Lodato, Francesco (2024) Remote Sensing and AI for Agriculture: Multiscale Monitoring Solutions for Sustainable Agro-Environmental Management. [Tesi di dottorato]

Documento PDF Lodato_Francesco___37.pdf Download (87MB)

Abstract

The agricultural sector is on the brink of a new digital revolution. With the availability of vast amounts of data generated from both ground-based and remote sensors, alongside advanced data processing capabilities provided by cloud-based systems, it has become feasible to monitor and manage the complex and dynamic agricultural landscape on a large scale. By applying machine learning (ML) and deep learning (DL) algorithms, this study aims to tackle the inherent unpredictability of agricultural production by providing more precise and scalable monitoring solutions. The first phase of the research focuses on a reduced geographical scale, integrating satellite data with ground-based sensor measurements to retrieve essential biophysical variables of crops. This approach allows for localized analysis of key factors such as plant health, growth stages, and stress conditions. The predictive power of the ML and DL models used will be carefully evaluated, assessing their accuracy and robustness in estimating these variables from the combined data sources. This integration provides a more detailed understanding of the factors influencing crop performance and enables targeted interventions to optimize resource use and improve yield. In the second phase, the focus will shift to a broader agro-industrial scale, where a specific crop will be mapped and analyzed using only satellite imagery. The objective is to assess the feasibility of remote sensing technologies in providing information on extensive agricultural areas without the need for ground-based data after an accurate mapping process. This phase will evaluate the capacity of satellite data alone to deliver detailed information on crop distribution, growth stages, and health status, critical for large-scale monitoring and management. Finally, the research will expand to an even larger geographical scale, demonstrating how vast and heterogeneous regions can be effectively monitored. This phase will showcase the ability to map extensive agricultural landscapes with significant environmental variability, providing a framework for implementing wide-scale monitoring systems. The study will explore how these systems can be leveraged for agricultural management decisions, enabling large-scale assessments of processes and addressing challenges related to climate variability and food security. A key aspect of this research is the focus on the algorithms employed. Special attention will be given to the explainability of the ML and DL models, ensuring transparency in how these models derive predictions from input data. In this way, this study aims to indirectly identify the fundamental mechanisms of crop growth and the biophysical processes that drive agricultural production. By making the models more interpretable, the research will contribute to a better understanding of how specific variables influence crop health and yield, thereby enhancing the reliability and applicability of AI-driven solutions in precision agriculture. Ultimately, this research aims to advance the field of precision agriculture by demonstrating how AI technologies, combined with remote sensing and ground-based data, can provide scalable, accurate, and explainable solutions for both localized and large-scale agricultural challenges.

Downloads

Actions (login required)

Modifica documento