Live Environment Streams

A curated, agent-optimized repository of live outdoor webcam streams across the globe. Originally created to power twitch.tv/alwaysrainsomewhere.

5,997 streams · 98 countries · 67 sources · 4,226 verified active · 4,229 directly usable

Data quality

Not all streams are equal. The schema includes fields that tell you exactly what you're getting:

Field	What it tells you
url_type	Whether the URL is a direct HLS manifest, a YouTube link, or an HTML page requiring extraction
coordinates_quality	Whether coordinates are camera-level (exact), city-level (city), or a capital fallback (country_centroid)
status	active = verified working by the production pipeline; unverified = not yet probed
last_verified	When the stream last passed a live probe (ffprobe-based, from the production selector)
resolution	Probed video resolution where known (e.g. 1920x1080)
source_url_requires	Special handling: browser for HTML pages, token_refresh for expiring auth tokens, or null

Unlike most webcam lists, active status here is not a one-time HTTP 200 check — it reflects continuous validation by a production streaming system that actually plays these feeds 24/7.

Recommended query for agent pipelines — get only streams you can use immediately:

usable = [
    f for f in streams
    if f["properties"]["status"] == "active"
    and f["properties"]["url_type"] in ("hls", "youtube")
    and f["properties"]["source_url_requires"] is None
]
# ~2,929 verified streams with direct URLs

Usage

import json
import random

with open("streams.geojson", "r") as f:
    data = json.load(f)

streams = data["features"]
selected = random.choice(streams)
props = selected["properties"]
coords = selected["geometry"]["coordinates"]

print(f"Loading {props['name']} at {coords}")
print(f"URL: {props['url']} (type: {props['url_type']})")
print(f"Environment: {props['environment']} | Resolution: {props['resolution']}")
print(f"Source: {props['source_family']} | Status: {props['status']}")

Filter by environment or quality:

# Verified coastal streams with direct HLS
coastal_hls = [
    f for f in streams
    if f["properties"]["environment"] == "coastal"
    and f["properties"]["url_type"] == "hls"
    and f["properties"]["status"] == "active"
]

# All streams for a specific country
us_streams = [f for f in streams if f["properties"]["country_code"] == "US"]

# HD streams only
hd = [f for f in streams
      if f["properties"]["resolution"]
      and int(f["properties"]["resolution"].split("x")[1]) >= 720]

GeoJSON schema

Each Feature in streams.geojson:

{
  "type": "Feature",
  "geometry": {
    "type": "Point",
    "coordinates": [-77.4875, 38.7749]
  },
  "properties": {
    "name": "I-95 at Route 123",
    "display_name": "Woodbridge",
    "url": "https://example.com/stream.m3u8",
    "country_code": "US",
    "environment": "traffic",
    "source_family": "vdot",
    "quality_tier": "standard",
    "scene_type": "traffic",
    "url_type": "hls",
    "resolution": "1280x720",
    "coordinates_quality": "exact",
    "status": "active",
    "last_verified": "2026-06-11T22:14:09+00:00",
    "source_url_requires": null
  }
}

Property reference

Property	Type	Description
name	string	Descriptive camera name
display_name	string	Human-readable location label (city/landmark)
url	string	Stream URL (HLS manifest, YouTube link, or HTML page)
country_code	string	ISO 3166-1 alpha-2 country code
environment	string	Scene category (see table below)
source_family	string	Data source identifier (see sources.json)
quality_tier	string	premium, high, standard, or low (production scoring tier)
scene_type	string	Granular scene label (traffic, beach, marina, etc.)
url_type	string	hls, youtube, or html_page
resolution	string/null	Probed video resolution (WxH), where known
coordinates_quality	string	exact, city, or country_centroid
status	string	active or unverified
last_verified	string/null	ISO 8601 timestamp of last successful live probe
source_url_requires	string/null	browser, token_refresh, or null

Environment values

Environment	Count	Description
traffic	3,551	Road/highway cameras
urban	1,375	City and town views
coastal	461	Beach and coastal
mountain	161	Mountain, ski, and volcano
harbor	119	Ports and harbors
lake	89	Lake views
nature	78	Nature and wildlife
waterway	71	Rivers, canals, waterfalls
landmark	54	Landmarks, temples, squares
marina	32	Marinas
transport	6	Airports and stations

