Speaker
Description
Crop growth models - or process-based models - simulate the dynamic responses of a range of varieties (G) as a function of environmental conditions (E) and management practices (M) and hence are appropriate tools to predict and explain G×E×M interactions (Chapman, 2008 ; Wang et al., 2019). Therefore such models could have practical applications for improving the design and analysis of multi-environment trials (METs) used for variety testing and help to identify specific fits between the tested varieties and their optimal cropping conditions (Jeuffroy et al., 2014). Developed for sunflower crop, the SUNFLO model (Casadebaig et al., 2011; Casadebaig et al., 2016) simulates performance-related variables (grain yield, seed oil concentration) and environmental-related indicators (abiotic stress effect on photosynthesis) of a wide range of varieties under contrasting cropping conditions. However it is not clear how actors in variety choice can integrate such models into decision support systems, and how their accuracy is impacted by operating crop growth models at a large geographical scale.
In this study conducted in the frame of a EU project aiming to foster the introduction of new varieties better adapted to varying biotic and abiotic conditions (H2020-INVITE, 2019-2024), the predictive accuracy of the SUNFLO model was evaluated over the key METs conducted in France for registration (GEVES) and post-registration (Terres Inovia) of sunflower varieties from 2003 to 2020, offering an unprecedented range of explored E, M and G modalities (1471 trials, 494 distinct cultivars).
While the model inputs related to management (sowing date, dates and amounts of N fertilization and irrigation) were directly collected from the structures that conducted the trials, the inputs related to climate and soil were derived from gridded public datasets. Daily weather data were systematically derived from the SAFRAN database (8 x 8 km cells) and soil data were extracted from the Geographic Database of Soils of France (millionth map). The cultivar-dependent parameters were collected from independent and routine trials conducted each year by INRAE and Terres Inovia. In this set of dedicated field and semi-controlled experiments, 9 parameters related to crop phenology, leaf architecture, plant response to water deficit, and biomass allocation were measured on a set of 133 cultivars.
Over these two trial networks, the flowering date (n = 3174), grain yield (n = 8652) and seed oil content (n = 6054) were predicted with an error (RMSE) of 5.9 days, 0.89 t/ha, and 5.9 % (relative error of 2 %, 27 % and 12 %, respectively). We concluded that grain yield prediction was not accurate enough to separate among elite varieties when using routine variety trials without additional data on the trial conditions (such as accurate weather records and a sound estimation of available soil water content).
While the model capacity to simulated GxE interactions was not granted, we showed that the simulated G and E effects were accurate separately. We proceeded by clustering trial locations and years into comparable environment-types based on the simulated abiotic stress patterns at trial level (Casadebaig et al., 2022). We showed that the sunflower growing area was composed of 4 types of situations (Figure): 45 % « cold », 30 % « optimal », 19 % « drought », and 6 % « heat ». This contextual information was embedded into a prototype of a decision-support system intended for choosing the best varieties (based on trial results) for each type of environment (based on simulation results).
References
Casadebaig P. et al., 2011. Agricultural Forest Meteorology 151, 163-178.
Casadebaig P. et al., 2016. European Journal of Agronomy 81, 92-105.
Casadebaig P. et al., 2022. Theoretical and Applied Genetics 135, 4049-4063
Chapman S.C., 2008. Euphytica 161, 195-208.
Jeuffroy M.H et al., 2014. Agronomy for Sustainable Development 34, 121-137
Wang E. et al., 2019. Journal of Experimental Botany 70, 2389-2401
| Keywords | crop model ; variety testing ; sunflower ; environment characterization ; multi-environment trial |
|---|
