the Spatial AI Challenge

Artificial intelligence (AI) is having its moment and spatial AI is right there at its heels. The terms “Spatial AI” and its geospatial counterpart, “GeoAI,” have gained increasing attention for the past few years. However, the core principles of utilizing location and spatial relations have been employed in computation and data analytics by a variety of fields for decades. Broadly speaking, spatial AI involves the development and use of AI to analyze location-based information, recognize spatial patterns, and make decisions in real-world or virtual spaces by leveraging spatial data and inference. The types of problems that are best addressed by spatial AI include, but are not limited to, augmented reality, autonomous navigation, environmental monitoring, robotics, and urban planning.

In October 2024, I-GUIDE launched its inaugural Spatial AI Challenge to develop innovative, responsible, and reproducible solutions focused on real-world issues such as disaster response, urban planning, or food and water security. By design,

The 2025 I-GUIDE Forum will focus on Geospatial AI and Innovation for Sustainability Solutions. Our three-day Forum, consisting of workshops and tutorials, presentations, posters, and discussions will be co-located with the Sustainability Research & Innovation Congress (https://sricongress.org), in downtown Chicago. This coordination will allow participants in the I-GUIDE Forum to take advantage of the momentum, energy, and synergies of several thousand people gathering to work on solutions to our most significant sustainability challenges! For one single registration you will be able to access both the Forum and the SRI Congress.

The sessions during the I-GUIDE Forum 2025 will focus on (1) Frontiers in Geospatial and Sustainability Sciences; (2) Geospatial AI and Data Science for Sustainability Solutions; (3) Innovation of CyberGIS and Cyberinfrastructure; and (4) Innovative Education, Training, and Workforce Development. The Call for Participation is now open! All information on submitting short papers or abstracts for presentations, posters, and workshops or tutorials can be found here. The final day to submit your work will be Monday March 24, 2025.

The 2025 Summer School – with a focus on Spatial AI for Extreme Events and Disaster Resilience – will take place August 4-8, 2025 at the University Corporation for Atmospheric Research (UCAR) Center Green campus in Boulder, Colorado.

The call for Team Leaders and Project Proposals is now open! We invite established scholars to apply as team leaders and invite anyone to suggest project topics, datasets, or methods for the Summer School program. If your research focuses on spatial AI and extreme events and disaster resilience, and you want a team of graduate students and other early career scholars to help you for a week, apply to lead a team! The deadline to apply to lead a team is Monday March 3, 2025.

The I-GUIDE Insider is a weekly digest of upcoming events and opportunities from the I-GUIDE project as well as recent publications and news. Sign up for the Insider to stay up to date with the I-GUIDE project!

Presenter

Alex Michels · University of Illinois

Explore OpenStreetMap (OSM) data using the I-GUIDE Platform. We will discuss some of the use-cases of OSM data, get hands-on experience with Python packages designed for OSM data, and walk through some example analyses.

Presenter

Song Gao · University of Wisconsin

** 2024-25 UCGIS Community Champion **

Spatial networks are essential for modeling geographic phenomena influenced by spatial interactions. This presentation will describe region2vec, a family of GeoAI-enhanced unsupervised community detection methods using graph neural networks (GNNs). Region2vec generates node embeddings based on attribute similarity, geographic adjacency, and spatial interactions, then applies agglomerative clustering to extract spatial communities. These methods optimize both node attribute similarity and spatial interaction intensity, making them valuable for applications like health service area delineation, ecological zoning, transportation planning, and political redistricting. The developed models will be hosted on the I-GUIDE platform for interdisciplinary use.

Presenter

Nattapon (Nathan) Jaroenchai · University of Illinois

This VCO introduces the concept of transfer learning from ImageNet to geospatial analysis. In our session, we explore how a model that has been pretrained on the vast ImageNet dataset can be effectively adapted to solve real-world geospatial problems, such as urban land cover classification. Participants will gain insight into the methodology of transfer learning, including the process of fine-tuning a powerful convolutional neural network to work with satellite and aerial imagery. We will compare a model trained from scratch with one that utilizes the pretrained features from ImageNet, highlighting the advantages in performance and efficiency.

Presenter

Yingjie Hu · University at Buffalo

** 2024-25 UCGIS Community Champion **

During natural disasters, people share urgent information on social media and chat groups, often including critical location descriptions. Accurately extracting these location descriptions can help responders reach victims faster. Traditional named entity recognition (NER) struggles with complex location descriptions, but recent geo-knowledge-guided GPT models have shown promise. In this presentation I will describe my Community Champion project and its contributions to the Spatial AI Challenge with its refined dataset of 1,000 Hurricane Harvey tweets with labeled location descriptions. A Jupyter Notebook with example code for processing the dataset and integrating geo-knowledge with OpenAI’s GPT API will be presented. Researchers can use this resource to refine prompts, improve location extraction methods, or experiment with open-source LLMs.

Groundwater depletion in the Western U.S., driven by unsustainable agriculture and climate change, threatens water security. Policies limiting irrigation may shift crop production globally, worsening deforestation and biodiversity loss. The National Science Foundation (NSF) recently highlighted the work of I-GUIDE researchers Thomas Hertel and Iman Haqiqi, both of Purdue University, as they study these complexities and warn of increased environmental pressures as global demand rises, particularly from South Asia and China.

Each year, I-GUIDE partners with the University Consortium for Geographic Information Science (UCGIS) via a Community Champions program, designed to expand the reach of I-GUIDE and support exchanges of knowledge and expertise for the greater GIS community. This year, the Community Champions are contributing their AI-related expertise towards our Spatial AI Challenge. The four individuals selected this year include:

• Seda Şalap-Ayça, Brown University: Rethinking Spatial Composite Indicators with the Lens of Machine Learning
• Song Gao, University of Wisconsin: GeoAI-enhanced community detection on spatial networks using graph neural embeddings
• Yingjie Hu, University at Buffalo: Extracting location descriptions from disaster-related messages using geo-knowledge-guided GPT models
• Xinyue Ye, Texas A& M University: Digital Twin Framework with Spatiotemporal Vision Transformers for Heat Resilience

Their data, models, and/or Jupyter Notebooks will be added to the I-GUIDE Platform this winter. These materials, and additional mentoring and consulting, will serve as helpful examples for the people participating in the Challenge. Additionally these UCGIS Community Champions will offer Virtual Consulting Office sessions or other online presentations to discuss their projects and offer guidance to those who are just beginning to work with Spatial AI.

The latest cutting-edge research and publications

from the NSF I-GUIDE Project Team

Planners managing coastal flood risk must balance computational constraints with simulation accuracy. This study introduces a deep learning model predicting storm surge based on storm parameters, landscape features, and boundary conditions. Trained on ADCIRC simulations for Louisiana (2020–2070), it achieved 0.086-m RMSE and 0.050-m MAE across 90 storms per landscape. A Kolmogorov-Smirnov test showed only a 1.1% rejection rate, confirming strong agreement with ADCIRC annual exceedance probability estimates.

Explore more of the latest work from the NSF I-GUIDE Team.

Click on the cards to view the full articles.