Learning guide 5 · Practical interpretation
How to read the Contrail Research map
The map puts observations and hypotheses in one view. Its value depends on keeping them separate: NOAA imagery is observed context, aircraft symbols come from tracking providers, and DeepWave trails/cloud cells are model output.
Before interpreting anything, check four things
- Time: Is the view live or playback? Note UTC, not only local time.
- Flight source: Does the status say OpenSky, Airplanes.live, holding, or stored? Those modes have different freshness and coverage.
- Layer state: Confirm which buttons are on. Satellite, radar, and model output can overlap visually.
- Map scale: Regional imagery and weather grids become coarse at street-level zoom. Apparent precision does not increase just because the basemap does.
The safest reading order: observed sky → observed satellite/radar → flight context → pressure-level weather → model hypothesis. Starting with the model invites confirmation bias.
Layer-by-layer guide
| Layer | What it is | What it is not | Best use |
|---|---|---|---|
| Basemap imagery | Reference map imagery, labels, boundaries, and place names. | Not a live image of the current ground or sky. | Location and geographic orientation. |
| Aircraft markers | Provider-reported positions and attributes at a recent timestamp. | Not a complete census or a continuously measured path. | Find candidate traffic and inspect altitude/time. |
| White trail lines | Rendered track segments where the experimental model found supportive inputs. | Not a satellite-detected cloud and not proof the aircraft produced a visible trail. | Prioritize track segments for comparison with observations. |
| NOAA Satellite Clouds | Regional nowCOAST geostationary satellite imagery. | Not a contrail classification; at local zoom it may be blocky or intentionally faint/hidden. | Observe regional cloud bands and changes. |
| NOAA Radar | NWS/nowCOAST base-reflectivity mosaic for precipitation context. | Not a general cloud image and usually not a contrail detector. | Identify rain/storm context and exclude precipitation features. |
| DeepWave Model cells | Experimental model-generated hypotheses. | Not observed cloud boundaries, chemical measurements, or calibrated probabilities. | Compare the model with independent imagery and field observations. |
| SkyPulse verification | User-submitted, timestamp/location-bound observation context when available. | Not automatically expert-verified or sufficient for flight attribution. | Add ground truth leads and document disagreements. |
How to read the status ribbons
The lower ribbons summarize weather and activity for the current view. They are compact interface cues, not standalone scientific conclusions.
- Atmosphere / gate
- A relative model support summary. Check the underlying source and pressure-level values before citing it.
- Wind
- Direction and speed used to orient or spread the hypothesis layer. It may represent a selected pressure level, not surface wind.
- Cloud state
- A relative summary of the modeled cloud layer currently displayed.
- Cruise activity
- Aircraft and rendered segment counts after current filters. Segment count is a software/display quantity, not the number of observed contrails.
- Provider label
- The source and freshness mode for visible aircraft. “Holding” means a recent cached frame is retained during a delayed refresh.
Experimental marine, infrastructure, or population-context ribbons do not establish contrail composition or causation. Treat them as contextual overlays only; temperature and humidity aloft remain the necessary atmospheric evidence.
A five-minute evidence workflow
Anchor the observation
Write down UTC time, location, camera direction, and visible facts. Keep the original image or video.
Toggle the model off
Inspect satellite and radar context independently. Ask whether the feature is actually visible in an observational layer and whether scale/resolution permits a match.
Inspect candidate flights
Check timestamp, altitude, heading, and track continuity. Search an earlier window to allow for drift. Remember that missing provider data are common.
Restore the model
See whether supportive trail segments and cloud cells overlap the observation. Treat agreement as one additional line of evidence, not confirmation.
Try to disprove the match
Test another pressure level, natural cirrus, a different candidate flight, timing error, and wind uncertainty. Record the result at the strength supported.
What it means when a layer is empty
| What is missing? | Possible explanation | Next check |
|---|---|---|
| Aircraft | No traffic, provider coverage gap, rate limit, filters, network failure, or playback window with no records. | Read provider/status label; widen time/area carefully; never infer “no aircraft” from one feed alone. |
| Model trails | Unsupported or missing weather, insufficient flight context, filters, or unavailable source data. | Inspect the source status and compare a nearby time. |
| Modeled clouds | No current model output or an unavailable source. | Check the layer status and compare independent observations. |
| Satellite texture | Clear sky, unavailable tile, local zoom suppression, scan/product delay, or cloud below display contrast. | Zoom regionally, check timestamp, then consult an independent official viewer. |
| Radar | No precipitation, location outside mosaic, or tile failure. | Radar absence says little about high ice cloud; use satellite/weather instead. |
Absence of a model mark is not proof of absence in the sky. A model can miss because the phenomenon is smaller than its grid, a source is incomplete, or the necessary variable is poorly represented.
How to write a finding someone else can audit
A concise but useful record includes:
- observation UTC time/range, approximate coordinates, direction, and original media identifier;
- visible description before interpretation;
- map mode, flight provider, retrieval time, filters, and candidate identifiers;
- weather source, valid hour, pressure level, temperature, humidity, wind, and missing values;
- satellite/radar source, scan time, product/channel, and whether the feature was resolved;
- the model layer displayed and its source time;
- leading explanation, strongest alternative, and the evidence that separates them;
- conclusion label: observed, supported, consistent, unresolved, or contradicted.
For interpretation guidance, see the methodology.