A high-yielding traits experiment for modeling potential production of wheat: field experiments and AgMIP-Wheat multi-model simulations

: Grain production must increase by 60% in the next four decades to keep up with the expected population growth and food demand. A significant part of this increase must come from the improvement of staple crop grain yield potential. Crop growth simulation models combined with field experiments and crop physiology are powerful tools to quantify the impact of traits and trait combinations on grain yield potential which helps to guide breeding towards the most effective traits and trait combinations for future wheat crosses. The dataset reported here was created to analyze the value of physiological traits identified by the International Wheat Yield Partnership (IWYP) to improve wheat potential in high-yielding environments. This dataset consists of 11 growing seasons at three high-yielding locations in Buenos Aires (Argentina), Ciudad Obregon (Mexico), and Valdivia (Chile) with the spring wheat cultivar Bacanora and a high-yielding genotype selected from a doubled haploid (DH) population developed from the cross between the Bacanora and Weebil cultivars from the International Maize and Wheat Improvement Center (CIMMYT). This dataset was used in the Agricultural Model Intercomparison and Improvement Project (AgMIP) Wheat Phase 4 to evaluate crop model performance when simulating high-yielding physiological traits and to determine the potential production of wheat using an ensemble of 29 wheat crop models. The field trials were managed for non-stress conditions with full irrigation, fertilizer application, and without biotic stress. Data include local daily weather, soil characteristics and initial soil conditions, cultivar information, and crop measurements (anthesis and maturity dates, total above-ground biomass, final grain yield, yield components, and photosynthetically active radiation interception). Simulations include both daily in-season and end-of-season results for 25 crop variables simulated by 29 wheat crop models.


ORIGINAL PURPOSE:
The original purpose of this dataset was to support a model inter-comparison (Guarin et al. 2022) as part of the Agricultural Model Intercomparison and Improvement Project (AgMIP, https://agmip.org/) (Rosenzweig et al. 2013). The field experimental data were from high-yielding trait experiments to investigate and improve wheat yield potential in high-yielding environments using improved physiological traits (Bustos et al. 2013;Garcia et al. 2013;Garcia et al. 2014). This dataset contains field measurements of selectively bred high-yielding wheat cultivars, including the highest reported wheat grain yield in the literature -16.6 t ha -1 dry weight (Bustos et al. 2013;Garcia et al. 2013), for benchmarking local and regional yield improvements and model improvement against an ensemble of 29 state-of-the-art wheat crop simulation models.

FIELD EXPERIMENTS:
A full description of the experiment sites and treatments are available in Bustos et al. (2013) and Garcia et al. (2013). The critical information for the simulation of the high-yielding treatments used in the AgMIP-Wheat Phase 4 project is summarized below for crop model setup and analyses. The experiment consisted of three sites located at the University of Buenos Aires Facultad de Agronomía experimental field, Buenos Aires, Argentina (34°35' S, 58°29' W, 26 m a.s.l.), the Norman E. Borlaug experimental station, Ciudad Obregon, Mexico (27°25' N, 109°54' W, 38 m a.s.l.), and the Austral University of Chile experimental station, Valdivia, Chile (39°47' S, 73°14' W, 19 m a.s.l.). The dataset includes two spring wheat genotypes, one check cultivar (Bacanora) from the International Maize and Wheat Improvement Center (CIMMYT) and one high-yielding doubled haploid (DH) line from the cross between Bacanora and Weebil with improved radiation use efficiency (RUE), light extinction coefficient (K), potential grain filling rate (GFR), and potential grain size (GWpot) and slightly decreased fruiting efficiency (FE) and grain filling duration (GFD) ( Table 1). The entire experiment consisted of 105 spring wheat DH lines, but only the best-yielding DH lines for each location are reported here.
Each growing year consisted of one to three replicates where the wheat crops were grown with ample N supply, full irrigation, and agronomic practices to reach potential yield for the local soil and weather conditions. All other crop factors including weed, disease and pest control, and potassium, phosphate, and sulphur fertilizers, were applied at levels to prevent yield limitation. The soil at Buenos Aires was a silty clay loam, classified as Vertic Argiudoll, and each replicate was sown in flat plots with five rows 2.1 m long by 0.9 m wide and 0.175 m between rows. The soil at Ciudad Obregon was a sandy clay, classified as Typic Caliciorthid, and each replicate was sown in a 2.5 m long by 0.8 m wide plot consisting of one raised bed with two rows per bed and 0.25 m between rows. The soil at Valdivia was a volcanic ash, classified as a Typic Hapludand, and each replicate was sown in either a continuous plot of three rows 1.5 m long with 0.15 m between rows (2008 and 2009) or split plots 2 m long by 1.5 m wide with 0.15 m between rows (2010). At each site, the temperature and solar radiation data were provided from a weather station located < 2 km from the experimental field and the rainfall, wind speed, and relative humidity data were obtained using the NASA POWER database (https://power.larc.nasa.gov) (Kratz et al. 2014;White et al. 2011). The average grain yield of each treatment for the three high-yielding locations is shown in Figure 1.

