Remote Sensing and Artificial intelligence as agronomic support for field action

The agriculture sector is under pressure from multiple perspectives such as global warming, floods, hailstorms, and the current geopolitical conflicts that led to the rise in production input costs, all occurring during a time of unprecedented world population growth. The agronomic scientific community is working on various fronts to find sustainable solutions to face the current situation, focusing especially on the efficiency of the use of production inputs, such as water resources and both mineral and organic inputs, to sustain necessary production levels. There are several sustainable agronomic approaches that have already demonstrated their ability to sustain production, such as conservation agriculture, organic farming, precision agriculture and their combinations. Thanks to the remote sensing data, nowadays we now have easy and free access to an ever-growing source of data that allow us to gain a new perspective of our fields, which is “no longer looking at our fields from below but having a constant general overview from above.” The large amount of data does not necessarily lead to a more efficient agricultural system; in fact in some cases, it can cause confusion or make it difficult to interpret and use such data. This is precisely where artificial intelligence, with its ability to find patterns in data, can help. It allows us to synthesize the information we have about a field and, based on that, identify the most effective agronomic management strategy. This thesis aims to analyze, over seven chapters, the viability and application of two major sustainable agronomic management strategies, such as conservative agriculture and precision farming, and explore how artificial intelligence can be used in defining agronomic management strategies for rainfed durum wheat. During the PhD, multiple sources of data were collected from various sites across Italy’s major durum wheat production regions, including phenology, yield components, multispectral and pedo-climatic data from different experimental sites with different experimental designs. The results can be summarized as follows: The repeated application of conservation agriculture results in higher production levels compared to conventional methods, provided no mineral fertilizer is applied. The only agronomic management practice that allows production levels to remain unchanged in the event of a 2 °C increase in air temperature is conservation agriculture. Multispectral imaging is an excellent source of data for monitoring the phenological development of crops and identify field areas experiencing stress. By using this data and automatic clustering procedures, validated agronomic prescription maps can be generated, enabling differential input application based on crop conditions. The combined use of different data sources representing the soil-crop-climate-agronomic management system can be used to train artificial intelligence algorithms, helping answer “if-then” questions and facilitating the identification of the best soil and fertilization management strategies. The integration of data source, domain expertise, and agronomic management could lead to a new agronomic management strategy that preserves the achievements of past research while merging this knowledge with new technologies, helping users better understand the context and take the most appropriate actions.