Forecasting the Harvest Date and Yield of Sweet Corn by Complex Regression Models

Abstract

Predicting sweet corn (Zea mays var. rugosa Bonaf.) harvest dates based on simple linear regression has failed to provide planting schedules that result in the uniform delivery of raw product to processing plants. Adjusting for the date that the field was at 80% silk in one model improved the forecast accuracy if year, field location, cultivar, soil albedo, herbicide family used, kernel moisture, and planting date were used as independent variables. Among predictive models, forecasting the Julian harvest date had the highest correlation with independent variables (R2 = 0.943) and the lowest coefficient of variation (cv = 1.31%). In a model predicting growing-degree days between planting date and harvest, R2 (coefficient of determination) = 0.85 and cv = 2.79%. In the model predicting sunlight hours between planting and harvest, R2 = 0.88 and cv = 6.41%. Predicting the Julian harvest date using several independent variables was more accurate than other models using a simple linear regression based on growing-degree days when compared to actual harvest time.