Causal Artificial Intelligence for Sustainable Agriculture

Agriculture is a critical sector for ensuring food security and sustainable development. However, modern agriculture is facing significant challenges given the changing climate and its consequences. A path for addressing or adapting to these challenges is the development of innovative solutions, that blend multiple sources of earth data and domain knowledge through Artificial Intelligence (AI), to provide actionable insights or advice for farmers and policymakers. Today, there is limited adoption of all these precise and climate-smart technologies, as we lack evidence of their efficiency in practice, but also due to the skepticism about the tangible environmental and mainly economical effect of following agricultural practices that information systems/AI proposes and not a human expert. This PhD project aims to explore how AI can serve sustainable agriculture by combining domain-knowledge with data-driven approaches using space-borne remote sensing and other heterogeneous data sources. Specifically, the research will focus on designing AI algorithms that not only make accurate predictions blending data of different sources but also: (a) adapt & generalize to different environments (to address out-of-distribution and data drift issues) and on using proper tools to (b) evaluate & quantify the effectiveness of AI-powered decisions. Through the aforementioned, we aim to increase the trust of the agricultural world in AI directly and indirectly, providing evidence-based results about the effectiveness of AI-powered decisions and increasing the reliability of AI models behavior in different unseen conditions, respectively. Supplementary to the above, the research will explore the integration of satellite imagery and remote sensing data with other heterogeneous data sources, such as weather data, soil data, and in-situ data to capture a comprehensive representation of the agriculture system. The proposed research will aim to support decision-making at various levels, from individual farmers to regional policymakers, and with various artifacts, from modeling to impact assessment.