A new approach to identifying Northern Great Plains vegetation using satellite data has earned a South Dakota State University doctoral student second place in the student poster competition at the 20th William T. Pecora Memorial Remote Sensing Symposium. The national conference, held Nov. 13-16 in Sioux Falls, brings together scientists who evaluate land and water resources using Earth-observation satellite data.
“I was very happy—and surprised because I had not registered for any kind of competition,” said Lan Nguyen, a Geospatial Science Center of Excellence doctoral candidate. He recalled explaining his research poster to a conference attendee carrying papers who then gave him advice on his work. He didn’t think much of it until the next morning when he checked his email and found he had won the $150 second place award.
GSCE co-director Geoffrey Henebry, Nguyen’s dissertation adviser and chair of his advisory committee, said, “We are very proud to have two SDSU students finish among the top three in this poster competition.” GSCE doctoral student Yenni Vetrita received third place for her poster on using satellite data to examine peatland fires. Her advisory committee chair is associate professor Xiaoyang Zhang and herdissertation adviser is former GSCE senior scientist Mark Cochrane, now at the University of Maryland Center for Environmental Science.
Nguyen completed two master’s degrees at Marshall University in Huntington, West Virginia, and worked as part of a geospatial sciences team for two years before coming to SDSU in 2015. He did his undergraduate work in Vietnam.
Importance of tracking vegetation
Nguyen’s model analyzes the seasonality of land surface objects in the Dakotas, Minnesota, Iowa and Nebraska. “Even though this area does not have a lot of people, there are a lot of changes in land use going on,” Nguyen said. Accurately tracking how land is being used and how that changes over time provides important data for those managing land and other natural resources.
To generate baseline data, Nguyen used the U.S. Geological Survey’s National Land Cover Dataset, which has fewer than 20 land cover classes, and the U.S. Department of Agriculture’s Cropland Data Layer, which has 100 classes of specific crops and other land cover types. “Different datasets are created with different purposes in mind—it depends on what you’re looking at,” he said
“Traditionally, a couple of satellite images, typically taken during the summer growing season, are used to identify land cover,” Nguyen explained. His approach utilizes images captured throughout the year from moderate resolution imaging spectroradiometers aboard the Terra and Aqua satellites, known as MODIS, and from finer spatial resolution imaging sensors on a pair of Landsat satellites. MODIS satellites can view the Earth’s surface every day or two, while the pair of Landsat satellites does so every eight days.
Identifying vegetation over time
Those datasets allow Nguyen’s model to more accurately identify the type of vegetation, such as crops, based on when they are planted and harvested. For instance, it’s difficult to differentiate corn from soybeans using just a few images from the summer growing season, he explained. However, he can distinguish between two crops “using multitemporal data because corn is planted earlier and then harvested later than soybeans.”
The vegetation index is based on sunlight reflected from the surface, Nguyen explained. The index uses red and near infrared light reflectance. “Vegetation reflects little red light, but a lot of near infrared light,” he said. “This difference is larger for vegetated areas, and smaller for areas of bare soil, building materials, or open water,” he explained.
His model links the vegetation index through time as viewed with the Landsat sensors with the temperature of the surface measured by the MODIS sensors. “This approach looks at land cover from the temporal dimension, one that may see some things more accurately and perhaps fill in some gaps,” Nguyen said. “That’s the whole idea of doing science.”