A comparison of yield prediction approaches using long-term multi-crop site-specific data

Authors: Huggins, David; Heineck, Garett; Casanova, Joaquin

Submitted to: Journal of the ASABE
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 5/13/2024
Publication Date: 5/25/2024
Citation: Huggins, D.R., Heineck, G.C., Casanova, J.J. 2024. A comparison of yield prediction approaches using long-term multi-crop site-specific data. Journal of the ASABE. 67(3):601-615. https://doi.org/10.13031/ja.15216.
DOI: https://doi.org/10.13031/ja.15216

Interpretive Summary: Predicting yield is important for farmers and researchers in making long-term management decisions. Current tools have several flaws. We use a long-term site-specific multiple-crop yield dataset in two modeling approaches and compare their performance. Different sources of data, including soil properties, topography, weather, and multispectral data are tested. Bayesian hierarchical modeling with spatio-temporal effects provides the best estimates, handles missing data, and provides uncertainty estimates in time and space.

Technical Abstract: Growers in the inland Pacific Northwest face numerous challenges in managing cropping systems. Climate variability, soil degradation, and topography lead to significant spatial and temporal variability in yield. Current yield modeling approaches can be difficult to use, lack transparency, and do not handle missing data well, nor do they provide uncertainty estimates, an important factor in making management decisions. To help this, we examine nearly two decades of rotations from the R.J. Cook Agronomy Farm, along with soil properties, topography, weather, and multispectral data, and test two modeling approaches to estimate yield: linear modeling (LM) and Bayesian hierarchical modeling (BHM). We find BHM with spatial and temporal random effects performs best in predicting relative yield, both using soil variables as predictors or remotely sensed data. Since the BHM approach handles missing data, allows farmer knowledge to be incorporated through priors, and gives uncertainty, this methodology lends itself well to decision support tools and on-farm study design.