Skip to main content
Open the data journey

The fourteen stages, as text

Wildfire Ahead · the data journey

One satellite sighting, followed through fourteen stages

At 14:22 a satellite passes over northern British Columbia and records a hot patch of ground. Every stage below says what it is for, what reaches it, what leaves it, the rule it enforces and who acts on it. The record only ever gains information. Nothing already on it is edited.

Stage 01 · Getting the data in

Where the data comes from

The system is plugged into 46 official data feeds — satellites that spot heat, weather services, fire agencies, drought monitors and maps of what is on the ground. It checks them on a schedule, day and night.

Why this step exists
Because nobody has all the information. It is scattered across satellites, weather services and fire agencies in two countries.
What comes in
A satellite passes over northern British Columbia and spots a hot patch of ground. NASA publishes it a few minutes later.
What goes out
The file exactly as NASA sent it, plus a note of when we asked for it.
The rule that protects this
Before anything is downloaded, somebody has written down who owns this feed, what we are allowed to do with it, and how out-of-date it is allowed to get.
Data steward
The person responsible for one data feed. They record where it comes from, what its known weaknesses are, and what the platform is allowed to use it for. Think of a librarian who knows which reference books are reliable and for what.
What the record gains here
Where it came from: NASA satellite feedWhen we fetched it: 14:22:07Who published it: NASA FIRMS

Stage 02 · Getting the data in

Translating each agency's format

Every agency formats its data differently — different column names, different units, different words for the same thing. A small translator sits at the door for each one so nothing downstream has to care.

Why this step exists
Because every agency writes their data differently, and the rest of the system should not have to know about any of it.
What comes in
A spreadsheet-style file using NASA's own column names and their own wording for how confident they are.
What goes out
The same file, untouched, plus a translated copy the rest of the system can read.
The rule that protects this
Adding a new agency means writing one new translator. It never means changing anything else.
Nobody — this part is automatic
Entirely automatic, and repeatable — the same file always translates the same way.
What the record gains here
Which translator ran: NASA satellite reader v2.1Translator version: 4

Stage 03 · Getting the data in

Keeping the original, unchanged

The file as it arrived is written down once and never edited again. Everything that happens later is a copy with its own history. If somebody asks two years from now what we actually received that morning, the answer still exists.

Why this step exists
Because if you can quietly change what arrived, nobody can ever prove what was actually known at the time. This is the single most important idea in the whole system.
What comes in
The provider's file and the exact request that fetched it.
What goes out
A permanent record: a fingerprint of the file, its size, and exactly when it arrived.
The rule that protects this
Corrections never overwrite. A correction is a new entry that points back at the old one, and the old one stays readable.
Nobody — this part is automatic
Automatic, and deliberately impossible to fake.
What the record gains here
Permanent record number: rcpt_01J9F2KFingerprint of the file: 9f3ac1…e42bSize: 184,220 bytes

Stage 04 · Getting the data in

Checking it before letting it in

The file is held in a sealed area and checked before anything else can touch it: is it really the kind of file it claims to be, are the coordinates plausible, are the timestamps sensible, has anything been tampered with.

Why this step exists
Because a corrupted or mislabelled file that gets into a live system is very hard to get back out.
What comes in
The stored file and its permanent record.
What goes out
Either permission for exactly this file to go further, or it stays held, with the reason written down.
The rule that protects this
A thermal image whose camera calibration is unknown is held, not scored. And if a file has changed since we stored it, it is refused before anything reads it.
Data steward
A person reviews anything held and decides whether it may go further. Because a person does it, there is always somebody accountable for what entered the system.
What the record gains here
Check result: Passed — allowed throughHow sensitive it is: Public — can be sharedWho let it through: A named data steward

Stage 05 · Making sense of it

Putting everything in the same units

Temperatures, coordinates, times and category names are all converted into one common form — while keeping the original values right beside them, so nothing is lost.

