Geosciences Professor Gives Presentation at U.S. Department of Agriculture

Photo Submitted

Xuan Shi

Xuan Shi, an assistant professor of geoinformatics in the Department of Geosciences in the J. William Fulbright College of Arts and Sciences, was recently invited to give a presentation at the Research and Development Division of the National Agricultural Statistics Service at the U.S. Department of Agriculture.

On Nov. 22, Shi gave a presentation titled "Geocomputation Over Heterogeneous Computing Infrastructure in the Era of Big Data Science" as a part of the Spatial Analysis Research Section seminar. Ten RDD researchers attended this seminar in person, while another five attended through webinar.

Shi's presentation included an overview of the latest developments in the field of geocomputation, as well as its possible applications for big data analysis and large scale image data classification. Big data analysis is the use of extensive data sets to expose unknown relationships, correlations, customer preferences, market trends and other information that can be useful to businesses.

A mission of the Research and Development Division is to produce national crop land cover classification annually. The Cropland Data Layer production is a typical geo-big-data computational process and is one of the fundamental national land cover data products that has been used nationally and internationally.

"The RDD is interested in my research on high performance geocomputation over big spatial data using hybrid computer architecture and systems," Shi said. "In general, geocomputation can be classified into different categories. When big data cannot be processed on a single computer, data and computation have to be divided and distributed onto multiple processors."

Shi said that many geocomputation processes may not have dependency and data communication in such a distributed computing environment, but varieties of geocomputation may have a strong dependency and data communication. Different solutions have to be developed to efficiently use varied supercomputing resources to complete the geocomputation tasks.

"RDD is seeking a more accurate, efficient and low cost solution with less human resource demand," Shi said. "My research on HPC solutions could potentially help RDD to further improve its operation."

Among the different applications introduced in this seminar, Shi said the Research and Development Division is particularly interested in the high-performance computing solutions that implement both unsupervised image classification by the Iterative Self-Organizing Data Analysis Technique Algorithm and supervised image classification by the maximum-likelihood-classification approach over big imagery data.

"By deploying large volume of CPUs and accelerators, we can process dozens to hundreds of gigabytes imagery data efficiently," he said.

For more information about the Research and Development Division or this seminar, contact Xuan Shi at xuanshi@uark.edu.