Isolate the dominant bias source
Separate cloud-shade errors from aerosol attenuation and panel temperature losses — the three primary drivers in the Gulf.
← All solar products
Illustrative solar and atmospheric imagerySOLAR FORECAST DIAGNOSTICS
Attribute forecast error to cloud, dust and temperature — hour by hour.
Compare observed versus forecast generation for each hour. Isolate the contribution of cloud cover, aerosol optical depth, dust risk and panel temperature to the error. Improve future forecasts with transparent, machine-readable attribution.
AVAILABLE · POST ENDPOINTSee how it works ↓
THE DECISION IT SUPPORTS
Why did this hour's generation differ from the forecast?
The attribution API takes hourly pairs of forecast and observed generation alongside auxiliary satellite signals. It decomposes each hour's error into contributing factors and returns a residual — what is still unexplained after the available signals are exhausted.
Built for
- Solar plant operators
- Air quality monitoring teams
- Energy traders and grid operators
- Environmental change analysts
WHY TEAMS USE IT
Turn broad data into a clearer next step.
Each result is designed to narrow a decision, show the inputs behind it, and surface what must still be checked locally.
Prioritise forecast improvement work
A consistent cloud bias calls for a different fix than a persistent dust undercount. Attribution makes the path clear.
Communicate error with evidence
Operations and trading teams receive an attributed explanation, not a raw number.
Track confidence per hour
Each hour carries a data-confidence flag — high, medium or low — based on how many signals were available.
FROM REGION TO SITE
See where each signal fits.
Images illustrate the operational setting. They are not live observations; use source timestamps and local measurements for decisions.
FORECAST VS ACTUALCompare modelled to observed generation
Hour-by-hour error attribution separates cloud, dust and temperature effects.
DUST SIGNALDistinguish aerosol attenuation from cloud shading
AOD and dust risk score help isolate the Gulf's dominant bias source.
DATA CONFIDENCEKnow when to trust the attribution
Each hour carries a confidence flag based on input completeness.
THE INFORMATION STACK
What goes in—and what it helps explain.
Availability is stated for every input. Missing site telemetry, irradiance readings, or atmospheric observations are identified rather than inferred.
INFORMATIONSOURCEHOW IT IS USEDSTATUS
01Hourly forecast (kWh/kWp)Plant forecast systemBaseline to attribute error against.REQUIRED
02Hourly observed (kWh/kWp)Plant SCADA / meterActual generation for comparison.REQUIRED
03Cloud cover (%)AtmosGrid GFS / satelliteEstimates cloud transmission factor.OPTIONAL
04Aerosol optical depthCAMS / GFS visibility proxyQuantifies beam attenuation by aerosol.OPTIONAL
05Dust risk score (0–100)AtmosGrid derived productUsed when AOD is not available.OPTIONAL
06Ambient temperature (°C)AtmosGrid GFS / plant sensorPanel efficiency correction vs STC 25 °C.OPTIONAL
A PRACTICAL WORKFLOW
From input to reviewable decision.
The workflow keeps model context, field evidence and human approval in their proper places.
- 01
POST hourly pairs
Supply forecast and observed kWh/kWp for each hour alongside any available auxiliary signals.
- 02
Receive per-hour attribution
Each hour returns a list of contributing factors with estimated magnitudes and confidence.
- 03
Inspect the residual
The unexplained residual reveals what the available signals cannot account for.
- 04
Review summary statistics
MAE and dominant error direction summarise the period.
- 05
Feed back into forecast tuning
Consistent cloud or dust bias informs retraining and recalibration.
HOW INFORMATION BECOMES USEFUL
Examples of decisions it can inform.
These are screening patterns, not automatic instructions. Operational thresholds and local advisories remain authoritative.
Large negative error on a clear day
- Signal
- Low cloud cover but elevated AOD and dust risk.
- Useful action
- Investigate soiling and aerosol attenuation. Schedule panel cleaning.
- Check before acting
- Confirm with on-site irradiance sensor before dispatch.
Persistent morning overforecast
- Signal
- Cloud cover systematically underestimated at sunrise.
- Useful action
- Apply a dawn cloud correction factor to the forecast model.
- Check before acting
- Check whether the error is seasonal or year-round.
Residual dominates despite full inputs
- Signal
- Cloud, dust and temperature explain < 30% of error.
- Useful action
- Investigate inverter curtailment, soiling, or tracking failure.
- Check before acting
- Attribution is atmospheric — mechanical issues require plant data.
FOR PRODUCT TEAMS
A transparent API contract.
The response exposes gaps and limitations so an application can explain confidence instead of hiding it.
/api/v1/energy/middle-east/solar-error-attributionExample request
{
"latitude": 24.29,
"longitude": 54.61,
"hours": [{
"utc_hour": "2026-10-02T08:00:00Z",
"forecast_kwh_per_kwp": 0.72,
"observed_kwh_per_kwp": 0.61,
"cloud_cover_percent": 18,
"aerosol_optical_depth": 0.31
}]
}decision responseExample response
{
"summary": {
"mean_absolute_error_kwh_per_kwp": 0.11,
"dominant_error_direction": "overforecast"
},
"hourly": [{
"error_kwh_per_kwp": -0.11,
"attribution": [
{ "factor": "dust_aerosol",
"estimated_contribution_kwh_per_kwp": -0.07,
"confidence": "satellite-derived" }
],
"data_confidence": "medium"
}]
}CAPABILITY BOUNDARIES
Useful now. Honest about what comes next.
Supported now
- Per-hour cloud cover attribution
- Aerosol / dust attenuation estimation
- Panel temperature efficiency correction
- Residual unexplained component
- Mean absolute error summary
Additional data required
- Calibrated ground irradiance sensor (pyrheliometer)
- Panel soiling and shading metadata
- Inverter SCADA telemetry
- Calibrated satellite AOD product (Copernicus CAMS credentialed)
BUILD RESPONSIBLY
Use the result to focus attention—not replace field judgment.
Pair AtmosGrid’s regional weather context with local irradiance sensors, site telemetry and operator review.