• Who are we?
  • In a nutshell
  • Our mandate
  • Our strategy
  • Our partnerships
  • CIRAD worldwide
  • Our values
• Research operations
  • Priority lines of research
  • Research Topics
  • Supply chains
  • Research units
  • Collective research tools
  • Research results
• Teaching & training
  • PhD training
  • Higher education
  • Available training
  • Scientific experience
  • E-learning
  • Grants
• Innovation & expertise
  • Partnerships for innovation
  • Skills and expertise
  • Products and services
  • Technology transfer
  • The CIRAD VIP newsletter
• Publications & resources
  • Documentary resources
  • Publications
  • Science for all
  • Vidéos
  • Website Directory

The high interannual rainfall variability that prevails in arid and semiarid regions like the Sahel is a serious constraint when trying to choose the best varieties. Farmers require varieties that will generate a steady income, thus that are not very sensitive to climatic variations. To make the best possible choice, plant breeders test the progenies of their crosses in multilocation trials lasting several years. They usually note a substantial interaction between the genotype and the environment, ie the progenies may be ranked differently depending on the year or the trial site. Theoretically, when breeding varieties for a target locality, selection should be carried out on a sample of years representing the interannual rainfall variability. In the Sahel, however, where this variability is high, 20 years of testing would be required to be able to compare varieties with sufficient precision. In practice, each variety is monitored for only two-three years, and breeders use the multilocation data to predict the potential behaviour of the variety over several years. This informal geographicalto-temporal deduction step is mainly dependent on the breeder’s expertise, but no satisfactory tool has been available to date that could validate the breeder’s conclusions.

From simulation of a variety’s growth to prediction of its behaviour

Historical meteorological data that account for the interannual climatic variability are nevertheless available for all cropping areas. Plant growth simulation models such as SarraH, which was designed by CIRAD and CERAAS, can predict genotype x environment interactions from these meteorological data. This model simulates a plant cover in its environment, ie growth patterns of a plant and its reaction to water stress, while also simulating the crop water balance. The parameters that control these phenomena are specific to each variety. The SarraH model can thus account for an environmental effect that may differ between varieties, ie a variety x environment interaction. However, the parameters of such a model are only available for a few varieties. Specific trials and many observations are required to be able to estimate them for new varieties (model parameterization)—this would be much too expensive to carry out for all varieties during their selection process, some of which will inevitably be discarded.

The APLAT (approximation via linearization in the neighbourhood of a control) method was developed at CIRAD within the framework of the PhD thesis research of a CERAAS scientist. APLAT overcomes this costly parameterization problem as only standard multilocation or multiyear experiments are required. However, in each trial, a control variety with known parameters must be present, and input data for the growth simulations must be recorded. For the SarraH model, a rain gauge should be used in these trials and they should be conducted in the vicinity of a more comprehensive meteorological station. APLAT is based on linearization of the crop’s response to variations in SarraH varietal parameters.

The APLAT method can generally be used to predict the behaviour of new varieties in environments where they have not been studied. This generic method permits genotype x environment interaction predictions via any crop simulation model, while avoiding expensive parameterization of the model for these new varieties. It improves the breeding accuracy by offering the operator the possibility of simulating a range of environments representative of soil-climate situations targeted by the breeding programme. It thus enhances plant breeding methods by reducing the impact of climatic variability on the accuracy of between-variety comparisons.

Conclusive tests in groundnut

APLAT was tested on groundnut varieties using the SarraH model. The data used to validate the method included historical data from selection trials carried out on groundnut over several years at the Bambey research station in Senegal, where climatic data are regularly recorded, as well as data from a multilocation trial carried out at eleven sites in Senegal during the 2005 rainy season. APLAT reduced the confidence intervals for between-variety differences by half for the multilocation trial.

Advances in plant breeding methods have had unexpected spinoffs for chemical analysis. The partial least squares (PLS) regression method had to be extended to trials with several sources of variability (plot, site, year) when implementing APLAT. This mixed PLS extension can also be directly used for a chemical analysis application, ie for calibration of near-infrared reflectance spectrometry (NIRS) absorption measurement methods from collected field data with several sources of variability, such as split plots.

Contact

Eric Gozé, E-mail , Annual Cropping Systems (UPR)
Ibnou Dieng,AGRHYMET

Partners

• Universities of Montpellier I and Montpellier II (France)
• Centre d’étude régional pour l’amélioration de l’adaptation à la sécheresse (CERAAS, Senegal)
• Institut national de la recherche agronomique (INRA, France).

For further information

• Dieng I., 2006. Mixed PLS models to predict genotype x environment interaction. In : XXIIIrd International biometric conference, International Biometric Society, 16-21 juillet 2006, Montréal, Canada.
• Dieng I., 2007. Prédiction de l’interaction génotype-environnement par linéarisation et régression PLS-mixte. Thèse de doctorat, université Montpellier II, 133 p.
• Dieng I., Gozé E., Sabatier R., 2006. Linéarisation autour d’un témoin pour prédire la réponse des cultures. Comptes rendus de l’Académie des sciences, Biologie, 329 : 148-155.
• Dieng I., Gozé E., Sabatier R., 2006. Prédiction de l’interaction génotype-environnement par linéarisation autour d’un témoin : nouveaux résultats pour la méthode Aplat. In : 38e Journées de statistiques, 29 mai - 2 juin 2006, Clamart, France. Société française de statistique, EDF recherche et développement.