Evaluation of five thermal time models for modelling spring wheat phenological development on the Canadian Prairies

Temperature (heat) is one of the most important weather factors that regulate plant growth and development. Understanding and accurately predicting crop development stage (phenology) is essential to many aspects of crop production including timing of crop management practices such as fungicides, herbicides, pesticides and fertiliser applications. The thermal time concept is used extensively to assess crop development rate (DR) as impacted by temperature and can be correlated to crop development. There are numerous different thermal time models intended to approximate crop phenological development, each with strengths and weaknesses. The most frequently used thermal time models include the growing degree-days (GDD), which relates DR linearly to temperature and the beta function (BF), which relates DR to temperature nonlinearly. The purpose of this analysis was to compare and contrast five different thermal time models in order to identify the best model for modelling spring wheat phenology on the Canadian Prairies. Crop and weather data collected during the 2009, 2010 and 2011 cropping seasons from Carman, Regina, Melita, Hamiota, Swift Current, Melfort and Saskatoon were used in the analysis. Five thermal time models including NDGDD (developed and used in North Dakota), GDD0 (base temperature zero), GDD5 (base temperature 5), Beta Function (BF) and weighted Modified Beta Function (MBF) were tested for their ability to predict phonological stages (from seeding to anthesis) of three widely grown spring wheat cultivars including AC Barrie, AC Intrepid and BW874. The ability of each model to predict time (calendar days) from seeding to anthesis was tested using wheat phenology data collected in 2011 from the above mentioned sites and data collected from 2003 through 2006 from five other experimental sites across the Prairies.

Results showed that accumulated GDD/daily growth rates calculated using the different models correlated well with spring wheat phonological development (seeding to anthesis) with R2 ranging from 0.91 to 0.94 and p<0.001. However, when the developed models were used to predict time (calendar days) from seeding to anthesis for cultivar AC Barrie, the BF and MBF performed slightly poorly compared to the other models. Overall, the predicted time (calendar days) from seeding to anthesis by the NDGDD, GDD0, GDD5, BF and MBF models were 64, 64, 63, 65 and 65 days, respectively; while the observed time was 60 days. A student t-test showed that the time (calendar days from seeding to anthesis) predicted by the NDGDD, GDD0 and GDD5 was statistically similar (p>0.05) to the observed time. However, the time predicted by the BF and MBF models was significantly higher (p<0.05) than the observed time. The root mean square error (RMSE) for the NDGDD, GDD0 and GDD5 was 5 while that for the BF was 6 and that for the MBF was 7. These findings suggest that the NDGDD model, which WeatherFarm.com has adopted and deployed across the Canadian Prairies for modelling spring wheat phenology, is a good model. Nevertheless, the model has to be constantly validated and updated as new wheat varieties come into production and the impact of climate change/variability becomes apparent.