Invasive plant species crowd out native species, outcompete crops, and cause havoc to rangelands. In the United States alone, the economic losses are estimated at over $20 billion each year. The most effective way to cut down on pesky plant invasions is to prevent them from taking root in the first place. The second best option is to detect arrivals early on and mitigate their spread.
For countless years, the best way to monitor invasive plants was by completing field surveys. These tactics are expensive, labor intensive, and slow. Delays in getting data gathered, compiled, and shared leads to missed opportunities of detecting and mitigating early.
David Moeller, a professor in the Department of Plant and Microbial Biology, sought to find a better way. Moeller, alongside graduate student Thomas Lake and research scientist Ryan Briscoe Runquist, found promise in a new method that leverages images from satellites.
Leafy spurge—the invasive plant in question—wreaks havoc on rangelands and pastures. It crowds out plants that are more nutritious for livestock and it’s toxic to them, causing sickness and death in cattle, sheep, and horses.
The research team took aerial images from two kinds of satellites to sort out how sharp the images needed to be to detect leafy spurge. Some images were high-resolution images taken infrequently and others were low-resolution images taken daily. From these images, and field data, researchers developed deep learning models by training a computer to identify leafy spurge from the images.
The deep learning models detected leafy spurge in the Twin Cities region with greater than 96 percent accuracy. The model leveraged a time series analysis and used numerous images across time to help detect the plant. The plant’s distinctive emergence, flowering, and senescence helped the model accurately identify the plants.
“Time series analysis is relatively new in this type of deep learning model so we were really excited to find that it worked so well at identifying leafy spurge populations,” says Runquist. “Deep learning allows scientists to discover patterns that previously were impossible to detect.”
The findings suggest that deep learning models can accurately identify individual species over complex landscapes with satellite imagery, even using lowerresolution images. This gives researchers hope of leveraging techniques to track invasive species—with limited lag time— using publicly available satellite images.
“Invasive species are more difficult than ever to manage and surveillance via satellites is a low-cost, rapid method for dynamically monitoring invasions,” says Moeller. —Claire Wilson