Evaluating multiple rice crop models for response to elevated temperature treatments in sunlit, controlled-environment chambers

Authors: BOOTE, KENNETH - University Of Florida; Baker, Jeffrey; LI, TAO - International Rice Research Institute; HASEGAWA, TOSHIHIRO - National Agricultural Research Center - Japan; YIN, XINYOU - University Of Wageningen; ZHU, YAN - Nanjing University; ALLEN, LEON - Retired ARS Employee; ADAM, MYRIAM - Cirad, France; BREGAGLIO, SIMONE - State University Of Milano; BUIS, SAMUEL - Inra, Génétique Animale Et Biologie Intégrative , Jouy-En-josas, France

Submitted to: Agricultural Engineering International Conference
Publication Type: Abstract Only
Publication Acceptance Date: 10/2/2019
Publication Date: 10/22/2019
Citation: Boote, K., Baker, J.T., Li, T., Hasegawa, T., Yin, X., Zhu, Y., Allen, L.H., Adam, M., Bregaglio, S., Buis, S. 2019. Evaluating multiple rice crop models for response to elevated temperature treatments in sunlit, controlled-environment chambers. Agricultural Engineering International Conference. Presentation. Lake Garda, Gargnano, Italy, 22 -25 October 2019.

Interpretive Summary: Crop simulation models are mathematical computer programs that were built to predict the effects of future climate change on agricultural crop yields. An international group of scientists and crop modelers compared predictions of 15 rice crop simulation models. These crop models were tested against data from CO2 enrichment and air temperatures experiments conducted on rice from around the world. In this report we focus on rice data sets collected at Gainesville, FL, USA in specially built, outdoor controlled environment plant growth chambers. This exercise identified weakness in some of the crop models. Specifically we found over prediction of grain yield at high air temperature treatments due to a lack of model sensitivity to high air temperatures under CO2 enrichment. These findings are being used to improve accuracy of these crop models to predict the impacts of climate change on future rice grain yields.

Technical Abstract: The AgMIP-Rice Team evaluated 15 rice crop models against growth and yield data measured on IR-30 cultivar grown in elevated temperature experiments conducted in sunlit, controlled environment chambers. Initially, modelers were given data only for selected moderate temperature treatments and were requested to calibrate model parameters for cultivar life cycle, grain yield, and biomass. Modelers then simulated additional elevated temperature treatments where no crop information was provided (blind phase). In the blind phase, only two models gave acceptable simulated responses to elevated temperature, and many models individually and as an ensemble, over-predicted grain yield at elevated temperature. The failures were associated with over-prediction of grain number at elevated temperature. Then modelers were provided with data from all temperature treatments of Exp 4 and allowed to modify model parameterization for temperature relationships. After calibration to Exp 4, eight of the models were improved (lower RMSE and higher d-statistic) for simulating biomass, grain yield, and grain number of Exp 4. Validation of the models with independent data of Exp 3 showed that model calibration to Exp 4 had improved predictions in Exp 3 (reducing RMSE by 11 to 36%, and increasing d-statistic averaged over all 15 models). Eight models were improved for RMSE and d-statistic with Exp 3 data, mostly caused by modifying temperature functions affecting grain number and thus grain yield. The better models included simulation of grain number. In uncalibrated state, three of the models were frequently as good or better than the Emedian or Emean of ensemble, so caution is warranted to conclude that ensemble Emedian and Emean are always best. The d-statistics of ensemble Emedian were 0.85, 0.96, and 0.95, respectively, for biomass, grain yield, and grain number of independent Exp. 3 data, while the respective modeling efficiencies of Emedian were 0.51, 0.89, and 0.84. Figures and statistics of model calibration and validation will be presented. Additional effort is needed to document temperature functions in the models, both the existing functions and those that were modified to improve model predictions.