Transportation datasets

NOAA International Best Track Archive (IBTrACS) data for Atlantic basin tropical storms and hurricanes. Contains storm tracks, intensities, and landfall information from 1851 to present.

Consumer Price Index for transportation, seasonally adjusted, monthly, 1947-present. Base period 1982-84=100.

FAA Wildlife Strike Database. Aircraft-wildlife collisions at US airports from 1990-2018. Contains species identification, damage assessment, airport location, and flight phase information.

Bureau of Transportation Statistics airline on-time performance data including flight delays, cancellations, and causes. Monthly data for domestic flights reported by major U.S. carriers under 14 CFR 234. Covers January 2003 through the most recently published month (~3 month lag).

Computer Aided Dispatch (CAD) incident records from the Austin Police Department, January 2017 to present with continuous updates. Nearly 4 million records including response times, priority levels, mental health flags, unit counts, and disposition. Covers all dispatched incidents across APD sectors and council districts.

Traffic crash records from the TxDOT CRIS (Crash Records Information System) database, rolling 10-year window from 2010 to present. Approximately 228,000 records with fatality and injury counts broken out by mode (pedestrian, cyclist, motorcycle, motor vehicle). Includes crash severity, speed limit, and location data. This is the Vision Zero dataset.

Philadelphia parking violation data from 2017. Contains ticket details including violation type, fine amount, location, and issuing agency. Great for urban data analysis.

Real-time traffic incident reports from the City of Austin, sourced from Austin-Travis County public safety agencies via the Combined Transportation, Emergency, and Communications Center (CTECC). Includes crashes, stalled vehicles, hazards, and other traffic events with geolocation, status, and responding agency.

Tour de France winners from 1903 to 2019. Contains winner name, nationality, team, time, and stage wins for each edition. Redistributed by the TidyTuesday project (R4DS Online Learning Community), which compiled it from Alastair Rushworth's `tdf` R data package.

Airline safety statistics comparing incidents and fatalities across two time periods (1985-1999 vs 2000-2014)

Bikeshare trip data from Austin MetroBike, 2018 to present. Approximately 2.3 million trips with checkout and return kiosks, trip duration, bike type (electric vs standard), and membership type. Shows ridership patterns, popular stations, and seasonality trends.

311 service requests from six major California cities - Los Angeles, San Francisco, San Diego, Oakland, Sacramento and San Jose - normalized to a common schema with a city column. Each row is one service request: city, category, status, submission channel, location, and open/close timestamps. Coverage differs by city (SF from 2008, Oakland from 2013, San Diego from 2016, Sacramento 2016 to April 2020, San Jose from 2017, Los Angeles 2025), so compare rates within a city over time rather than counts across cities. Built from the six per-city datasets listed under inputs; see each for its own cadence and caveats.

California DMV counts of registered vehicles broken out by ZIP code, model year, fuel type, make, and duty class. Eight annual snapshot files (each labeled by its "as of" vintage year) from data.ca.gov, unioned into one table with a `vintage` column marking which DMV release each row came from. The source `date` column is NOT the vintage label - it is the DMV snapshot date embedded in the file, which is off by up to a year from the filename/vintage (the 2023 vintage's `date` reads 12/31/2022; the 2019 vintage's `date` reads 10/1/2018). Always filter or group by `vintage`, not `date`, for year-over-year analysis. `model_year` includes a rolling "less-than" sentinel bucket (e.g. `<2012` in the 2026 vintage, `<2006` in the 2019 vintage) grouping all older vehicles, so it is not a stable category across vintages - only years at or above the newest vintage's sentinel cutoff are comparable across all eight files. Total statewide row counts drop from roughly 767K (2023 vintage) to roughly 504K (2026 vintage). This is a change in how DMV aggregates the file, not a decline in the vehicle fleet: the 2023 vintage has 767,054 rows summing to 31,113,830 vehicles while the 2024 vintage has 499,623 rows summing to 31,127,451 vehicles - a stable statewide total with 35% fewer rows. The exact mechanism is not documented by DMV and is not small-cell suppression (the minimum `vehicles` value is still 1 in both vintages, and make/model-year/ZIP cardinalities barely moved). Because the denominator's composition changed, be careful comparing row-level granularity across the 2023/2024 boundary. `zip_code` includes the non-ZIP sentinel "OOS" (out of state).

MyLA311 service requests submitted to the City of Los Angeles in 2025, covering everything from bulky item pickups and illegal dumping to graffiti removal, streetlight outages, and homeless encampment reports. Each record has the request type, responsible department (owner), status, submission channel, address, ZIP code, coordinates, and City Council district. The City of Los Angeles publishes a separate MyLA311 dataset for each calendar year back to 2015; this dataset tracks the current (2025) year, which updates daily.

EIA Form 860 annual generator-level inventory for U.S. utility-scale power plants (1 MW+ combined nameplate capacity), from the PUDL raw EIA-860 archive on Zenodo. Covers currently operating generators, with the year each came online going back to the early 1900s, plus planned retirement dates where reported. Unlike the EIA-860M monthly snapshot, this dataset reflects the annual Form 860 filing and preserves each generator's full operating history in a single year-online column, making it usable to chart capacity buildout over time by state, fuel, or technology. Includes six annual vintages (2014, 2016, 2018, 2020, 2022, 2024) so the same fleet can be tracked as it evolves. A report_year column marks which filing year each snapshot came from, alongside online_year for when each generator itself came online.

EIA-860M preliminary monthly generator inventory for U.S. utility-scale power plants (1 MW+ combined nameplate capacity). Includes operating and planned generators with capacity, fuel type, technology, state, and expected online/retirement dates. Single monthly snapshot (latest available month).

Total federal investment outlays for major public physical capital, research and development, and education and training from 1962 to 2027, from OMB Historical Table 9.1. Defense and nondefense breakdowns. Values in millions of current US dollars.

Utility plant-in-service and capital additions/retirements per utility per year, from FERC Form 1 Schedule 204 as cleaned by the PUDL project (Catalyst Cooperative). Covers ~1994 to present. Includes starting and ending balances, additions, retirements, adjustments, and transfers by FERC account category.

Heavy weight (Class 8) truck retail sales (HTRUCKSSAAR). Monthly data from 1967. Seasonally adjusted annual rate in millions of units. Logistics carriers curtail fleet expansions immediately when demand forecasts weaken.

Georeferenced inventory of existing US data center facilities from the Department of Energy's IM3 project at Pacific Northwest National Laboratory. Facility locations are derived from OpenStreetMap and enriched with US Census Bureau county and state boundaries (2024), operator names, and building footprint area. Coordinates are WGS84 (EPSG:4326). Each row is one facility record in one of three spatial layers, given by the type column: campus (the full site boundary), building (an individual data center structure), or point (a location known only as a single point, with no footprint). The layers deliberately overlap. A single campus and each of its constituent buildings are all present as separate rows, so summing sqft or counting rows across all three layers double counts floor area and sites. Use the buildings, campuses, or points views to work within a single layer, or filter on type directly. A facility straddling a county line also appears once per county with the same id, so count distinct id rather than rows for a facility count. Because the underlying source is crowd-sourced, coverage is incomplete and skewed toward well-documented operators. The operator field is blank for about 35% of rows and name for about 19%. This is a location and floor-area inventory, not a power inventory. There is no megawatt or energy field. For projected future buildout including IT power capacity, see im3/projected-data-center-locations.