Street View Seismic Risk Mapping Project

ISEF Category: Engineering Technology: Statics and Dynamics

Subcategory: Civil Engineering  ·  Difficulty: Advanced  ·  Setup: Home Setup  ·  Time: 1 to 2 Months

The Hook

A city block can look sturdy from the street and still hide a lot of risk. Earthquakes do not just hit buildings, they expose weak shapes, bad spacing, and aging materials. You can test whether a computer model can spot those warning signs from public images and map data. That turns your project into a real screening tool, not just a data exercise.

What Is It?

This project asks a simple question with a hard answer: can you estimate how earthquake-vulnerable a block is without walking inside every building? You use public street imagery and building footprint data from OpenStreetMap, then turn visible clues into a score. Think of it like a weather app for structures. The app does not measure the storm itself, but it predicts where damage is more likely.

Your model might look for clues such as building shape, height, spacing, roof style, and signs of older construction. These features matter because buildings fail in different ways during shaking. A tall, narrow building behaves differently from a short, boxy one. Your job is to turn those visual differences into numbers, then see whether your score lines up with published HAZUS estimates and with what you can verify in one neighborhood.

Why This Is a Good Topic

This is a strong science fair topic because you can test a real engineering idea with public data and clear outcomes. You are not guessing, you are building a model, comparing it to a known framework, and checking whether it works in one place. The project connects to earthquake safety, urban planning, and disaster response. You can learn image analysis, feature engineering, validation, and error checking, all in one project.

Research Questions

• How does a Street View based vulnerability score compare with published HAZUS block estimates?
• What is the effect of adding building footprint shape features on model accuracy?
• Does including visible roof and facade cues improve prediction of seismic vulnerability?
• To what extent does the model agree with ground-truth observations from a local neighborhood survey?
• Which image features contribute most to the final vulnerability index?
• How does model performance change when you train on one district and test on a different district?

Basic Materials

• Laptop with enough storage for image files and map data.
• Internet access for Google Street View, OpenStreetMap, and USGS or HAZUS-related public data.
• Spreadsheet software for organizing blocks, labels, and measurements.
• Python installed with data analysis and image processing libraries.
• Simple notebook for field notes and ground-truth observations.
• Smartphone camera for documenting local building features during a neighborhood walk.
• Measuring wheel or distance app for rough block-level checks if needed.
• Access to a local map or GIS viewer for comparing footprints and parcels.

Advanced Materials

• Computer with a GPU for faster image feature extraction.
• Python environment with geospatial and machine learning libraries.
• GIS software for footprint cleaning, buffering, and map matching.
• Access to a university library or public safety report archive for earthquake-related building references.
• High-resolution aerial or parcel data if your city provides it.
• Image annotation tool for labeling structural features in Street View images.
• Statistical software for agreement testing, error analysis, and model comparison.
• Field survey sheet for structured neighborhood ground truth collection.

Software & Tools

• Python: Cleans data, extracts features, and trains your prediction model.
• Google Earth Pro: Helps you inspect blocks and compare viewpoints with map data.
• QGIS: Matches building footprints to blocks and manages geospatial layers.
• ImageJ: Measures visual features in images when you need simple image-based metrics.
• GeoPandas: Handles footprints, boundaries, and other spatial data in Python.

Experiment Steps

1. Define the exact vulnerability score you want to predict and the block-level unit you will study.
2. Build a clean dataset that pairs each block with Street View images, footprint geometry, and any reference score you can get.
3. Choose the visible building features you will measure, then decide which ones you can score by hand and which ones need automation.
4. Create a baseline model first, then add more features and compare whether prediction improves.
5. Plan a validation check against HAZUS outputs and a small ground-truth survey in one neighborhood.
6. Set up an error analysis plan so you can explain which blocks the model gets wrong and why.

Common Pitfalls

• Mixing data from different years, which can make the Street View image and the building footprint describe different buildings.
• Using block boundaries that do not match the HAZUS comparison unit, which makes the validation unfair.
• Letting the model learn neighborhood wealth instead of structural risk, which creates a misleading score.
• Recording footprints with missing or merged buildings, which hides the shape cues that drive the prediction.
• Comparing images with different camera angles or lighting, which changes the visible building features and lowers consistency.

What Makes This Competitive

A stronger version of this project does more than build a map. It explains why the model works, where it fails, and which features carry real predictive power. You can raise the level by comparing at least two modeling approaches, using careful spatial validation, and testing whether the method still works in a new neighborhood. Clear error analysis and honest uncertainty estimates make the project much stronger.

Project Variations

• Use only one neighborhood type, such as older residential blocks, and test whether the model still separates higher and lower risk buildings.
• Replace hand-scored features with a computer vision model that classifies visible structural clues from Street View images.
• Compare street-image based scores with another hazard layer, such as flood exposure or fire spread potential, to see whether the same buildings stay high risk across hazards.

Learn More

• USGS HAZUS resources: Search the USGS site for HAZUS earthquake loss estimation materials and technical summaries.
• FEMA HAZUS technical documentation: Search FEMA for HAZUS user and technical manuals that explain building damage models.
• OpenStreetMap: Use the OpenStreetMap wiki and map data to learn how building footprints are stored and edited.
• NOAA Digital Coast: Search for mapping and GIS guidance that helps with coastal and urban spatial analysis.
• MIT OpenCourseWare, Introduction to GIS: Find free course materials that explain spatial data, map layers, and geoprocessing.
• PubMed: Search for review articles on post-earthquake building damage assessment and image-based structural analysis.