Uncategorized

We are excited to share that our new paper, “Spatial CoT: A Spatial Concept Transformation Guided LLM Reasoning Framework for Complex Geospatial Question Answering”, led by SEAI Lab PhD student Zeping Liu, has been published in the International Journal of Geographical Information Science (IJGIS)!

Inspired by Prof. Werner Kuhn’s core concepts of spatial information, this work asks a simple question: can classic GIScience theory directly guide how large language models reason about complex geospatial problems? Spatial CoT makes the LLM identify the spatial core concepts involved in a question and construct a concept transformation graph as a blueprint before answering. The framework further introduces a spatial concept transformation knowledge graph and a propose–retrieve–refine retrieval-augmented pipeline to refine the reasoning path.

Across multiple geospatial benchmarks, Spatial CoT consistently improves reasoning performance and reduces geospatial hallucination, with especially large gains for smaller open-source models.

Much of this work was completed during Zeping’s internship with the Spatial Statistics team at Esri. It is a collaboration among researchers from UT Austin, Esri, Utrecht University, and Emory University. Congratulations to Zeping and all collaborators!

📄 Read the paper: https://doi.org/10.1080/13658816.2026.2696352

SEAI Lab PhD student Zeping Liu has published a new paper, “GAIR: Location-aware self-supervised contrastive pre-training with geo-aligned implicit representations,” in the ISPRS Journal of Photogrammetry and Remote Sensing.

This work introduces a novel multimodal geospatial representation learning framework that aligns remote sensing imagery, street-view images, and geographic location information—three modalities with fundamentally different characteristics and spatial scales—into a unified embedding space. By learning geo-aligned representations across these modalities, the framework enables more effective cross-modal understanding and transfer for a wide range of geospatial AI applications.

Congratulations to Zeping on this exciting achievement!

At the AAG 2026 Annual Meeting, SEAI Lab members received multiple prestigious awards. Among them, Dr. Jielu Zhang was honored with the William L. Garrison Award for Best Dissertation in Computational Geography. Notably, our lab director, Prof. Gengchen Mai, is also a past recipient of this distinguished award. Congratulations to Jielu on this outstanding achievement!

In addition, second-year Ph.D. student Zeping Liu received the IGIF Scholarship Award. Congratulations to him as well!

Building on this success, SEAI Lab members look forward to another productive year of impactful research and continued excellence.

🚀 Excited to share that our paper “LocDiff: Identifying Locations on Earth by Diffusing in the Hilbert Space” led by Dr. Zhangyu Wang has been accepted at NeurIPS 2025!

LocDiff introduces a new multi-scale latent diffusion framework for image geolocalization, achieving state-of-the-art accuracy and strong generalization to unseen regions. 🌍📍

Congratulations to Zhangyu and all collaborators!