Download and reuse

Everything behind the map, the charts and the Sea page can be downloaded and reused under its licence. Nothing here needs an account.

What you can use, and how to credit it

Aggregates only: the GNSS files hold counts per H3 cell and region, never aircraft identifiers. The daily files are made automatically and not reviewed by a person, apart from the 20 validated days (on the Validation page).

GNSS degradation: monthly releases

One GitHub release per month, named pnt-data-YYYY-MM, with one set of files per UTC day. A day's files appear the next morning, after the daily run at 05:40 UTC (about 12:40 AM CDT), with a second try at 12:40 UTC for late archives. Download one file directly with https://github.com/metacog-osint/noise-floor/releases/download/pnt-data-YYYY-MM/<file>.

All releases: github.com/metacog-osint/noise-floor/releases.

Gaps are at the source: adsb.lol published no archive for 31 Dec 2025 or 6 May 2026, its 5 May 2026 archive holds no traces, and its 7 May 2026 archive covers about 6 hours.

Columns in the per-cell table

"Bad" means an aircraft with half or more of its reports in that cell (or cell-hour) below the threshold. Reports are ADS-B version 2, the aircraft's own GNSS position only (type adsb_icao), airborne (1,000 ft or more), sampled at most every 30 s per aircraft. Aircraft whose avionics were faulty that day (bad where their neighbours were fine) are dropped. Full method: Methods and the method card.

The hourly table has the same counts per cell and hour (0–23, UTC). The minute table has h3, minute (0–1439, UTC), bad_nacp, latched and, from 25 Oct 2026, heavy.

Columns in the position-loss tables

Strong jamming can stop an aircraft sending any GNSS position, so it drops out of the tables above instead of counting as degraded. Multilateration (MLAT: receivers timing the same signal) can still place many of them. These tables count aircraft at 10,000 ft or more, any version, per H3 cell (resolution 4): loss.parquet per hour (0–23, UTC), loss_daily.parquet once per day with lat and lon. They exist only for days processed after the daily run started making them (late Oct 2026).

mlat_only / heard is a lower bound on position loss: aircraft that neither send a position nor are placed by MLAT can't be counted, and MLAT coverage differs by place and has grown over time (east of Kaliningrad it went from a few cells to most of them between mid-2025 and mid-2026). Compare places and years with mlat_any in view. Method and checks: position-loss.md.

Read one day in Python

import io, urllib.request
import pyarrow.parquet as pq

day = "2026-10-01"
url = f"https://github.com/metacog-osint/noise-floor/releases/download/pnt-data-{day[:7]}/{day}.parquet"
table = pq.read_table(io.BytesIO(urllib.request.urlopen(url).read()))
rows = [r for r in table.to_pylist() if r["aircraft"] >= 5]
jam = [r for r in rows if 100 * max(r["bad_nacp_live"] - 1, 0) / r["aircraft"] >= 10]
print(len(rows), "cells with 5 or more aircraft;", len(jam), "with severe GPS error")

Open the data in ArcGIS or QGIS

The tables give each cell as an H3 cell ID with its centre. This script turns one day into hexagon polygons in GeoJSON, which QGIS, ArcGIS Pro and ArcGIS Online all open:

import io, json, urllib.request
import h3, pyarrow.parquet as pq          # pip install h3 pyarrow

day = "2026-10-01"
url = f"https://github.com/metacog-osint/noise-floor/releases/download/pnt-data-{day[:7]}/{day}.parquet"
rows = pq.read_table(io.BytesIO(urllib.request.urlopen(url).read())).to_pylist()
features = []
for r in rows:
    if r["aircraft"] < 5:
        continue
    ring = [[round(lon, 5), round(lat, 5)] for lat, lon in h3.cell_to_boundary(r["h3"])]
    if max(p[0] for p in ring) - min(p[0] for p in ring) > 180:    # skip the few cells that cross 180 degrees
        continue
    ring.append(ring[0])
    props = {k: r[k] for k in ("h3", "aircraft", "bad_nacp_live", "score_live")}
    features.append({"type": "Feature", "properties": props, "geometry": {"type": "Polygon", "coordinates": [ring]}})
with open(f"{day}.geojson", "w") as f:
    json.dump({"type": "FeatureCollection", "features": features}, f)
print(len(features), "cells written to", f"{day}.geojson")

• QGIS (free): Layer → Add Layer → Add Vector Layer, choose the file, then style by score_live with a graduated colour ramp.
• ArcGIS Pro: the JSON To Features tool reads GeoJSON into a feature class.
• ArcGIS Online: in Map Viewer, add a layer from a file. A whole day is about 24,000 cells and 8 MB; to keep the file small, keep only cells in your region, or only those with bad_nacp_live above 0.
• The Sea page's ship downloads are already GeoJSON and CSV (with lat and lon columns), so they open directly.

The ODbL applies to whatever you build from these files: credit adsb.lol contributors and Noisefloor, and share derived databases under the same licence.

Site files

The pages read small JSON files that you can use directly. They're refreshed after each daily run.

How to cite

Noisefloor (2026). GNSS degradation from ADS-B, daily, method v1.4. The Intelligence Company. https://noisefloor.theintelligence.company. Data derived from adsb.lol, ODbL 1.0.

Code: github.com/metacog-osint/noise-floor.