ratel, and the chinkara, which benefit from the same vegetative cover used by leopards.