Science · R&D · Innovation
Our research focuses on three questions: how to process meteorological data without losing its origin, how to bring preparedness forward for weather-driven natural disasters, and how to measure the risk that shifts with climate change. We explain our method openly — not the details that make the product different.
Our goals
Goals are not measured success. Next to each one we write down how progress will be measured.
01 · Data processing
Processing multi-source meteorological data while preserving its meaning and origin.
Criterion · share of outputs without a lineage record · open verification reports
02 · Natural disasters
Floods and flash floods, storms, strong wind, heatwaves, snow and ice, coastal events.
Criterion · time from warning to the first preparation action · in exercise and event records
03 · Climate change and adaptation
Translating climate-scale change into the language of local operating thresholds.
Criterion · trends in local threshold frequency · decisions that become adaptation plans
Our scientific model
Weather information is only the beginning. The model has four layers: each inherits the previous layer’s output together with its origin, and no layer silently changes the evidence that came before it.
Calculation, rules and language do not replace one another
calculates
Produces the values, ranges and verification metrics for your point from forecast fields.
Every number you see on screen comes only from here.
applies
Applies your thresholds and the sector pack’s rules deterministically.
Every trigger shows which rule ran, with which value.
explains
Translates the result into the language of your work and writes briefing and checklist text.
It does not calculate, set thresholds or decide.
Public scientific architecture
Forecast fields, official warnings, station observations and satellite products are taken in with their name, time and content identity.
Source evidenceVerification
Multi-model evidence for your point is ordered by lead time. The model range is shown openly; calibration comes only after verification.
Hazard
Your thresholds and exposure meet the weather evidence: which job, at what time, how much it may be affected.
ExposureImpactRiskTiming
Recommendation, checklist, decision record and exercise. Every step can be traced back to its source.
PreparationActionOutcomeLearning
We explain openly how the layers work. Details of threshold calibration, sector rule sets and the order in which evidence is combined are part of the product. This is explanatory architecture — not a formula, a scoring function, a live-engine claim or a ranked provider list.
Public scientific principles
Source remains source.
A provider’s record is never re-labelled as ours.
Missing is not zero.
A missing value stays missing; it never becomes “no rain” or “low risk”.
Agreement is not probability.
Models resembling each other is not a measured likelihood.
Severity is not confidence.
How strong an event could be says nothing about how sure we are.
Official warning is not advisory.
The issuing authority’s warning keeps its own lane; our output never takes its place.
No silent data fusion.
Sources are combined only by documented rules, and the combination stays visible.
No public claim without evidence.
A capability is described at the stage its published evidence supports.
Numerical analysis
Verification is done on the forecast’s own grid and at the exact valid time. The error definition, the metrics and the matching rule are public: an academic should be able to reproduce the same result.
The value is taken from the model’s own grid and the sampling method is recorded. No height correction is applied silently.
The same valid time is viewed from the T+6, T+12, T+24 and T+48 runs, so you see how forecast skill changes with lead time.
Forecast and observation are matched at the exact valid time, with zero tolerance. Nearby timestamps are not silently substituted for exact observation times; an unmatched sample stays out, with its reason recorded.
The range between models is shown openly. Agreement between models is not presented as a probability.
Mean error, mean absolute error and root-mean-square error — by lead time, station and model. Missing data are not treated as zero.
Models are not ranked until a verification campaign is complete and its results are published.
The canonical unit is the kelvin. Observations come from official national station networks: AEMET in Spain, Météo-France in France.
8stations4lead times3models56runs
5,376expected samples · 14 days
The campaign is a controlled pilot panel and makes no claim to represent a whole country. Its launch gates are being checked; results will be published when the campaign is complete.
Big-data processing
Model fields, warning feeds, station measurements and satellite imagery arrive in different formats and at different speeds. Our pipeline processes all of them with the same discipline.
GRIB2, NetCDF, CAP and station data pulled from each institution by name
Kept byte for byte, unchanged, with a content identity
Meteorological codes read with standard tools; meaning preserved
Variable, valid time and lead time in one schema
Sampled to station, site and route points
Forecast and observation matched at the exact time, with a unique key
Daily export and independent verification
A raw file is never overwritten. Every product can be traced back to the bytes it came from.
Every record carries a SHA-256 content identity: the same data is not counted twice, and changed data is noticed.
Deterministic identities and idempotent jobs: an interrupted job can be safely run again.
Two concurrent processes cannot write the same record; long jobs keep their lock alive.
Forecasts, observations and official warnings are kept in separate tables, with separate meanings.
Capacity is planned from measured data; an estimate is never labelled “measured”.
