[Defended thesis] Daniel Pasquel : Metrics to assess the spatialisation of crop models for Precision Agriculture: application to a vine water status model

Metrics to assess the spatialisation of crop models for Precision Agriculture: application to a vine water status model
 

My name is Daniel Pasquel, and I'm doing my thesis at Inrae, within the UMR Itap and Mistea in Montpellier. After graduating from the Institut Agro de Montpellier with a degree in Agronomic Engineering, specializing in Sustainable Plant Production, I decided to pursue my studies with a thesis on agronomic modeling, a subject that has always interested me. This thesis is an opportunity for me to acquire new methods for evaluating future crop models.
My topic concerns the performance evaluation of spatialized agronomic models, with the aim of proposing new, more relevant methodological approaches for their assessment. The issues at stake are to understand how crop models are spatialized and for what reasons, and then to propose evaluation methods. My case study concerns a predictive model of water stress in grapevines (WaLIS). This model will have to be spatialized (in order to understand the mechanics of spatialization) and the proposed methods will be tested on this model.
This subject is of particular interest to me, as it deals with an innovative aspect of evaluating the performance of crop models, taking into account the spatial nature of the data, a major issue for the application of precision agriculture. My specialization in crop production, particularly viticulture following my final year internship, means that I can bring my agronomic knowledge to bear in this field, particularly when it comes to interpreting results.
From an application point of view, this subject represents an interesting prospect for the future improvement of crop models by proposing to evaluate the effect of spatialization on the prediction performance of these models. Knowing whether high-resolution auxiliary data (soil or plant physiology data) used to spatialize models improves their performance is vital for the agricultural sector, in order to make these models as efficient as possible.

• Starting date : 1st November 2020
• University : Institut Agro
• PhD school: GAIA
• Scientific field: Agronomy, modeling, precision agriculture
• Thesis management : James Taylor, UMR Itap, Inrae
• Thesis supervisors: Sébastien Roux, UMR Mistea, Inrae / Bruno Tisseyre, UMR Itap, Institut Agro 
• Funding: #DigitAg – Inrae
• #DigitAg : Cofunded thesis – Axe 6 : Modélisation et simulation (systèmes de production agricole), Axe 5: Fouille de données, analyse de données, extraction de connaissances – Challenge : sujet transversal

Keywords: Sensitivity analysis, Geostatistics, Spatialized crop models

Abstract:  There exists enormous knowledge gaps in how to best manipulate crop models to accept and to be updated with spatial (and temporal) high-resolution ancillary data, the implications that changing the model has on predictive power (and subsequent management options), and how to properly assess model performance at varying scales. Agricultural scientists need help achieve this and the crux of this thesis will be method development to generate metrics to assess the effect of incorporating multi-temporal spatial crop and environmental observations into existing crop models. The intent will be to incorporate aspects of spatial variance decomposition into a Sobol-based sensitivity analysis. The intent is to improve understanding of how to spatialize model predictions for enhanced spatial management. It will not and cannot address all issues, but will start to provide tools to achieve this. The intent is not to arrive at the best spatialize model, but to develop tools that will help all models arrive at this point. Spatial crop (agri-environmental) models will generate outputs with a change in extend, coverage and/or support from traditional crop model applications. The need for correct methods of sensitivity analysis has been previously discussed and proposed, but only for large scale, regional applications. High resolution, sub-field, agronomic applications are an area of sensitivity analysis that requires further work.

Social network: LinkedIn