Integrating Deep Learning and Satellite Imagery for Biodiversity Assessment

by Yanfei Shan

The Environmentally-Extended Multiregional Input-Output (EE-MRIO) database is a mainstream tool for effectively modeling global supply chains and their environmental impacts. It can scientifically address complex questions such as: “How much biodiversity loss is caused by land conversion in Brazil due to beef demand in Germany?” or “How much global water stress may result from crop cultivation?” However, these datasets face intrinsic limitations, such as restricted spatial resolution, data discontinuity, data quality deficiencies in the Global South, and outdated temporal trends. In contrast, satellite data offers a promising solution to these challenges, with its extensive spatio-temporal coverage and continuous monitoring capabilities.

Motivated by these challenges and research gaps, my PhD research focuses on improving data reliability and accuracy in assessing the sustainability of global agricultural systems by integrating remote sensing, climate data, and multi-regional input-output (MRIO) models. I am developing machine learning and multimodal deep learning approaches to quantify global agricultural water stress at high temporal and spatial resolutions. In parallel, I am exploring foundational methods to integrate large language models (LLMs) with EE-MRIO analysis, aiming to enable novel AI-driven insights in sustainability science.