Frederick County Fire & Rescue Incidents Dashboard

A real-time incident tracking and analytics dashboard for Frederick County, Maryland fire and rescue calls. The project scrapes incident data from frederickscanner.com and publishes an interactive dashboard and RSS feed.

Live dashboard: newsappsumd.github.io/fredscanner

About

This project is produced by the Philip Merrill College of Journalism at the University of Maryland and funded by a grant from the Scripps Howard Foundation.

How It Works

1. Scraping (fredscanner.py) — Fetches the latest incident data from frederickscanner.com every 30 minutes via GitHub Actions. New incidents are appended to incidents.csv.
2. Threading (threads.py) — The source re-posts an incident every time its dispatch changes (units added, event reclassified). This groups those re-posts into a single incident thread — chaining posts at the same location within a 60-minute window when their event types are compatible (e.g. a fire alarm escalating to a building fire), while never merging administrative posts (station transfers) or posts at station/training-facility addresses. make_rss.py and the dashboard both use it, so counts and feed entries reflect incidents, not individual dispatch posts.
3. Geocoding (geocode.py) — Dispatch locations are CAD strings, not mailing addresses (“5500 BLOCK UPSHUR SQ”), so this cleans them up and geocodes them via the free Census Bureau Geocoding Services API, caching results in geocode_cache.csv. A location is only ever geocoded once, whether it matches or not, so an unmappable address isn’t retried forever. Highway locations given only as a mile marker (“I70EB / 38MM”) have no street to look up and are skipped — those show up in the Highway Hotspots panel instead. Historical locations are geocoded in one batch via a manually-triggered backfill (.github/workflows/geocode-backfill.yaml); newly-seen locations are geocoded incrementally as part of the regular 30-minute run. Because there’s no per-location city to disambiguate the query, the geocoder occasionally matches a same-named street in a different county entirely; any match falling outside Frederick County plus a buffer for its mutual-aid neighbors is rejected as unreliable rather than plotted (REGION_BOUNDS in geocode.py). Run python geocode.py --revalidate to re-check an existing cache against the bounds without any new API calls.
4. RSS Feeds (make_rss.py) — Generates a feed of the 50 most recent individual posts (site/latest.rss) and a filtered feed of large responses, working fires, and rare incident types, one entry per incident thread (site/priority.rss). A growing incident updates its existing feed entry instead of appearing as a new one. Also writes the CSV into the site/ directory with added ThreadID and Lat/Lon columns.
5. Dashboard (site/index.html) — A static HTML page deployed to GitHub Pages that loads and visualizes the incident data client-side, grouping posts into threads the same way.

Dashboard Features

• Search by event type, location, or responding units
• Filter by event type and date range
• Key metrics: total incidents, average per day, most common event type — counted per incident thread, not per dispatch post
• 7-day and 30-day percentage change indicators
• Interactive line chart of incidents over time
• Monthly Trends — a full-history bar chart and table of incidents per month (count, average per day, most common event, change vs. prior month), independent of the date-range/search filters
• Incident map — geocoded incidents for the current filters, plotted with Leaflet and clustered by zoom level; markers show event, time, location, and units on click
• Frequent Locations leaderboard — top locations by incident count for the current filters, click-to-filter
• Highway Hotspots — vehicle accidents on interstates and state routes, grouped by highway and mile marker (or nearest cross street), ranked by crash count, click-to-filter
• Incident records table with repeat dispatches grouped into one row; click a row with an updates badge to expand each individual post
• RSS feed for the latest individual posts, plus a filtered priority-alerts feed (one entry per incident thread) for large responses, working fires, and rare incident types

Data

The incident data is stored in incidents.csv with the following columns:

Column	Description
Time	Time of the incident (e.g., “1:17 pm”)
Event	Incident type (e.g., “HOUSE FIRE”, “VEHICLE ACCIDENT - BLS”)
Location	Address or block location
Units	Responding units and radio information
Date	Date in MM/DD/YYYY format
Datetime	ISO-formatted datetime (YYYY-MM-DD HH:MM:SS)

The raw incidents.csv in the repo is never modified beyond appending new posts. The published copy at site/incidents.csv adds ThreadID and Lat/Lon columns so the dashboard can group posts into incidents and plot them on the map; rows sharing the same ThreadID are re-posts of the same incident.

Geocoding results are cached in geocode_cache.csv (Location, Lat, Lon, Status), keyed by the exact raw Location string. Status is Match, No Match, Skipped (mile-marker-only highway locations that were never sent to the geocoder), or Out of Region (the geocoder returned a match, but it fell outside Frederick County and its mutual-aid neighbors, so it’s almost certainly a same-named-street mismatch and isn’t trusted).

Setup

Requirements

• Python 3.x
• Dependencies: pip install -r requirements.txt

Running Locally

# Run the test suite
python test_threads.py
python test_geocode.py

# Scrape latest incidents
python fredscanner.py

# Geocode any newly-seen locations
python geocode.py

# Generate RSS feeds and prepare site data
python make_rss.py

The dashboard can be served from the site/ directory using any static file server.

Automation

The GitHub Actions workflow (.github/workflows/scrape.yaml) runs the tests, scraper, geocoder, and RSS generator every 30 minutes and deploys the updated site to GitHub Pages.

.github/workflows/geocode-backfill.yaml is a manually-triggered (workflow_dispatch) one-off job that geocodes the full history of locations at once via the Census batch endpoint. Run it once to populate geocode_cache.csv from an empty or missing cache; routine incremental geocoding of new locations happens automatically in the regular workflow above.

License

MIT License. See LICENSE for details.