Atmospheric pressure and wheat yield modeling

• Published in: Agricultural meteorology Vol. 19; no. 1; pp. 23 - 34
• Main Authors: Steyaert, Louis T., LeDuc, Sharon K., McQuigg, James D.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 1978
• ISSN: 0002-1571
• DOI: 10.1016/0002-1571(78)90035-3

Principal components of monthly sea level pressure representing large scale general circulation features such as persistent upper level troughs and ridges, blocking highs, semipermanent pressure cells, standing eddies, etc., are the predictors in a linear regression on yield for large wheat production regions in the United States, Canada, and the Soviet Union. The purpose of this modeling is to estimate national level of production and to demonstrate a link between crop forecasting and extended atmospheric outlooks. Long term, reliable records of pressure data are used. The models also benefit from the quality, reliability, and availability of foreign crop data for large areas. The monthly pressure field exhibits significant spatial collinearity which determines the time orthogonal principal components. The use of these principal components in the regression leads to a fewer number of required predictors, more stable signs on regression coefficients, minimal variance inflation of regression coefficients, and the ability to objectively partition the principal components into “real” and “noise” relationships to yield. Correlation fields are determined by correlation of pressure data to area weighted temperature or precipitation. These are used to evaluate the physical interpretation of space orthogonal eigenvector fields. The correlation fields and eigenvectors are used together to ensure that the signs on regression coefficients for principal components in the final yield equation make agronomic sense. The statistical structure of the models is discussed. Generally, the models have an explained variance of approximately 0.90 and a standard error of 1.5 quintals per hectare. For 1975 and 1976 these wheat yield models provided operational estimates which are generally within two quintals per hectare of official estimates.