Skip to main content
ARS Home » Research » Publications at this Location » Publication #189662


item Meek, David
item Singer, Jeremy

Submitted to: Applied Statistics In Agriculture Conference Proceedings
Publication Type: Proceedings
Publication Acceptance Date: 12/8/2004
Publication Date: 4/14/2005
Citation: Meek, D.W., Singer, J.W. 2005. An example of developing covariates for problems in precision agriculture. In: Milliken, G. (Ed). Proceedings of the 16th Conference on Applied Statistics In Agriculture, April 25-27, 2004. Manhattan, KS. p. 270-278.

Interpretive Summary: Statistical methodology for analyzing crop yield results from research in precision agriculture employs mainly variations of on well-known procedures. Such tools can provide sound inference about differences between local yields but no understanding of what is driving site differences is achieved. In this work, an alternative is provided via an example that shows crop yield relates very well to total light intercepted by the plant canopy over the growing season. The result is of interest to agronomic researchers and related academic and government agencies as well as crop consultants and related agricultural industries.

Technical Abstract: Methodology for precision agriculture is, perhaps, too focused on methods that allow for spatial correlation in the ANOVA error term. While sound inference about differences between local yields can be computed, no understanding of what is driving these differences is achieved. A completely general form for a spatial model can include suitable covariates. Most research in precision agriculture includes gathering a variety of site-specific information. Through the presentation of the analysis of data from a published soybean [Glycine max (L) Merr.] study, one specific type of covariate is developed - a duration index for soybean canopy light interception over the growing season. The relationship of the index to grain yield is reasonably well determined (R² = 0.82). We, therefore, suggest that the quest for modeling an appropriate covariate or covariates is primary. Treating spatial variation by other methods should only be used when the quest has failed.