Why this step exists
Because one agency measures in Celsius and another in Kelvin, and nobody should be doing that arithmetic in their head at three in the morning.
What comes in
Values in whatever units and reference system the agency happens to use.
What goes out
The same information in the platform's standard form, with the original still attached.
The rule that protects this
There is a real difference between “we looked and there was nothing”, “we did not look”, “this does not apply here”, “we are not allowed to tell you” and “the answer is zero”. The system keeps all five apart and never quietly turns one into another.
Nobody — this part is automatic
Entirely automatic and repeatable.
What the record gains here
Where on the map: 54.207° N, 118.412° WMap system used: World standard (WGS 84)How hot: 412.6 K (139 °C)Original values kept: Yes

Stage 06 · Making sense of it

Making different agencies comparable

Different agencies use the same words to mean different things. One rates confidence out of a hundred; another uses low, medium and high. A reviewed conversion table maps between them — and, importantly, records exactly what detail the conversion throws away.

Why this step exists
Because two agencies can both say “high confidence” and mean genuinely different things — and if you merge them without saying so, you have invented a number nobody measured.
What comes in
A tidy record that still uses one agency's private vocabulary.
What goes out
A record that can be compared with other agencies' — and a note saying how safe that comparison actually is.
The rule that protects this
If a conversion loses detail, it has to say which detail. Anything that cannot be mapped is marked unknown rather than being pushed into the closest-looking category.
Data steward
Proposes and looks after the conversion table between two agencies' wordings.
Science lead
A scientist decides whether the two agencies really are describing the same thing. Once they approve it, that version of the table is frozen, so results stay reproducible.
What the record gains here
How comparable: Converted — usable, with caveatsConversion used: NASA confidence → platform scale, v1.2What it lost: A 0–100 scale became three levels

Stage 07 · Making sense of it

Working out if it is the same fire

One fire can appear in several systems at once, each with its own reference number, and sometimes on both sides of the border. The system proposes that two records describe the same fire — but it never glues them together.

Why this step exists
Because the same fire shows up in a satellite feed, a provincial database and a US database under three different numbers — and counting it as three fires would be wrong in a way that matters.
What comes in
A single observation with no idea what else in the world describes the same event.
What goes out
A suggestion — “this looks like the same fire as that one, and here is why” — waiting for a person to agree or disagree.
The rule that protects this
Both original records always survive, which is what makes a wrong link fixable. And if an agency's official reference number says these are different fires, that beats any score the computer produces, however confident it looks.
Data steward
A person decides whether the suggestion is right. Their decision is recorded and overrides whatever the computer proposed.
What the record gains here
Possible match: Same fire as BC-K52104How sure: 86%Match across the border: None found

Stage 08 · Making sense of it

Making it findable

The record is written into an index: where it is, when it is from, how detailed it is, where it came from and what it may be used for. Anyone with permission can now find it without knowing it existed.

Why this step exists
Because information nobody can find is the same as information you never collected.
What comes in
A cleaned-up, comparable record.
What goes out
An index entry that other systems can search.
The rule that protects this
Being in the index does not mean it is accurate, current, or allowed to be published. Those are separate questions with separate answers.
Nobody — this part is automatic
Automatic — and if the index were ever lost, it could be rebuilt from the permanent records.
What the record gains here
Index entry: Fire detections, 10 SeptemberWhat we may do with it: Open, credit NASAWhat it was built from: 3 earlier records

Stage 09 · Making sense of it

Preparing what the models learn from

Before a model can learn anything, the system assembles what was known at that exact moment — the weather, the dryness, the terrain, the fire history — using only information that genuinely existed then.

Why this step exists
Because a model that accidentally saw the future during training looks brilliant in testing and fails the moment it meets a real fire.
What comes in
Everything the platform holds that is relevant to that place and time.
What goes out
A frozen, dated package of inputs that can never change afterwards.
The rule that protects this
Nothing that was discovered later is allowed in. The system is built so that looking into the future is impossible, rather than merely discouraged.
Data steward
Publishes the package and freezes it, so any result built on it can be reproduced exactly.
What the record gains here
Frozen input package: Week 37, version AEverything known by: 14:31Future-peeking checks: 6 passed

Stage 10 · Making sense of it

Working out what it means

Now the science runs. Models estimate how likely this really is to be a fire, and how it might behave — always inside the conditions they were actually tested for.

