Skip to main content
ARS Home » Southeast Area » Tifton, Georgia » Southeast Watershed Research » Research » Publications at this Location » Publication #390659
Research Project: Enhancing Water Resources, Production Efficiency and Ecosystem Services in Gulf Atlantic Coastal Plain Agricultural Watersheds

Location: Southeast Watershed Research

Title: Machine learning and AI for characterizing agricultural production regions

Author
item POWELL, JOSEPH - University Of Georgia
item STONE, AUSTIN - University Of Georgia
item Coffin, Alisa
item SEYMOUR, LYNNE - University Of Georgia
item AHN, JEONGYOUN - University Of Georgia
item MADDEN, MARGUERITE - University Of Georgia

Submitted to: US-International Association for Landscape Ecology
Publication Type: Abstract Only
Publication Acceptance Date: 2/21/2022
Publication Date: 4/13/2022
Citation: Powell, J., Stone, A., Coffin, A.W., Seymour, L., Ahn, J., Madden, M. 2022. Machine learning and AI for characterizing agricultural production regions. 2022 Annual Meeting of the US-International Association for Landscape Ecology--North American Chapter; Virtual. 2022.

Interpretive Summary:

Technical Abstract: Understanding dynamic agroecological systems is central to long-term food security. To study this, the United States Department of Agriculture, Agricultural Research Service (USDA-ARS) established the Long-Term Agroecosystem Research (LTAR) Network. This study focuses on the Gulf Atlantic Coastal Plain (GACP) LTAR site in Tifton, GA, where scientists have been observing cropping, hydrologic and climatic systems in the Little River Experimental Watershed (LREW) since 1968. To assess the broader impacts of environmental stressors and agricultural management practices in the LREW it is necessary to extrapolate observations beyond the bounds of experimental and monitoring studies. Our goal is to identify appropriate scaling methods for the extrapolation of data, using statistical methods to identify, validate and guide the improvement of scaling models towards characterizing the GACP agroecoregion. We are assessing Machine Learning and Deep Learning methods such as Random Forest, Support Vector Machine (SVM), K-Nearest Neighbor (KNN) and convolutional neural networks (CNN) to advance our characterization of the agroecoregion. A combination of desktop and cloud-based platforms were used to compute vegetation indices from remotely sensed imagery acquired by satellites, aircraft and uncrewed aerial systems (UAS). The UAS imagery was collected biweekly at 9-cm spatial resolution with DJI Matrice 100 Pro equipped with a MicaSense RedEdge multispectral sensor throughout the growing season in 2018 and 2019. Orthorectification provided a rich data set of high spatial and temporal resolution imagery for rotational crops of cotton, peanuts and corn. Satellite imagery included 10-m Sentinel 2A/B MultiSpectral Instrument (MSI) and 30-m Landsat Operational Land Imager (OLI) data acquired at 5-day and 16-day intervals, respectively. Utilizing ArcGIS Pro and TensorFlow, we scaled plot-level measurements of agricultural crop type and above ground biomass to imagery of increasingly coarse spatial resolution. This project is contributing to broader regionalization efforts within the continental network of LTAR sites to identify agroecoregions.