URL types

Type	Count	How to use
hls	3,754	Pass directly to ffmpeg or yt-dlp
html_page	1,768	Extract the stream URL from the page (most are SkylineWebcams — fetch the page and pull the m3u8)
youtube	475	Use yt-dlp to extract the HLS manifest

Coordinates quality

Quality	Count	Accuracy
exact	2,373	Camera-level geocoded
city	3,614	City/interchange level (~10 km)
country_centroid	10	Capital fallback — do not use for geo-queries

Source breakdown (top 20)

Source	Streams	Type	Countries
skylinewebcams	1,684	Scenic/tourism	65+
vdot	1,168	Virginia DOT traffic	US
opencctv	746	Open CCTV feeds	Multi
mdsha	404	Maryland SHA traffic	US
deldot	261	Delaware DOT traffic	US
autobahn_nrw	252	German Autobahn (NRW)	DE
trafficvision	245	Traffic aggregator	Multi
manual	224	Hand-curated feeds	Multi
skyline-youtube	167	SkylineWebcams via YouTube	Multi
itic	88	Thailand traffic	TH
youtube	86	YouTube Live	Multi
worldviewstream	82	Scenic network	Multi
balticlivecam	64	Baltic live cams	Multi
camguide	53	Scenic network	Multi
webcamera_pl	52	Polish webcams	PL
iowa_dot	50	Iowa DOT traffic	US
atcs-tasikmalaya	31	Indonesia traffic	ID
beachcam_meo	31	Portuguese beaches	PT
OpenTrafficCamMap	31	Open traffic map	US
nysdot	23	New York State DOT	US

Full per-source metadata (all 67 sources, with active counts) in sources.json.

Country coverage (top 24)

Country	Streams	Country	Streams
US	2,210	MT	59
IT	828	GB	37
TH	462	PT	36
DE	290	PH	33
KR	241	TW	31
ES	226	CZ	28
GR	164	HR	23
ID	220	FR	26
PL	148	RO	26
TR	147	KG	21
JP	129	NZ	13
NO	74	IS	13

98 countries total. Recent expansion focused on rain-prone, previously under-covered regions: Japan (129), New Zealand, Iceland, Norway, Ireland, SE Asia, and the Pacific Northwest.

Directory structure

streams.geojson          # Master file (all 5,997 streams)
sources.json             # Per-source metadata (license, notes, refresh method, active counts)
data/                    # Per-country GeoJSON subsets (e.g. data/US.json)
scripts/
  validate.py            # Probe stream URLs, update status/last_verified
  geocode_swc.py         # Fix fallback coordinates for SWC streams
  regenerate_country_files.py  # Rebuild data/ from streams.geojson
  stats.py               # Print dataset statistics
.github/workflows/
  validate.yml           # Weekly automated validation

Known limitations

1. SkylineWebcams pages (~1,750 streams): Published as page URLs because their direct HLS manifests use auth tokens that expire within hours. Fetch the page and extract the m3u8 at play time. Marked url_type: "html_page", source_url_requires: "token_refresh".

2. YouTube extraction fragility (475 streams): YouTube Live extraction via yt-dlp breaks periodically — keep yt-dlp current. Some streams rotate video IDs when the broadcaster restarts; prefer resolving from the channel when a watch URL dies.

3. City-level coordinates (3,614 streams): Geocoded to the nearest town rather than the exact camera position. Fine for weather lookups (~10 km), not for precise mapping.

4. unverified ≠ dead (1,771 streams): These simply haven't been probed by the production pipeline yet. Dead and rejected streams are excluded from this dataset entirely.

Contributing

To add a new source:

1. Add streams to streams.geojson following the schema above — all fields are required (resolution may be null).
2. Add source metadata to sources.json.
3. Run python scripts/regenerate_country_files.py to update data/.
4. Run python scripts/stats.py to verify counts.
5. Update this README with new totals.

#dataset #computer-vision #live-weather #machine-learning #webcams #hls-streams #geojson #environment-monitoring #traffic-cameras #ai-agents #geospatial