Reliability of Predicting Spring Wheat Yield with DSSAT Using Early Season Weather Data.

Crop simulation models (CSM) can be effective predictive tools to help producers make critical in-season crop management decisions. The objective of this research was to evaluate the Decision Support System for Agrotechnology Transfer (DSSAT) in predicting hard red spring wheat (HRSW) yield at various points in the growing season to determine how early the model was accurately able to predict final grain yield. Stochastic, deterministic, and analogue modeling approaches were used to simulate HRSW yield from three growth stages for five locations throughout North Dakota, with historic weather data utilized from North Dakota Agricultural Weather Network (NDAWN). A regional spring wheat cultivar was calibrated from vegetative and reproductive data from North Dakota Research Extension Centers with an r² value of 0.46 and RMSE of 815 kg ha^-1, and an r² value of 0.70 and RMSE of 2.2 days between simulated and observed data for anthesis date and yield. Model performance was evaluated by analyzing the difference between the anthesis date and yield for the entire growing season and the predicted anthesis date and yield for each growth stage and modeling approach. The smallest difference indicated better model performance. The optimum approach for yield prediction was through a stochastic modeling approach with an average deviation of 486 kg ha^-1 from full season simulations. An analogue and deterministic approach were non-significantly different with an average deviation of 684 and 695 kg ha^-1, indicating neither approach was better suited for yield prediction. The optimum growth stage to predict end season yield was from Zadoks growth Stage (ZGS) 58 with an average deviation of 331 kg ha^-1. Prediction from ZGS 45 was significantly different with an average deviation of 609 kg ha^-1 and ZGS 21 with 889 kg ha^-1.