How MarineRadar Trends Works

MarineRadar Trends measures what people pay attention to at sea — which vessels, ports, places and lighthouses the world is searching for and opening on MarineRadar — and turns it into a simple 0–100 interest index. This page explains exactly how those numbers are made, what they can support, and what they cannot. If something here ever disagrees with what the site shows, treat this page as the source of truth and tell us.

What we measure

Every figure in Trends is built from anonymous attention signals: typed searches and the opening of vessel, port, place and lighthouse pages. We count unique devices, never people — one device that opens a vessel ten times counts once for that day. Trends is a measure of audience interest, not of ship movements.

The interest index (0–100)

We never publish raw counts. Instead, each subject is expressed as a relative index from 0 to 100, recomputed on a fixed schedule (roughly every six hours). Because the index is scaled relative to the busiest subject in a given refresh, it answers “how much attention did this get compared with everything else right now?” — not “how many people.” Visitor traffic never touches the underlying database: the public site reads only the pre-computed indices.

Read indices as a within-snapshot ranking, not a time series. The 0–100 scale is re-based each refresh, so a value today is not directly comparable to the same subject’s value last week. Use the shape of a trend and its rank, not the absolute number.

When we publish a number — and when we don’t

Small samples make unreliable rankings, so we apply minimum thresholds and stay quiet below them:

A ranking entry needs a floor of distinct devices behind it before it appears (roughly ten in a day).
Geographic slices need a higher floor (around 25 devices) and are suppressed below it.
A daily Top 10 and a weekly Top 50 are supported; finer cuts (per-city, deep long-tail) are not, and we don’t publish them.

The panel, and how it’s skewed

The audience behind these numbers is global but not a representative sample of the world. It skews heavily toward Android devices, and toward the regions where MarineRadar is most used. We disclose this rather than bury it: the index reflects MarineRadar’s own audience, whose attention MarineRadar’s own pages partly steer. It is a real, first-party signal — and a biased one, like every audience metric.

Privacy by design

The signals are collected to be useful in aggregate and forgettable individually:

Approximate location is rounded to a coarse grid (about 11 km) before it is stored.
IP addresses are dropped once a country is resolved — we keep the country, not the address.
Raw search text is redacted after 90 days.

These rules are engineered into the pipeline, not applied after the fact.

What this data is not

Not AIS or ship positions. Trends measures audience attention. It does not show where ships are, and it cannot explain why anything happened.
Not absolute. A 0–100 index is a relative measure. It is not a vessel count, a visitor count, or a market-share figure.
Not a scientific poll. The panel is skewed (see above); read it as a directional signal.
Not seasonal — yet. Trends launched recently, so there is not enough history to support “normal for this time of year” claims. We won’t make them until there is.
Not a weekday pattern — yet. On launch-period data the day-of-week index collapses almost all signal onto one or two days, which is a normalisation artifact rather than a real weekly rhythm. The seasonality page therefore withholds that breakdown, and says so in place of the chart, until at least five of seven days carry signal.

Corrections

Trust means being fixable. When we get a published number or claim wrong we correct it in public and record every one in the corrections log. To report an error or request the data behind a chart, email connect@marineradar.com or see the press & data kit.

Most recent correction

30 July 2026: The seasonality page published a day-of-week breakdown that collapsed almost all signal onto one or two days. On launch-period data that is an artifact of how the index is scaled, not a real weekly rhythm, so the chart and the accompanying “peak weekday” figure should not have been presented as findings. Both are now withheld, with the reason shown in their place, until at least five of the seven days carry signal.

See the full corrections log →

Changelog

What has changed in how these numbers are built and described, newest first. We publish this because a measurement that quietly changes shape is worse than one with known limits — if you quoted a figure from this site, this list tells you whether the ground moved under it.

30 July 2026 Correction: Weekday breakdown withheld until the data supports itOn launch-period data the day-of-week index collapsed almost all signal onto one or two days — a normalisation artifact, not a real weekly rhythm. The seasonality page now withholds both the weekday chart and the “peak weekday” figure and explains the gap in their place. Every claim about weekday coverage on that page — the summary copy, the page description and the structured data — is gated on the same test, so none of them can promise a breakdown that isn’t shown. It returns on its own once at least five of the seven days carry signal.
30 July 2026: First chokepoint board publishedA permanent page for the Strait of Hormuz, charting the last 30 complete days of attention and placing the most recent day within that window’s own range. It deliberately does not claim “above normal”: because the index is re-based each refresh there is no fixed baseline to compare against yet, and the window’s highest day is the normalisation ceiling rather than a measured level. Both limits are stated on the page.
21 July 2026: Public corrections log openedEvery fix to a published number or claim is now recorded in the corrections log linked above, with the most recent entry mirrored on this page. Before this, corrections were made but not published.
21 July 2026 Correction: Explorer wording corrected: attention is not AISExplorer copy that described these figures as “AIS-driven” was wrong and has been removed. The index reflects what people are watching — searches and page opens — not ship positions. Reduced in scope on 17 July and fully removed on 21 July.
17 July 2026: This methodology page publishedFirst public statement of the method: what is measured, how the 0–100 index is built and why it is not comparable across refreshes, the publication floors, the panel’s Android and regional skew, the privacy rules already engineered into the pipeline, and an explicit list of what this data is not.
17 July 2026: Location coverage stated inline, next to the geographic chartsThe geographic view now names its own coverage where the charts are — what share of users and of entities actually carry coordinates — alongside the roughly 11 km rounding applied before storage and the minimum-device floor below which geographic slices are absent rather than shown small. Previously a located minority could be read as the whole audience.
17 July 2026: Data-freshness timestamps addedTrends pages now show when the figures on them were last computed, so a number is never mistaken for live where the refresh is periodic.
15 July 2026 Correction: Growth figures floored and cappedExtreme growth percentages produced by tiny denominators — “+40,100%” and similar — were floored at a minimum device count and capped, and the sentinel value behind “+9999%” was removed. The “MarineRadar AIS network” source line was removed from stories at the same time, and attribution was reframed to state honestly how much of the signal is traced to a known source.

← Back to Maritime Trends