Why this step exists
Because a hot patch of ground is not the same thing as a fire, and somebody has to estimate how likely it is — with an honest sense of how sure that estimate is.
What comes in
The frozen package of inputs, and a specific version of a model.
What goes out
An estimate with an honest range around it — or a refusal to answer, if this situation is not one the model was tested on.
The rule that protects this
A model gives an estimate. It never gives an order, and it can never approve its own use — the person who built it is not allowed to be the person who signs it off.
Model engineer
Builds and trains the model against a plan that was written down and frozen before the work started.
Independent checker
Checks the work independently. Deliberately not anyone who helped build it — otherwise it is marking your own homework.
Release approver
Decides whether it may be used, for what, and where — together with a second named person.
What the record gains here
Which model version: Detection model 3.4.1Exact version fingerprint: sha256:c17e…How it may be used: Decision support only

Stage 11 · Making sense of it

Saying how sure we are

Every estimate is wrapped in the same standard packaging: when it was made, how long it is good for, what area it covers, how uncertain it is, what it assumed, and when you should not trust it.

Why this step exists
Because a number without a range, a time and a set of assumptions is the kind of number that gets someone hurt.
What comes in
A bare number.
What goes out
The same number, but now you can see exactly what it does and does not tell you.
The rule that protects this
If the situation is outside what the model was tested on, it says so and declines — rather than producing a confident-looking number that happens to be wrong.
Nobody — this part is automatic
Automatic, and enforced — a result that arrives without its limits attached is rejected by an automatic check.
What the record gains here
Answer available: YesHow sure, as a range: 71% to 94%Good until: 16:22Known limitations: 2 stated

Stage 12 · Putting it to work

Combining what every sensor saw

Everything that saw this patch of ground is brought together into a single candidate. Confidence builds as more sensors agree — but sensors that tend to agree with each other anyway count for less, because they are not really independent.

Why this step exists
Because the same fire is seen by several sensors, and two satellites that look alike are not two independent opinions.
What comes in
Separate sightings from two or more sensors, over overlapping ground.
What goes out
One candidate fire with a confidence history you can scroll back through — 41% at 07:02, 94% by 07:18.
The rule that protects this
When two sources disagree, you see the disagreement. It is never averaged into a single comfortable number that hides it.
Detection operator
The person watching for new fires. They work through the candidates and decide what each one actually is, recording their reason.
Detection viewer
Anyone allowed to see the candidates and how confidence has changed, without changing anything.
What the record gains here
Candidate number: cand_8827Confidence over time: 41% → 94%Which sensors saw it: Two satellitesDisagreements kept: 1

Stage 13 · Putting it to work

Showing the working

Every number the system shows can be opened up: what went into it, how much each thing mattered, where each piece came from, how old it is, and what disagreed. No black boxes.

Why this step exists
Because an agency will not act on a number it cannot argue with, and it should not have to.
What comes in
A candidate with a confidence number.
What goes out
The same candidate, with a complete explanation attached to it.
The rule that protects this
No important number is ever displayed without the working behind it.
Situation analyst
The person building the picture of what is happening. They open the working, see what moved the number, and judge whether it holds up.
What the record gains here
What drove the number: 6 factors, 5 sourcesOldest thing used: 41 minutes agoWhat would settle it: A thermal patrol over ridge 14

Stage 14 · Putting it to work

Getting it to the people who decide

Finally it reaches people. It appears on the operations screen for the crews working the fire, in a ranked list of what to check first — and, once somebody with the authority approves it, as a public record with the precise location blurred.

Why this step exists
Because none of the previous thirteen stages matter unless the right person sees the right thing in time to act on it.
What comes in
A candidate with its full explanation.
What goes out
A ranked job on somebody's screen, a layer on the map, and — after approval — something the public can read.
The rule that protects this
The public side is completely separate, so a surge of public traffic can never slow down or reach the operational side. Public locations are deliberately blurred to a minimum radius before they leave.
Incident commander
The person in charge of fighting this fire. They decide. The platform's only job was to put the geometry, the trade-offs and the unknowns in front of them faster than before.
Firefighter in the field
Gets the job on a phone that keeps working with no signal, and records what they actually find when they get there.
Public information approver
The person authorised to decide what the public is told. Nothing reaches the public without them.
What the record gains here
Where it ranks: 3rd of 47 to checkPublic release: Waiting for approvalWhich agency owns it: BC northeast

Open the data journeyOpen the slide deckThe short briefingWildfire Ahead home