Dynamic nitrogen fertigation with reflectance sensors: exploration of statistical modeling approaches to optimize N fertilization in processing tomato (Solanum lycopersicum L.)

Dynamic nitrogen (N) fertigation guided by reflectance sensors presents a valuable opportunity to improve the N use efficiency in vegetable cropping systems, and this thesis explores multiple modeling approaches to optimize N fertilization in processing tomato using multispectral and hyperspectral reflectance sensors. The development of threshold curves of the green vegetation index (GVI) to trigger the N fertigation represents the simplest approach to guide the dynamic fertigation, and multiple modelling strategies to build the GVI thresholds were explored in the thesis. Besides the GVI threshold adopted, field validation trial demonstrated that dynamic N fertigation guided by GVI thresholds can save up to 38-60% of N fertilizers as compared to conventional N fertilization. Dynamic fertigation guided by the GVI helps maintain yield levels, reduce fertilization costs and greenhouse gas emissions, and potentially enhance fruit quality. Among the GVI threshold curves, monitoring the GVI in crops under non-limiting N conditions in the past growing seasons represents an effective, simple, and fast modeling approach. Furthermore, different approaches to defining the optimal N rate were investigated. A low-computational demanding approach, consisting of the integration of the N balance sheet with the GVI threshold curve, was successfully validated in a field trial. On the other hand, reflectance data were used to retrieve crop traits, including aboveground biomass (AGB), N uptake, leaf area index (LAI), and the Nitrogen Nutrition Index (NNI). In turn, these crop parameters were used to initialize a crop model, the critical N uptake curve, to calculate the optimal N rate. Different regression approaches were explored to retrieve such parameters, including linear regression with vegetation indices, nonlinear regression with vegetation indices hybridized with agroclimatic data, and linear and nonlinear non-parametric regression (machine learning regression algorithms). The thesis encompasses the research findings as well as future priorities for investigation.