Skip to main content
ARS Home » Plains Area » Temple, Texas » Grassland Soil and Water Research Laboratory » Research » Publications at this Location » Publication #394069
Research Project: Development of Enhanced Tools and Management Strategies to Support Sustainable Agricultural Systems and Water Quality

Location: Grassland Soil and Water Research Laboratory

Title: Machine learning algorithms improve MODIS GPP estimates in United States croplands

Author
item Menefee, Dorothy
item Lee, Trey
item Flynn, Kyle
item CHEN, JIQUAN - Michigan State University
item ABRAHA, MICHAEL - Michigan State University
item Baker, John
item SUYKER, ANDY - University Of Nebraska

Submitted to: Frontiers in Remote Sensing
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 10/9/2023
Publication Date: 11/2/2023
Citation: Menefee, D.S., Lee, T.O., Flynn, K.C., Chen, J., Abraha, M., Baker, J.M., Suyker, A. 2023. Machine learning algorithms improve MODIS GPP estimates in United States croplands. Frontiers in Remote Sensing. 4. Article 1240895. https://doi.org/10.3389/frsen.2023.1240895.
DOI: https://doi.org/10.3389/frsen.2023.1240895

Interpretive Summary: Primary productivity in U.S. LTAR croplands was estimated using machine learning and remote sensing data. Eddy covariance data was used as a ground truth. This study showed that the machine learning method was able to successful estimate cropland productivity.

Technical Abstract: Machine learning methods combined with satellite imagery have the potential to improve estimates of carbon uptake across U.S. croplands. Studying carbon uptake patterns across the U.S. using research networks, like the Long Term Agricultural Research (LTAR) network, can allow for the study of broader trends in crop productivity and sustainability. In this study, satellite gross primary productivity (GPP) estimates were obtained from the Moderate Resolution Imaging Spectroradiometer (MODIS) satellite’s GPP product for three LTAR cropland sites planted to maize (Zea mays L.) and soybean (Glycine max L.). The three sites selected were those with the most available data among LTAR sites, Kellogg Biological Station (KBS, 2 towers and 20 site-years), Upper Mississippi River Basin (UMRB - Rosemount, 1 tower and 12 site-years), and Platte River High Plains Aquifer (PRHPA, 3 towers and 52 site-years). The MODIS GPP product was initially compared to in-situ measurements from Eddy Covariance (EC) instruments at each site. Next, machine learning algorithms were used to create refined GPP estimates using air temperature, precipitation, crop selection (maize or soybean), agroecosystem, and the MODIS GPP product as inputs. The optimal model for simulating GPP across all sites was a Stacked Ensemble type with r2 values of 0.97, 0.93, and 0.94; and MSE (gC m-2 8 days-1) of 73.18, 195.95, and 188.10 (daily RMSE gC m-2: 1.07, 1.75, and 1.71, respectively) for training, cross-validation , and validation, respectively. The machine learning methodology was able to successfully simulate GPP across three agroecosystems and two crops.