Microbial-based predictive modeling of wheat yield and grain baking quality

It is very difficult to predict crop yield and produce quality based solely on soil physicochemical parameters, as the net effect these parameters is strongly affected by microbes. For instance, the form of soil nitrogen changes due to nitrification and denitrification activities, which will influence N mobility, availability, and energetic efficiency for plant growth. It is crucial to include microbial parameters to better predict crop yields and produce quality. However, microbial communities vary spatially and temporally, and are very complex, so it is uncertain if robust models could be derived from microbial data. My goal for this thesis was to create microbial-based models to predict wheat grain quality and yields across time and space. I used two sampling schemes: 1) early season sampling of 80 wheat fields across the province of Québec (Chapter 2) and 2) repeated sampling of a single wheat field across a growing season (Chapter 3). For both these experiments, I measured a wide array of microbial parameters: 16S rRNA gene and ITS region amplicon sequencing, qPCR quantification of key N-cycle genes, and microbial community level carbon usage. Grain baking quality and grain yields were measured at the end of the growing season. I used linear regression with stepwise forward/backward (Chapter 2) or LASSO selection (Chapter 3), limiting the models in most cases to less than 10 microbial indicators. In Chapter 2, I was able to explain observed variation of wheat grain quality and yields with an accuracy of up to 90% across all fields. Many of the inputs selected in the models had a link with soil nitrogen availability (e.g., ammonia-oxidizers and denitrifiers abundance). My microbial-based models also outperformed similar models based on commonly measured soil parameters (pH, total C, total N, C/N ratio, water content). However, in this Chapter, I had sampled the fields early in the growing season, and it was not certain that this was the best to create my pre