Verification and maturity
Every scientific capability stands on one of five steps. A step is climbed only when its evidence is published. No capability is at the “in production” step today.
Source: public capability manifest
| Capability | 1Code present | 2Locally verified | 3Verified with a live source | 4Calibrated | 5In production |
|---|---|---|---|---|---|
| 2 m air temperatureCanonical variable · kelvinCurrent stage: Verified with a live source | reached | reached | current stage | not reached | not reached |
| Forecast–observation matching2 m air temperature onlyCurrent stage: Verified with a live source | reached | reached | current stage | not reached | not reached |
| Temperature verification metricsBias, MAE and RMSE · 2 mCurrent stage: Locally verified | reached | current stage | not reached | not reached | not reached |
| Heavy precipitation evidenceNot calibratedCurrent stage: Verified with a live source | reached | reached | current stage | not reached | not reached |
| Wind hazard evidenceNot calibratedCurrent stage: Verified with a live source | reached | reached | current stage | not reached | not reached |
| Confidence engineNo scored confidenceCurrent stage: Code present | current stage | not reached | not reached | not reached | not reached |
| Impact risk modelNot publishedCurrent stage: Code present | current stage | not reached | not reached | not reached | not reached |
| FloodHazard evidenceCurrent stage: Not developed yet | not reached | not reached | not reached | not reached | not reached |
| Climatological anomalyNeeded for the climate goalCurrent stage: Not developed yet | not reached | not reached | not reached | not reached | not reached |
Canonical native-lead forecasts, provenance, and official-warning source state are shown separately. Fusion, hazard probability, confidence, and impact risk remain unfinished. This is not a certification.
Statuses come from the public research configuration. VERIFIED is not used as a marketing badge.
Preparedness research
Official warnings remain authoritative. MetOry preparation and advisory outputs are decision support. The two lanes stay parallel and are never merged.
Lane 1
Issued and owned by the national authority, shown unchanged and with attribution. A missing feed does not mean there is no warning.
Lane 2
A separate reading for your activity. Not an official warning, not a safety order and not an all-clear.
Weather evidence or an official warning becomes relevant.
The information reaches the responsible person or team.
Receipt and awareness can be recorded.
Operational relevance is reviewed in context.
Preparation steps are reviewed, assigned and tracked.
A drill, a preparation action or a real response can be recorded.
Outcome and lessons are captured for next time.
Where it applies, an organisation can record a structured trail of warning, delivery, acknowledgement, assignment, action, exercise and review. This describes product and research direction; not every state is live today.
Illustrative / synthetic scenario. Not a real customer activity, not an observed outcome, and not an official warning.
Process records are not a claim that MetOry prevented damage, saved money or reduced risk by a percentage, unless that result is independently measured and verified.
Research & Applied Science
Each programme carries its current status from configuration — not a marketing claim.
Forecast, observation, satellite / Earth-observation and official-warning evidence, retained with source, product, time and provenance.
Where suitable observations exist, forecast evidence can be tested against real observations. Historical and prospective verification methods are under active research.
Studying how meteorological hazards interact with location, exposure and operational context. Not a published loss or probability product.
Studying whether warning information reaches the right person, is acknowledged in time and results in preparation.
Evaluating preparedness procedures through exercises, records and post-event learning.
Event-day programme
Tens of thousands of people travel to one place at one time, stand outdoors for hours and leave together. The event-day product applies the MetOry evidence chain to that day in four phases and keeps every value traceable to its source.
A stadium gathers its crowd in the two hours before kick-off and releases it in one hour after the final whistle. Rain chance, gusts and rain amount at those moments decide covers, gates, queues and the journey home.
A circuit lies outside the city: the journey is longer, the grandstands are open and the race itself is weather-sensitive by rule. The forecast along the access roads and at the circuit are read separately, for the time each one matters.
An exhibition day runs eight hours or more with a steady flow in and out. Entry and exit are spread, so the phases are long and the transport phase repeats; the day is read as a whole and hour by hour.
On the evidence chain
Nothing on this tab is an official warning, a probability or a safety order. Official warnings for the venue's area stay with the issuing authority and are listed on the warnings page.
Collaboration
We welcome joint verification campaigns, exercise research and adaptation studies with universities, meteorological services, municipalities and civil-protection bodies — and dialogue on responsible data use, verification and interoperability.
This describes the direction of MetOry’s work and does not imply official affiliation, endorsement or funding by the European Union, EUMETSAT or any meteorological authority.
Named sources are not ranked, and attribution is not endorsement. MetOry does not replace national meteorological authorities. This page explains method and maturity: it is not an operational dashboard and not an official warning. Goals are not measured success.
