P0903 Random Regression Models: Incorporating Weather and Soil Information for Predicting of Genotype by Environment Interactions

Ani A. Elias , Purdue University
Kelly R. Robbins , Dow AgroSciences, Indianapolis, IN
Dev Niyogi , Purdue University
James J. Camberato , Purdue University
Rebecca Doerge , Purdue University, West Lafayette, IN
Mitch Tuinstra , Purdue University
Multi-environment field studies in crops are highly influenced by environmental factors.  Diverse soils and continuously changing weather conditions not only effect over-all performance, but often result in strong genotype by environment interactions (GE). Many commonly used GE analyses treat each location as a unique environment with GE estimates often nested within location.  In reality locations do not always represent distinct environmental events, as many environment components vary continuously throughout the growing season and determine the extent to which locations are environmentally correlated.  Ideally, multi-environment models should utilize detailed environmental information to obtained more accurate GE estimates.  In this study, random regression models (RR) incorporating continuous measurements of weather variables and soil data for prediction of GE are examined.  Data consisted of multi-environment field trials with detailed weather and soil data collected throughout the growing season.  The soil and weather data were decomposed into principle components (PC) to use as environmental indexes in the RR.   Performance of both the RR and PC were evaluated using Akaike’s information criterion (AIC), and predictive ability was examined using cross validation.  Preliminary results show several PC have predictive ability when incorporated into RR models, indicating these methods may provide a more powerful tool for dissecting GE.