SIMULATION OF FIELD EXPERIMENTS:
The treatments described above were simulated by 29 wheat crop models (Guarin et al. (2022); see CIM_AgMIP_model_names.txt). Simulations were carried out using standard AgMIP protocols (Rosenzweig et al. 2013;Asseng et al. 2015) in two steps, one step for model calibration for the check cultivar Bacanora, and the second step for 'blind' simulations of the high-yielding DH line. The simulation results reported here are for both steps. For step one modelers had access to all the experimental data for the check cultivar, Bacanora, for the five growing seasons at Valdivia, Chile (2008-2009, 2009-2010, 2012-2014and 2014-2015, one growing season at Buenos Aires, Argentina (2009Argentina ( -2010 and four growing seasons at Ciudad Obregon, Mexico (2009Mexico ( -2010Mexico ( , 1990Mexico ( -1991Mexico ( , 2015Mexico ( -2016Mexico ( and 2016Mexico ( -2017 2017). Detailed initial soil conditions were not available for each location. Therefore, as water and nitrogen (N) were managed to limit any stress, total initial soil mineral N (NO 3 − and NH 4 + ) content was assumed to be equal to 140 kg N ha -1 . To ensure no water stress, supplementary irrigation was provided. For each experiment, the dates and rates of irrigation were calculated using the DSSAT-NWheat model (Kassie et al. 2016) automatic irrigation routine. Modelers used either the irrigation dates and rates provided by DSSAT-NWheat or their model-integrated unlimited water and N routine to prevent any simulated water or N stress. For step two, a 'blind' simulation study was conducted for the best-yielding DH lines at each location using the same initial growing and management conditions from the calibration, but the measured data were not provided. One additional season at Valdivia, Chile was included (2010)(2011). Only calculated trait percent changes (Table 1 final column) and instructions describing how to modify the calibrated cv. Bacanora traits for the high-yielding DH line (Guarin et al. 2022) were provided to simulate growth for the seasons that the DH line was grown, i.e., three seasons at Valdivia, Chile (2008-2009, 2009-2010, and 2010-2011, one season at Buenos Aires, Argentina (2009Argentina ( -2010, and one season at Ciudad Obregon, Mexico (2009Mexico ( -2010. The RUE and K of the DH line were calculated using the average of the two best-yielding DH lines from Chile because detailed light interception data was only measured in Chile. The FE, GWpot, GFD, and GFR were calculated using the mean percent change between the best-yielding DH line and Bacanora from each of the three locations (Table 1). In addition to the five 'blind' DH line treatments, the five Bacanora treatments corresponding to these treatments were re-simulated in step two for comparison to step one to ensure model consistency. Model outputs include emergence date, anthesis date, maturity date, grain dry mass yields, total aboveground biomass, leaf area index, number of grains per square meter, grain dry weight, harvest index, crop N dynamics, crop transpiration and evapotranspiration, soil water and N dynamics, and intercepted photosynthetically active radiation (PAR). Not all models simulated all variables. Variables not simulated are indicated by "NA". Simulation results are reported for each individual model and for the multi-model ensemble median (e.median).

DATA FORMAT, STRUCTURE, AND AVAILABILITY:
An overview of the main tables and files from the data is given in Table 2. Experimental (means of crop measurements) and simulation (model output) data, model input (cultivar information and crop management), soil description, initial conditions, and daily weather data (incoming solar radiation, maximum and minimum air temperature, rainfall, wind speed, dew point temperature, vapor pressure, and relative humidity) for simulation setup are provided in a Microsoft Excel (version 2019) and a JavaScript Object Notation (JSON) file. These files follow the AgMIP Crop Experiment (ACE) data schema. The ACE JSON file was created from the Microsoft Excel file by using the data translator available at https://github.com/agmip/translator-excel-python. The ACE JSON can be used to create model input files using QuadUI desktop utility for ACE input and output data translation (http://tools.agmip.org) or model-integrated translators (Porter et al. 2014). Data are also provided in tabdelimited text format. All text files are UTF-8 encoded. The names, descriptions, and units of the variables (key) are provided in the Microsoft Excel file and in text files with their correspondence and conversion factors in the International Consortium for Agricultural Systems Applications (ICASA) standard (White et al. 2013). Data available at https://doi.org/10.7910/DVN/VKWKUP.