Skip to Main content Skip to Navigation
Theses

Co-développement d'un modèle d'aide à la décision pour la gestion intégrée de la flore adventice. Méta-modélisation et analyse de sensibilité d'un modèle mécaniste complexe (FLORSYS) des effets des systèmes de culture sur les services et disservices écosystémiques de la flore adventice

Abstract : The main threat to agricultural crops are weeds with herbicides being the primary cropping management practice. Due to the negative impact of herbicides on health and environment, their use must be reduced. To replace herbicides, numerous cropping practices need to be implemented. This makes weed management more complicated and, together with necessity of scheduling operations at long-term and the multiplicity of cropping system impacts, explains why models are so useful for designing innovative cropping systems. The aim of this thesis was to develop a Decision Support System (DSS) intended for crop advisors to help design cropping systems that are less dependent on herbicides. Our approach consisted in identifying the structure of the DSS in interaction with future users while using an existing research model for the biophysical content of the tool. The research model, FLORSYS, is a mechanistic model of a “virtual field” simulating the weed flora dynamics from both cropping systems and pedoclimatic conditions. As output, it provides weed impact indicators, both for crop production and ecosystem services. This work was achieved via three steps: (1) increasing the speed of FLORSYS simulations by metamodelling, via polynomial chaos extension, the slowest submodel, i.e. the radiation interception. This submodel is based on a 3D representation of each plant of the crop:weed canopy where the transfer and interception of the radiation is computed at a voxel (3D pixel) scale. (2) The identification of the most influential cropping practices on weed impacts indicators via datamining on a wide range of simulated cropping systems with FLORSYS. To widen the range of cropping systems, random cropping systems were added to the study. Random forest allowed the construction of charts of the most important cropping practices contingent on different objectives; classification and regression trees gave the optimal combinations of practices leading to those objectives. (3) Simultaneously, the co-development of the structure of the DSS with future users was done via surveys and workshops to design cropping systems. These interactions aimed at defining which questions should be answered with the DSS, which inputs and outputs, and with which format, should be used. In step (1), the simulation time of the voxelised FLORSYS and the FLORSYS metamodelled at the scale of the plant highlighted the precision/speed trade-off of the model. The metamodel is faster for agricultural practices requiring precise knowledge of the position and volume of plants. In other cases, decreasing the precision by using the voxelized FLORSYS with an increased voxel size is much faster. Step (2) shows no general conflict between crop protection and ecosystem service indicators. The conservation of both crop protection and ecosystem services is thus possible, except for bee food offer. To reduce weed impacts on production, tillage, especially in summer, and herbicide use are the main cropping techniques. In step (3) two use cases were identified: (1st) the redesign of a cropping system requires a synthetic tool, based on meta-decision rules (e.g. rotating winter and spring crops, ploughing once every three years); (2nd) adjusting cropping practices requires a detailed tool, with detailed inputs on the cropping system. The synthetic DSS is composed of charts of the most important cropping system practices, of decision trees, and an emulator of FLORSYS based on random forests. Its capacity to predict known facts makes it an essential discussion medium for cropping system design workshops, particularly to suggest the outlines of multi-performant cropping systems. The detailed DSS consists of a pre-parameterized version of FLORSYS for different regions, to allow the fine tuning of cropping systems to constraints and objectives of farmers.
Document type :
Theses
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-01829696
Contributor : Abes Star :  Contact
Submitted on : Wednesday, July 4, 2018 - 11:39:07 AM
Last modification on : Thursday, July 30, 2020 - 10:56:06 AM
Long-term archiving on: : Monday, October 1, 2018 - 1:09:23 PM

File

78751_COLAS_2018_archivage.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-01829696, version 1

Collections

Citation

Floriane Colas. Co-développement d'un modèle d'aide à la décision pour la gestion intégrée de la flore adventice. Méta-modélisation et analyse de sensibilité d'un modèle mécaniste complexe (FLORSYS) des effets des systèmes de culture sur les services et disservices écosystémiques de la flore adventice. Sciences agricoles. Université Bourgogne Franche-Comté, 2018. Français. ⟨NNT : 2018UBFCK014⟩. ⟨tel-01829696⟩

Share

Metrics

Record views

173

Files downloads

173