IMOLA – Intelligent Map recOgnition LAb

NSF-funded project building a national-scale historical road network database (US1884+) from USGS topographic maps

Sponsor: National Science Foundation (NSF)
Co-PIs: Dr. Yao-Yi Chiang (UMN), Dr. Stefan Leyk (CU Boulder)
Status: Ongoing (2024 – 2027)

Overview

IMOLA (Intelligent Map recOgnition LAb) is a large-scale NSF project focused on extracting, digitizing, and analyzing historical U.S. road networks from 178,000+ USGS topographic map sheets (1884–2006).
The core goal is to build US1884+, a comprehensive spatiotemporal database of attributed road networks with uncertainty measures, enabling research across geography, history, urban development, demography, and environmental change.

By converting scanned maps into GIS-ready vector and raster layers, IMOLA unlocks historical landscape information that predates satellite imagery, supporting long-term studies of urbanization, infrastructure development, and population change.

Main Research Contributions

  • US1884+ historical road network database
    A national, multi-temporal dataset of vector and raster road layers with feature-level uncertainty for 120+ years of U.S. history.
  • DaVinci map processing system
    Development of advanced computer vision capabilities (LCA, LDTR, contextual learning, multimodal knowledge fusion) for automatic line extraction with reduced manual annotation requirements.
  • Uncertainty modeling framework
    Processing-, production-, and application-related uncertainty analysis for trustworthy spatiotemporal inference.
  • Case studies & community integration
    Demonstrations connecting historical road networks with population change, land-use, settlement expansion, flooding risk, and environmental exposure research.

Technical Approach

IMOLA integrates deep learning, contextual reasoning, and geospatial knowledge graphs:

  • Automatic training data generation using Label Correction Algorithm (LCA) for weakly supervised learning.
  • LDTR (Linear Detection Transformer) for high-connectivity line extraction and vector graph reconstruction.
  • Prior-knowledge fusion through context-aware multimodal learning and semantic spatial rules.
  • Scalable processing and dissemination through the DaVinci pipeline, Docker distribution, open data portals, and GIS tutorials.

All outputs will be publicly accessible through the project website, Harvard Dataverse, and UMN U-Spatial.

Impact

IMOLA enables the first nationwide historical road network time series spanning 100+ years, supporting:

  • Urban growth and transportation evolution studies
  • Demographic change modeling and historical settlement analysis
  • Landscape & environmental research, hazards, and climate studies
  • Long-term spatial data integration with census, land cover, and building datasets

The project lowers barriers to using historical cartographic archives and establishes foundational infrastructure for large-scale spatiotemporal research across social, behavioral, environmental, and geospatial sciences.