|
 However, these estimates are based primarily on
farmer self-reporting and are only compiled once every five
years, potentially limiting accuracy.
In a new study, University of Illinois scientists demonstrate a
way to accurately map tilled land in real time by integrating
ground, airborne, and satellite imagery.
“We’ve shown remote sensing can quantify regional-scale tillage
information in a cost-effective manner. This field-level
information can be used to support growers in their management
practices, as well as to support agroecosystem modeling and
provide tools to the USDA to verify their census data,” says the
study’s lead author, Sheng Wang, a research assistant professor
in U of I’s Department of Natural Resources and Environmental
Sciences (NRES) in the College of Agricultural, Consumer and
Environmental Sciences (ACES). He is also a research scientist
in the Agroecosystem Sustainability Center (ASC) at U of I.
Wang and the research team took photos of the ground at
participating field sites throughout Central Illinois,
generating 6,719 GPS-tagged images. Then they arranged for an
airplane equipped with high-powered hyperspectral sensors to fly
over the region. The airborne system scanned 40,000 acres per
hour and captured rich spectral signatures of the ground at a
scale of about half a meter.
Wang fed the ground photos into a computer that learned to
differentiate bare ground from crop residue, a hallmark feature
of no-till and conservation tillage. After training on labeled
ground images, the computer could interpret and predict
hyperspectral images from the airborne sensor with about 82%
accuracy. Using this ground-to-air upscaling as a model, the
computers then developed an algorithm to scale up again, this
time from the air to space, using satellite data.
Compared to upscaling directly from the ground to
the satellite, which was only accurate about 22% of the time
according to a separate analysis in the study, the airborne
layer increased mapping accuracy to 67%.
[to top of second column] |
“In remote sensing, we're always trying to link
ground-truth data with spectral signals from satellites, but that
represents a big scale mismatch. The intermediate-scale
hyperspectral data helps to augment ground-truth data because it can
provide both high resolution and accuracy. It’s a major innovation;
nobody has done this in the agricultural world. This cross-scale
technology significantly advances our capability to create
ground-truth information,” says Kaiyu Guan, associate professor in
NRES, founding director of the ASC, and senior author on the study.
Although the method was tested in Champaign and surrounding Illinois
counties, Guan says the team is working to scale the technology to
the broader Midwest and the nation. Now that airborne sensors and
computers have been trained to detect evidence of tillage using
ground images, it should be possible to forego or minimize ground
photos in the next iteration.
The article, “Cross-scale sensing of field-level crop residue cover:
Integrating field photos, airborne hyperspectral imaging, and
satellite data,” is published in Remote Sensing of Environment [DOI:
10.1016/j.rse.2022.113366]. The research was supported by the U.S.
Department of Energy ARPA-E SMARTFARM projects and the Foundation
for Food & Agriculture Research (FFAR) Seeding Solutions Award.
Partial funding was also provided by a Foundation for Food and
Agriculture Research Seeding Solutions Award, the National Science
Foundation, the USDA-NIFA AIFARMS project, and the C3.ai Digital
Transformation Institute.
[Sources: Sheng Wang andKaiyu Guan
News writer: Lauren Quinn ]
 |
