CALIFORNIA is facing a dire threat: oriental fruit flies. Officials sprang into action laying traps and spreading insecticide after 13 of the insects were spotted in the state. They issued an emergency statewide alert on 8 August, and the next step is a lockdown of huge stretches of farmland in a crippling quarantine.
This is no overreaction. The oriental fruit fly, Bactrocera dorsalis, is among the most feared insects in agriculture, responsible for crop losses that can run into billions of dollars. It is known to infest 230 crop species, with the larvae that hatch in the fruits leaving them rotting.
"They are one of the world's worst fruit and vegetable pests because of their rapid breeding, broad range of host plants and invasive abilities," says Gary Steck, an entomologist at Florida's Department of Agriculture.
Now a system developed in Taiwan, where the pest is endemic, aims to harness artificial intelligence to warn of imminent outbreaks, limiting the need for such drastic action.
In Taiwan, fruit fly populations are normally monitored using traps that are manually checked every 10 days. Cheng-Long Chuang and colleagues at the National Taiwan University in Taipei wanted to automate the counting process, so they placed infrared beams in the traps. Each trap records when the beam is broken, indicating that an oriental fruit fly has entered, attracted by a chemical designed to lure the insect. The results collected are sent via radio to a local station every 30 minutes, allowing real-time measurements of the population.
Part-funded by the Taiwanese government, the team have so far set up 240 traps on fruit farms around the country. Machine learning algorithms pool the continuous data arriving from each of these traps and predict when the local fruit fly population is about to explode.
To help in this prediction, the traps are also fitted with weather sensors that monitor temperature, humidity, wind speed and rainfall. Fruit fly population surges tend to match changes in weather - when it is humid, the level of insects is expected to rise, for example.
In Taiwan's current system, a red alert is issued when the number of flies caught in a trap surges beyond 1024 in a 10-day period. But the AI system can learn what counts as a normal level of fruit flies in an area and adapt its warnings on the basis of the current weather and time of year. It can also work out where the pest is likely to be breeding.
When a potentially devastating infestation is predicted, it automatically sends a text message to government officials' cellphones, providing the time, location and severity of the potential outbreak. The warning should allow authorities to pre-empt the outbreak by putting down insecticide.
Tested on historical data taken from the network of traps, the AI system was accurate in predicting an outbreak 88 per cent of the time (Computers and Electronics in Agriculture, doi.org/h6b).
"It is good to know the real-time status of my farm without physically going there," says Zai-Lang Jiang, who owns a guava orchard in Yuanlin, and is taking part in the experiment. "Also, I can reduce some production cost by avoiding unnecessary pesticide spreading."
Chuang believes the system can have immediate benefits for local farmers and help slash the annual fruit-crop loss. "If this system is widely deployed in most of the fruit orchards in Taiwan, we expect it to reduce the damage caused by the oriental fruit fly by 50 per cent," he says.
Farmers around the world will hope it works. The oriental fruit fly lives mainly in south-east Asia, but it is also seen in Hawaii, California and Florida, usually carried in on fruit that is illegally imported. As the effects of climate change play out, it is expected to expand its stomping ground into higher latitudes as temperatures rise (Bulletin of Entomological Research, doi.org/b6jjzd).