Summer School 2026 Projects : I-GUIDE

Project Description: Volunteered Geographic Information (VGI) maps (e.g., Open Street Map) are widely used for urban analytics, disaster response, and environmental applications. However, its quality is uneven: while highly accurate in developed urban areas, VGI often suffers from incompleteness and positional errors in rural regions and the Global South due to limited contributions and expert effort in calibration. This spatial bias can introduce uncertainty into downstream analysis, particularly in data-sparse regions. This project aims to develop a systematic approach to evaluate and calibrate VGI maps using multimodal remote sensing data (e.g., Landsat satellite remote sensing imagery and LiDAR remote sensing data). The key research questions are: 1) How does VGI accuracy and completeness vary across geographic and socioeconomic contexts? 2) Can remote sensing data detect discrepancies in VGI data such as roads and buildings? 3) How can we develop scalable (AI) approaches to automatically improve VGI quality in data-sparse regions? During the Summer School, the team will collaborate to design evaluation metrics and workflows, extract features from imagery and LiDAR, and develop models for detection and calibration. We will design a scalable workflow for benchmarking and enhancing VGI data quality across diverse geographic contexts.

Team Leader: Fangzheng Lyu, Department of Geography, Virginia Tech University

Project Description: The Everglades are a critical freshwater resource for South Florida, serving more than nine million people and replenishing the Biscayne Aquifer. Beyond water supply, they play key roles in flood control, storm protection, and sustaining a highly diverse ecosystem with thousands of species. As the largest subtropical wilderness in the United States, they also contribute significantly to the economy, generating about $31 billion annually. Existing flux-towers in the Everglades provide point measurements at high frequency, but their spatial coverage is limited, and scaling infrastructure is challenging due to cost and maintenance. On the other hand, remote sensing provides broader spatial coverage but at lower temporal resolution, involves indirect measurements, and requires calibration and validation. First-principle models provide an understanding of phenomena of interest but require parametrization and validation. In this summer project, we will explore an integration framework that can combine information from these three sources. The broad question we would like to answer is: how can in situ measurements from flux towers, remote sensing observations, and first-principles models be effectively integrated to improve the accuracy, spatial scalability, and predictive capability of carbon, water, and energy flux estimates in the Everglades? The Summer School team will collaborate on data integration and processing, machine learning, and model formulation and validation. That week we will focus on one specific tower, one variable of interest, a simplified model, and two remote sensing sources.

Team Leader: Leonardo Bobadilla, Knight Foundation School of Computing & Information Sciences, Florida International University

Project Description: Urban residents spend roughly 80% of their time indoors, yet traditional environmental assessments, such as satellite NDVI, measure greenness from a top-down perspective. This fails to capture the human-centric, window-level visual experience of urban greenery, which is crucial for mental well-being and mitigating environmental stressors like extreme heat. By leveraging Google Photorealistic 3D Tiles and semantic segmentation (DeepLabv3+), we can now quantify 3D window-level green visibility, revealing that vertical building position significantly shapes visual exposure to nature. This project scales this framework into a comparative urban digital twin to address the equity implications of indoor-out green views. We aim to answer two primary questions: (1) How does 3D window-level green visibility intersect with socioeconomic vulnerabilities across diverse urban morphologies? (2) How does this vertical access to greenery correlate with exposure to localized environmental hazards, specifically extreme urban heat? To achieve meaningful progress in the short time available and without excessive data wrangling, the multidisciplinary team will converge their distinct skill sets using a "micro-to-macro" approach. The team will collaborate to utilize the I-GUIDE Platform’s HPC to run and parallelize the spatial AI pipeline on a targeted neighborhood sample, leverage a pre-computed dataset of multiple cities to run geospatial equity models, and synthesize these outputs, translating the statistical spatial models into actionable policy insights for designing climate-resilient cities.

Project Description: Heat exposure is a growing concern in the United States, particularly in urban areas where extreme heat events are becoming more frequent. High temperatures contribute to a wide range of adverse outcomes, including heat-related health issues, stress on energy supply, and damages to infrastructure systems. Urban environments are especially vulnerable due to high building density, limited natural vegetation, and unequal distribution of cooling centers. As a result, an important challenge for research and planning is to characterize the heat exposure patterns. A number of existing assessments measure heat exposure based on environment features, which overlook the fact that people regularly move across various environments. This project adopts an activity-based perspective by integrating urban heat data with human mobility flows to estimate the individual-level heat exposure in an urban environment. Students will construct human mobility networks and capture the heat exposure patterns based on these mobility patterns. Spatial AI techniques will be used to analyze, interpret, and predict these patterns, as well as identify important environment features, such as green spaces. The output will illustrate how people move around different thermal conditions, offering a data-driven understanding of urban heat exposure through the lens of movement.

Team Leader: Yuqin Jiang, Department of Geography and Environment, University of Hawaii at Manoa

still in draft...

Wildfires are increasing in frequency and intensity, yet many prevention and mitigation decisions still rely on static administrative boundaries or ad hoc clustering of incidents. Meanwhile, public wildfire data streams (e.g., satellite active fire detections) provide abundant point observations but are noisy and difficult to translate into actionable plans. This project builds an AI-agent-centered planning system that uses an LLM as a “geospatial planner” to transform natural-language planning goals into auditable, executable spatial workflows and optimized mitigation districts. The central contribution is an agentic LLM workflow—not just a model that predicts risk. The LLM agent will (1) interpret stakeholder intent (coverage, capacity, “do not split boundaries,” equity priorities), (2) retrieve and summarize relevant geospatial layers and their limitations, (3) generate a structured optimization specification (objectives + constraints) for regionalization, (4) call tools to run regionalization and sensitivity experiments, and (5) produce a policy-facing prevention/mitigation memo that is grounded in the data and accompanied by provenance (what layers and parameters supported each recommendation). The wildfire case serves as a concrete, high-impact use case because active fire detections are naturally point-based and suitable for generating contiguous “planning districts” for staged resources and prevention investments.

During the week, students will...

Summer School 2026 Projects

Summer School 2026: Spatial AI and Convergence Science

University of Illinois Urbana-Champaign · July 13-17, 2026

Detecting and Correcting Spatial Bias in VGI Using Remote Sensing

Integrating Flux Towers, Remote Sensing, and Models in the Everglades

Seeing Green in 3D: Assessing Vertical Urban Space

Urban Heat Exposure Revealed by Human Mobility Patterns

Project #5 (Yunfan) - Agentic Wildfire Mitigation Planner: LLM-Guided Regionalization for Disaster Analysis, Planning, and Prevention

Project #6 (Eric Shook)