Using Automatic Differentiation to study the sensitivity of a crop model

Claire Lauvernet (1) , L. Hascoet (2) , F.X. Le Dimet (3) , Frédéric Baret (4)

1 UR MALY - Milieux aquatiques, écologie et pollutions
2 IMRA Europe - Sophia Antipolis
3 UGA [2016-2019] - Université Grenoble Alpes [2016-2019]
4 UEEAA - Unité expérimentale Environnement et Agronomie d'Avignon

Automatic Differentiation methods are often applied to codes that solve partial differential equations, e.g. in the domains of geophysical sciences, such as meteorology or oceanography, or Computational Fluid Dynamics. In agronomy, the differentiation of crop model has never been performed since the models are not fully deterministic but much more empirical. This study shows the feasability of constructing the adjoint model of a crop model referent in the agronomic community (STICS) with the TAPENADE tool, and the use of this adjoint to perform some robust sensitivity analysis. This aims at giving a return of experience from users working in the environmental thematic, and presents a somewhat unusual field of application of Automatic Differentiation.

Mots clés: ANALYSE DE SENSIBILITE; COUVERT VEGETAL; ADJOINT MODE; AGRONOMIC CROP MODEL; SENSITIVITY ANALYSIS

Domaines: Sciences de l'environnement

Claire Lauvernet, L. Hascoet, F.X. Le Dimet, Frédéric Baret. Using Automatic Differentiation to study the sensitivity of a crop model. 6th International Conference on Automatic Differentiation, Jul 2012, Fort Collins, United States. 10 p. ⟨hal-01058645⟩