INSECT ECOLOGY AND SUSTAINABLE SYSTEMS FOR INSECT PEST MANAGEMENT IN THE SOUTHEASTERN REGION
Location: Crop Protection and Management Research
Title: Use of statistical and conceptual path models to predict corn yields across management-zones on the Southeast coastal plain
Submitted to: Journal of Agricultural Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: February 14, 2013
Publication Date: February 1, 2013
Citation: Savabi, M.R., Scully, B.T., Strickland, T.C., Sullivan, D.G., Hubbard, R.K. 2013. Use of statistical and conceptual path models to predict corn yields across management-zones on the Southeast coastal plain. Journal of Agricultural Science. 1(2)32-51.
Interpretive Summary: Corn yield is a response to a complex set of direct and indirect components of the agricultural environment, which often interact positively or negatively. Yield is further influenced by the genetics of the hybrid and the pathways that contribute to that outcome. In this research, grain yield of the corn hybrid PX 31G98 is assessed through a set of pathways that include both the components of yield and the soil and water properties that are crucial to successful crop production. The analytical statistic know as “Path Coefficient Analysis” is compared to the more common multiple regression method to help illuminate not only the correlative relationship between 7 yield components and 27 soil properties, but also the causal relationships. Research fields were delineated into management-zones, with Path Analysis statistics applied to individual management-zones and then across all zones combined. Path Analysis results were compared to commonly used multiple regression statistics and both revealed that the average number of kernels per ear and the test weight of 100 kernels were the best predictors of yield. Path Analysis further identified the soil properties of soil carbon, clay content and carbon/nitrogen ratio within the 15 to 35 cm depth as having the largest influence on crop yield. Path Analysis provided more insight into the complex relationships between variables than did simple correlation and/or stepwise multiple regression analysis and is a useful tool for understanding yield across highly irregular crop environments and management-zones.
Corn, cotton and peanuts dominate the highly irregular agricultural landscape of the U.S. southeastern Coastal Plain, and efficient management of soil, water and nutrients is crucial for successful crop production. In this research, Path Analysis is investigated as a diagnostic tool to separate the direct and indirect effect of soil properties, corn yield components and crop yield, and as a technique to better explain the underlying causal relationship between these properties and components. Two fields of 1.54 and 1.62 ha in Berrien County, Georgia, USA (83° 21’ 09.96” W, 31° 22’ 37.89” N) were delineated into six management-zones based on the “fuzzy-c means” cluster technique. Statistical analyses were combined and applied across the entire research site, then to each management-zone. Crop yield was defined as a function of seven yield components and 27 soil, water and nutrient properties. Results from standard multiple regression procedures were compared to the results derived from three conceptual Path Analysis models, which separated yield components and soil properties into direct and indirect effects. Stepwise multiple regression revealed that the average number of kernels per ear and the test weight of 100 kernels were the best predictors of yield (R2=99.6), while eight soil properties provided the best prediction of yield (R2 = 59.2) across all management-zones. The most predictive Path Analysis model for yield only used yield components as 1st and 2nd order variables, while another model used yield components and soil properties as 1st, 2nd, and 3rd order variables, and identified soil carbon, clay content and carbon/nitrogen ratio within the 15 to 35 cm depth as having the largest influence on ear width, which subsequently influenced crop yield. Path analysis provided more insight into the complex relationships between variables than did simple correlation and/or stepwise multiple regression analysis and may provide utility in assessing yield across highly variable production environments and management-zones.