I am Yifan Yang, a third-year Ph.D. student in the Department of Geography at Texas A&M University, advised by Dr. Lei Zou in the Geospatial Exploration and Resolution (GEAR) Lab. My research builds autonomous and accountable GeoAI: systems that reason about places from incomplete, conflicting, and changing observations, transfer across regions, and know when the evidence is not enough. Disasters are my current testbed, where these conditions are most extreme. Beyond my research, I founded AutonomousGeoAI4Science, an open community hub for papers, tools, benchmarks, and collaboration. My doctoral committee members are Dr. Heng Cai, Dr. Andrew Klein, and Dr. Zhengzhong Tu.

🔭 Vision: To transcend the boundaries of screens and make the real world our playground of intelligence — where AI, space, and humanity coexist and co-create.

Before starting my Ph.D. studies, I earned a master’s degree in Spatial Data Science from the University of Southern California. During my master’s, I worked with Dr. John P. Wilson on the Urban Trees Initiative, where I explored geospatial approaches to urban sustainability. Under the mentorship of Dr. Siqin Wang, I also completed a research internship at the Spatial Data Lab, Harvard University, which further strengthened my expertise in Geospatial AI and interdisciplinary data science.

Prior to graduate school, I received a bachelor’s degree in Software Engineering from Hainan University in China. I’m also an AI4Science enthusiast, and my hobbies are basketball🏀, football⚽, golf🏌️‍♀️, poker🎴, and talk shows🎆.

📄 Curriculum Vitae: View CV (last updated: October 19, 2026)

Research interests:

My current research interests mainly lie in Geospatial AI, especially in:

• Spatiotemporal Complementarity: Any single view, time slice, or model observes a place only partially. The pre-disaster baseline, the overhead view, and the ground-level view carry evidence that cannot substitute for one another, and disagreement between them is information, not noise. How can a model restore the missing baseline, arbitrate conflicting observations, and turn partial evidence into a defensible judgment?
• Place Specificity vs. Transferability: Disasters are geographically specific, yet response demands methods that are general. Task-specific models win on specificity; generative foundation models win on zero-shot transfer across hazards and regions, and no single model achieves both. How should this trade-off be managed at the level of the framework, and how do we evaluate semantic correctness rather than mere perceptual realism?
• Accountability as a Constitutive Condition of GeoAI: Once GeoAI moves from emitting labels to recommending decisions, evidence lineage, process validity, and the ability to refuse when evidence is insufficient are no longer add-ons but preconditions for the output to count as geographic knowledge. How do we build agentic GeoAI that anticipates, inspects, and communicates place-based impacts, and remains auditable while doing so?

Keywords: