"The method we developed is data intensive because you have to monitor resistance across the landscape in different fields," he said. "And then you have to relate the spatial variation in resistance to the abundance and distribution of potential refuges and treated fields. In some fields you have high resistance; some fields you have less. And the hypothesis is that you have less resistance in some fields because you have more refuges near these fields."

The group developed spatially-explicit, statistical models based on the first four years of data that included aerial remote sensing maps and documented pesticide applications in the study area. The models identified the crops affecting the spatial variation in resistance and the maximum distance at which these crops affected resistance. They then used a separate data set from the next four years to predict resistance at the landscape level with these models. The successful prediction of resistance confirmed that refuges of cotton delayed the evolution of resistance and fields treated with pyriproxyfen accelerated the evolution of resistance.

"We had the resistance data from 84 cotton fields. What we didn't have was information on the crops surrounding these fields. We had to use remote sensing, digging back through satellite images for the last eight years, and then analyzed the images to map the crops surrounding each cotton field."

Carrière said that the method and framework developed by their research could help refine the refuge strategy for many key pests. Carrière points out that "This is important because some pests targeted by a Bt crop or insecticides are generalist feeders. They can come from many types of crops or uncultivated habitats. A pest targeted by Bt corn, for example, could come from soybeans or tomatoes or sorghum. So, then how do you know which of these habitats are efficient refuges, and how do you know the distance at which these refuges will produce enough susceptible insects to delay the evolution of resistance?"