SCIENCE & ENGINEERING NEWS

Athens, GA. — They’re out there chomping away. Gypsy moths and spruce budworms can defoliate entire forests in short order without aerial spraying, so several times a year, specially equipped airplanes and helicopters set out to deposit loads of pesticide over thousands of acres.

The good news is that proper planning and precise aerial spraying can kill the pests before they grow out of the larval stage and begin to gnaw through the forest canopy. The bad news is that there’s much room for improvement in how the pesticides are targeted on forests.

That’s why a team that includes a computer scientist from the University of Georgia is, for the first time, using artificial intelligence tools called genetic algorithms to control how aerial spraying application models do their work. The new system has been tested successfully in the lab and could be in field trials by the end of this year.

“This research is about what happens to the spray right when it leaves the aircraft,” said Don Potter, a researcher with UGA’s Artificial Intelligence Center and associate professor in the department of computer science. “Our job is finding a better way to get the spray from the air to the ground.”

The project is funded by the U.S. Forest Service, and includes scientists from that USDA-run organization and also from a company in Princeton, N. J., called Continuum Dynamics.

Potter presented research on the spray modeling and application technology June 20 in New Orleans at the 13th Annual International Conference on Industrial and Engineering Applications of Artificial Intelligence and Expert Systems. The corresponding research paper won the Best Paper Award at the same conference. A description of the work will also be published this fall as a chapter in Practical Applications of Soft Computing Techniques (Kluwer Publishing Co.).

UGA Artifical Intelligence Center graduate student Wan Ying Bi has worked with Potter on the project. Scientists from the Forest Service include Dan Twardus, Harold Thistle, Mark Twery and John Ghent. Milt Teske collaborates from Continuum Dynamics.

The problems facing aerial sprayers fighting the gypsy moth and the spruce budworm are nothing new. Decades ago, sprayers saturated some areas of the Northeast with DDT, which effectively controlled the rapacious forest-eating moths. Unfortunately, there were considerable unintended effects from DDT, including the systematic softening of bird eggshells. Federal officials banned DDT, and the race was on to find another pesticide.

Scientists found a better spray with a naturally occurring insecticide called Bacillus thuringiensis (Bt),which is specific to caterpillar species. Bt is much more ecologically sound because it controls gypsy moths and spruce budworms without causing harm to other species.

Along the way, scientists and government officials realize that broadcast spraying wasn’t the best way to control forest pests, either. What they needed was a computer program that could control sprayers by taking into account a number of variables such as wind speed, airplane altitude and humidity.

The earliest models were developed by the U.S. Army but they were limited because they did not incorporate so-called “near-field wake dynamics,” which involve vortices created by the aircraft. The next generation spraying model was called the Forest Service Cramer-Barry-Grim (FSBCG) spray dispersion model, and it, too, was an improvement, but like the Army models on which it was based, the newer version had problems calculating near-wake flow fields.

Another model called AGDISP followed FSCBG, with which it was later combined to form the best system yet for controlling how aerial spraying systems do their work. Finally, during the 1990s, a much-improved model called AgDRIFT came into use, and sprayers had their best computer help so far in targeting pests from the air.

“Problems still remained, though, and the Forest Service started to look at things in another way,” said Potter. “The earlier simulations assume you know all the parameters, and you’d input them, and the model would give you answers about how best to spray. What they needed was a system where a pilot could put in the desired spray results and then let the system figure out how best to achieve these results. It would be like listening to an expert.”

Potter and his colleagues from the Forest Service and Continuum Dynamics thought the answer might lie in genetic algorithms. These computer programs are decision-making rather than just simulation models. A genetic algorithm uses the initial input from a human expert in a field and then uses that knowledge to make extremely fine distinctions in problem-solving and to recommend best solutions.

Though genetic algorithms had never been used in sprayer-modeling before, the results so far in the laboratory have been encouraging, Potter said. The team’s spray application is called SAGA – an acronym for Spray Advisor Genetic Algorithm. The researchers combined a genetic algorithm with a slightly modified version of the AGDISP model to achieve their success.

The variables that SAGA controls include aircraft altitude, nozzle type and number, airplane characteristics, environmental characteristics such as humidity, and aircraft flight pattern. The knowledge of the expert system is embedded in such a way that the computer makes an extremely sensitive analysis of how the insecticide can best be sprayed.

The pilot would then theoretically take the results generated by the SAGA system – which, according to Potter might take, depending on complexity, 30 minutes to two hours – and input them into an on-board computer that would completely control all spraying from the aircraft. The result would be a precision spray that takes advantage of an advanced analysis of numerous variables involved.

The next step will be to incorporate Global Positioning Satellite (GPS) information in the program. This would add important information on terrain to the program, Potter said.

When SAGA is finished, it will be available to private sprayers as well – though it would likely have to be modified for crop spraying. “There’s no reason it shouldn’t work on crops, though,” said Potter.

============================================================