One architecture, any ground truth

DATUM.

The forecast, corrected to exactly where you are — not the region around you.

datum /ˈdā-təm/ noun
In surveying, the fixed reference point every other measurement gets corrected against. Datum does the same for weather: one real, local reading becomes the reference every forecast model is checked — and corrected — against. It's the foundation of an ecosystem that merges precision local weather data with the world's leading global forecast models into one corrected forecast neither could produce alone.
How the architecture works

Grade every model. Blend the winners. Correct for what's left.

National weather agencies have run this exact play for decades: take every model available, weigh each by how well it verifies against real observations, and correct for what's left over. That process — data assimilation — has always lived behind billion-dollar infrastructure, built for regions, not for one specific rooftop, mast, or ridgeline. Datum runs the same grade-blend-correct logic against a single point of real ground truth, at whatever scale a sensor already there can reach.

MODELS AccuWeather Meteoblue GFS ECMWF ICON GROUND TRUTH DATUM GRADE → BLEND → CORRECT
01

Plug in a ground truth

Any real sensor already producing readings qualifies — a backyard station, a boat's wind instrument, a farm probe, a remote ridgeline unit. No new hardware, no new network.

02

Grade every model against it

AccuWeather, Meteoblue, and named models like GFS, ECMWF, and ICON are scored against that ground truth, at every lead time from tomorrow to a week out.

03

Blend and correct

Models are combined by that track record and corrected for each one's own known bias at that exact point — a single number more accurate than any source alone.

04

Ask, don't scroll

One button, one conversation, spoken back in plain language. The same interface works whether the ground truth sits in a yard, on a mast, or on a ridge.

Four solutions, one architecture

Built on the same engine. Pointed at different ground.

Every solution below runs the identical grade-blend-correct logic — the only thing that changes is what real sensor it's pointed at.

Dooryard
Personal weather, finally personal
Ground truth: the personal weather station already mounted outside
Home and yard weather, corrected to your exact coordinates — frost dates, storm timing, whether you need a jacket before the bus comes.
“Is it going to frost my tomatoes tonight?”
View full details →
Beaufort
Marine weather, true to the wind
Ground truth: the wind instrument already on the mast
Marine wind and weather, corrected to wherever you've dropped anchor — gust timing, wind shifts, true wind at the top of a beat.
“Will the wind back before we round the point?”
View full details →
Treeline
Backcountry & alpine weather, true to the ridge
Ground truth: remote ridgeline stations already run by huts, ski patrols, and avalanche centers
Backcountry and alpine weather, corrected to your exact ridge — wind-loading, whiteout timing, and melt-freeze cycles.
“Will this ridge still be wind-loaded by noon?”
View full details →
Furrow
Row crop & orchard weather, true to the field
Ground truth: ag-weather stations already common on working farms
Row crop and orchard weather, corrected to your exact field — frost protection timing, spray windows, and irrigation scheduling.
“Will it drop below freezing in the low block tonight?”
View full details →
Solutions currently in production

Where else this architecture is headed

Windsock — small airfields & gliding
Ground truth: AWOS-style sensors already at private strips and clubs
Crosswind limits and thermal timing corrected to the one runway or ridge you actually fly.
In production
Rooftop solar & storage
Ground truth: the array's own inverter output, an implicit cloud-cover sensor
Production forecasting and storage dispatch timing for one specific roof, not a regional irradiance map.
In production
The asset behind the architecture

The solutions bring the ground truth in. What compounds behind them is worth more.

Each Datum solution is built to serve one customer's ground truth — but every station added quietly makes every other station's forecast a little smarter too. This is the exact parallel to what happened at Weather Underground: a consumer product whose real asset turned out to be the network itself, dense enough to see what the big government models couldn't — valuable enough that IBM's acquisition fed it directly into GRAF, IBM's own forecast model. Synoptic Data runs a version of the same business today, aggregating access to over 170,000 mesonet stations. Datum is positioned to build the same kind of asset on purpose, from day one.

Cold start disappearsNew signups inherit a warm-start bias correction from similar existing stations instead of waiting weeks.
A bias map, not a bias pointModel error becomes a continuous surface across a region, not one station's number.
The data becomes a resellable assetThe same path Weather Underground took — verified ground truth has value beyond any one app.
MORE VERIFIED GROUND TRUTHS BETTER BIAS MAPS & COLD-START BETTER PRODUCT MORE CUSTOMERS