A group led by Dr. Darren Drewry in the Department of Food, Agricultural and Biological Engineering (FABE) at Ohio State University recently published an article in the peer-reviewed journal Agricultural Water Management on the use of machine learning (ML) to estimate total ecosystem water use (also known as evapotranspiration or latent energy flux) in agricultural systems.
The data for this research was collected at the Sustainable Advanced Bioeconomy Research (SABR) experimental site outside of Ames, Iowa, in collaboration with colleagues at Iowa State University. SABR comprises a set of four fields (soybean, corn, bioenergy sorghum and miscanthus) each monitored with unique climate and eddy covariance flux sensors. Dr. Drewry’s team in FABE has augmented this site with proximal sensors for land surface temperature (LST) and normalized difference vegetation index (NDVI), observations that carry information on vegetation physiological response to climate and canopy phenology.
This research focused on the development and cross-validation of 64 sets of ML models for each crop to produce a comprehensive evaluation of the combinations of climate and proximal sensing variables that produced the best predictive performance for latent energy flux. This work demonstrates the power of ML to predict field-scale water use with just a few carefully selected predictor variables. The results also demonstrated the strength of proximal remote sensing in particular to enhance water use predictions, advancing data-driven methods for this critical ecosystem function.