Measuring instead of estimating
Why this page exists.
Without an instrument, the calculator works out WBGT from a weather forecast. It has to infer the two quantities that matter most — the reading a black globe would give in the sun, and the natural wet bulb temperature — because a forecast does not contain them. That inference is good, and it is measurably imperfect: driving the chain from a forecast rather than from site sensors costs about 1.5 °C, which is the single largest term in the uncertainty range shown next to every estimate.
A Kestrel 5400 has an actual black globe and reports WBGT to a published ±0.7 °C. That is the prize, and this build has not won it yet: the byte offsets we can read do not cover the globe or the meter's own WBGT, so what a connected meter contributes today is measured air temperature, humidity and wind. The calculator models a WBGT from those plus the forecast's solar radiation, and the uncertainty range does not narrow — three measured inputs out of four does not earn a narrower one, and the ±0.7 °C figure belongs to the meter's own WBGT output, which we cannot read.
A measurement describes where the instrument is. If the meter sits in shade and the crew works in full sun, the crew's exposure is higher than the number on this screen. Nothing in software can detect that — put the meter where the work happens.
Kestrel meter
A 5-series meter or a DROP logger, over Bluetooth, directly from this browser.
Kestrel's 5000-series meters with the LiNK option broadcast their readings over Bluetooth Low Energy. This page can subscribe to that broadcast and read temperature, humidity, wind and — on a 5400 — the black globe and WBGT.
The chooser also lists Kestrel DROP loggers (D1, D2, D3, D3FW). They speak a different Bluetooth service, so this page works out which family you picked from what the device offers rather than from what it is called, and shows only the fields your model actually has.
The link does not follow you to the calculator. A Bluetooth connection belongs to one page — that is the specification, not a limitation of this site — so walking from here to the calculator drops it every time. You do not need to come here first: the calculator has its own Connect Kestrel button, and connecting there is one tap instead of two. This page is for checking what the meter is actually sending.
The meter needs about eight minutes to settle after a large change in conditions, and about fifteen to be fully accurate; Kestrel publishes those figures. Carried straight out of an air-conditioned truck it reads low, which is the unsafe direction, so the calculator holds off using a reading until it has stopped drifting.
The meter supplies measured air, never a measured WBGT. The published description of what each byte in a Kestrel broadcast means covers the 5500, which has no black globe. Air temperature, humidity, wind, pressure and wet bulb decode from that description; the black globe and WBGT sit in bytes it does not describe, and there is no honest way to guess which. So the calculator takes the air temperature, humidity and wind and models a WBGT from them plus the forecast's solar radiation — no Kestrel measures the sun — and it labels the result modelled. The uncertainty range does not narrow, because three measured inputs out of four does not earn a narrower one.
Those offsets are still unconfirmed against a 5400. Check the values below against the meter's own screen before you trust a schedule built on them.
A DROP shows readings here and drives nothing. No DROP has a black globe and no DROP measures wind, so it cannot report a measured WBGT — and it cannot feed a modelled one either, because that needs air temperature, humidity, wind and sunshine, and a DROP supplies two of the four. What a DROP is good for is checking a forecast against the air where you actually are: temperature on every model, humidity, dew point and heat index from a D2 up, pressure and density altitude on a D3.
A D3FW also reports a wet bulb, and this page shows it without using it. It is a psychrometric wet bulb — a properly ventilated one — and heat-stress limits are written against the natural wet bulb, which is the same sensor left in the wind and sun it happens to be standing in. The two differ, we have no verified correction between them, and the calculator withholds the 5 series' wet bulb for exactly the same reason.
The connection does not follow you to the calculator. A Bluetooth link belongs to one page, so walking from here to the calculator drops it — that is how browsers are built, not a fault. What survives is this browser's permission for your meter, so the calculator reconnects on its own where it can, and offers a Connect Kestrel button where it cannot. A DROP is deliberately not remembered for the calculator, because there is nothing there for it to do.
If a meter that worked a minute ago will not reconnect, it is most likely not listening: a Kestrel stops advertising about 30 seconds after it disconnects. Toggling Bluetooth off and back on at the meter itself wakes it up.
A forecast's error is not one number, it is one number per place: the same provider that runs two degrees cold in a paved yard can be right on a grass verge two hundred metres away. Naming the site keeps each one's comparison separate below, instead of averaging two microclimates into a figure that describes neither. Leave it blank and readings are grouped by coordinate instead. The name stays in this browser.
| Reading | Decoded |
|---|
Raw packets
The last few packets exactly as they arrived, and every signed 16-bit value that could be read from each. This is the mapping tool: move the meter between shade and sun, watch which offset tracks its globe reading, and that offset is the one we are missing. Also switch the meter between °C and °F while watching. If these numbers move when you do, the meter transmits in whatever unit it displays, and the decode has to read that setting before it can be trusted.
Weather station
Your own Ambient Weather Network station, read straight from this browser.
If your instrument uploads to Ambient Weather Network — a KestrelMet station, or a Kestrel heat stress tracker through its gateway — this page can read it directly. A Kestrel publishes its black globe reading and its natural wet bulb alongside WBGT, which are the two quantities a forecast cannot contain, so nothing about the result is estimated.
Two different keys. The API key identifies you and is created
on your Ambient Weather account page. The application key identifies the
software and is issued separately — if the account page does not offer one, Ambient
Weather asks you to request it from support@ambientweather.com.
Both are required; pasting the same key into both fields returns an error.
Your keys are stored in this browser only. Requests go from here straight to Ambient Weather — they never pass through our server, and we never see either key. Anything with access to this browser profile can read them, which is the trade for not handing them to us; “Forget keys” deletes them.
Weather Underground PWS
Your own personal weather station uploading to Weather Underground, read straight from this browser.
Connect a Weather Underground station
If you own a personal weather station that is actively uploading to Weather Underground, this page can read it directly. Weather Underground reports solar radiation alongside temperature, humidity and wind — a black globe reading and WBGT are not part of its data, so those stay estimated, same as any other weather-station source here.
The API key must be your own, and your station must be uploading. Create it at wunderground.com/member/api-keys. The key only returns data for a station that is actively sending observations — a registered station that has gone quiet returns nothing, not an old reading.
Free-tier use is non-commercial, and Weather Underground requires a visible credit wherever its data is shown. This page shows that credit next to every reading below and in this panel; see wunderground.com/company/legal for the full terms.
Weather data from Weather Underground.
Your station ID and key are stored in this browser only. Requests go from here straight to Weather Underground — they never pass through our server, and we never see your key. “Forget keys” deletes them.
PWSWeather (via Xweather)
Your own PWSWeather station, read through Xweather's API, straight from this browser.
Connect a PWSWeather station
Your PWSWeather station ID and station key will not work here. Those two — the ones
from pwsweather.com/pwsupdate/pwsupdate.php?ID=&PASSWORD= — are
upload-only: they let your station push readings in, and they cannot read anything back out.
Pasting the station password into a field below will fail with no useful explanation, because it is
simply the wrong credential for this job.
Reading a station back out requires a separate, free Xweather account: a client ID and a client secret, issued for PWS contributors at signup.xweather.com/pws-contributor. Your station ID (the same one your station uploads under) is still needed below to say which station to read.
“Secret” sets an expectation this storage does not meet. Your client secret is stored in this browser's local storage, same as every other key on this page — readable by anything with access to this browser profile. That is the same trade you already accepted for Ambient Weather and Weather Underground above, for the same reason: it is your credential, for your station, and it never reaches our server. It is said here plainly because the word “secret” implies more protection than local storage provides.
Your station ID, client ID and client secret are stored in this browser only. Requests go from here straight to Xweather — they never pass through our server, and we never see them. “Forget credentials” deletes them.
Compare weather stations
How closely the forecast has matched your own instrument, at each place you have used it.
Every hour a connected meter drives the calculator, this browser keeps one paired record: what the meter read, and what the forecast said for that same hour at that same place. Enough of those and the pattern stops being anecdote — you can say whether the forecast runs consistently cold at a site, by how much, and how tightly the two track each other.
A plus sign means the forecast is running cold. Bias is the meter's reading minus the forecast's, averaged. Positive means your instrument reads hotter than the forecast did — conditions on the ground are worse than predicted, which is the direction that gets people hurt. Negative means the forecast is over-stating the heat.
RMSE is typical error size, and it punishes occasional large misses more than MAE does — RMSE much larger than MAE means the forecast is usually close but sometimes badly wrong. The fit line and R² come from a least-squares regression of measured against forecast: a slope near 1 with a non-zero intercept is a forecast that is simply offset and can be corrected for; a slope far from 1 is one that compresses or exaggerates the range.
R² is a fit, not a forecast skill score. Hours within one day are strongly correlated with each other, so this figure is flattered by how weather works and should not be read as an out-of-sample accuracy.
Stored on this device
Everything this site keeps in your browser, what it is for, and a button to delete it. Nothing here is ever sent to us — the full account is on the privacy page.
| Key | What it holds | Size | Delete |
|---|---|---|---|
| Reading your browser storage… | |||