Satellite support for search and rescue drift prediction
When a vessel or person goes missing at sea, drift models define the search area. Satellite-derived currents, wind fields, wave height, and spaceborne AIS last-known-position data can sharpen those models and then directly image the probability zones.
Sensors
- Spire Global spaceborne AIS: Provides last-known-position and last-known-heading for the missing vessel. Spire operates over 100 LEO satellites with global AIS refresh typically under 20 minutes for open-ocean tracks, giving the drift model its starting point.
- Sentinel-1 SAR (C-band, 5.405 GHz): 12-day exact repeat at the equator, 6-day with both satellites active. IW mode delivers 10 m ground resolution over 250 km swaths. Used both to derive near-surface wind speed via NRCS backscatter (CMOD5.n geophysical model function) and to directly search probability areas for small targets. Detection of a person in the water remains unreliable below sea states of about 1.5 m Hs; a life raft or vessel debris is more tractable.
- Copernicus Marine Service (CMEMS) altimetry products: Merges data from multiple radar altimeters (Sentinel-6 Michael Freilich, Jason-3, SARAL/AltiKa and others) into gridded absolute dynamic topography and surface geostrophic current fields at roughly 0.25-degree resolution, updated daily. Significant wave height (Hs) fields from the same altimeters feed directly into Leeway model uncertainty bounds.
- VIIRS Day/Night Band (DNB): 750 m resolution, daily global coverage. At night, DNB can detect vessel lights, flares or activated EPIRBs with visible-light emission, extending search capacity into darkness. Sensitivity is roughly 2 × 10⁻⁹ W cm⁻² sr⁻¹, sufficient for a vessel distress flare at moderate range.
- Copernicus Marine Service ocean analysis and forecast (NEMO/IFS): Hourly to 6-hourly surface current and wind stress fields from the global ocean physics analysis, assimilating altimetry, SST and Argo float data. These fields drive the forward and backward drift simulations used to define probability density maps.
Why the first hour of data determines the search area for the next 72
Drift prediction for search and rescue is a compound error problem. Every uncertainty in the object's starting position, the wind field, the surface current, and the object's own aerodynamic and hydrodynamic profile accumulates over time. A 10 percent error in surface current speed translates to roughly 8 nautical miles of positional error after 24 hours in a 2-knot current. That is the difference between a focused, survivable search and an area too large to cover before a person in the water succumbs to hypothermia.
Satellite data attacks this problem at three points: it anchors the starting position through spaceborne AIS, it constrains the forcing fields through altimetry-derived currents and SAR-derived winds, and it can directly image the highest-probability zones once the model has run. None of these inputs is perfect. The honest case for satellite data is not that it solves drift prediction, but that it reduces the dominant error terms in a tractable way.
The Leeway framework and where satellite inputs slot in
The Leeway drift model, developed and published by the US Coast Guard Research and Development Center and adopted widely by national maritime rescue coordination centres, represents a drifting object as the vector sum of ocean current, wind-driven current, and object-specific 'leeway': the downwind and crosswind motion caused by wind acting on the object's above-water profile. Leeway coefficients have been measured empirically for dozens of object types, from life rafts to kayaks to a person in a survival suit. The USCG publishes these coefficients; they are not proprietary.
Satellite altimetry feeds the current field. CMEMS global ocean physics products assimilate altimeter-derived sea surface height to produce geostrophic current estimates. These are then combined with Ekman current components derived from wind stress. The resulting surface velocity field is the primary forcing input. Sentinel-1 SAR wind retrievals, using the CMOD5.n model function applied to C-band normalised radar cross-section, provide spatially detailed near-surface wind speed at 500 m to 1 km effective resolution, resolving mesoscale structures that coarser NWP grids miss. Significant wave height from altimetry enters the model as a proxy for wave-induced Stokes drift, which can add 0.1 to 0.3 knots of net transport in swell-dominated seas. That sounds small. Over 48 hours it shifts the probability centroid by 10 to 15 nautical miles.
Spaceborne AIS as the drift model's anchor
A vessel's last AIS transmission is the starting point for any drift simulation. Terrestrial AIS receivers cover coastal waters to roughly 40 nautical miles offshore; beyond that, spaceborne AIS is the only record. Spire Global's constellation of over 100 LEO satellites provides global AIS coverage with median revisit times that allow open-ocean tracks to be reconstructed to within roughly 20-minute intervals under normal traffic conditions.
The last confirmed position, heading, and speed from spaceborne AIS define the initial probability ellipse for the drift simulation. If the vessel was underway, a dead-reckoning step projects forward to the estimated time of distress before the Leeway model takes over. If AIS was switched off or failed before the distress event, the last known position becomes a much wider prior, and the resulting probability area grows accordingly. This is an honest limit of the method: spaceborne AIS cannot help if the vessel never transmitted or if the gap between the last transmission and the distress event was many hours.
Direct imagery of probability areas: what SAR and VIIRS can and cannot find
Once the drift model produces a probability density map, Sentinel-1 can be tasked to image the highest-probability zones. A 250 km IW swath at 10 m resolution can cover a substantial search area in a single pass. SAR detects objects by their radar cross-section contrast against the sea surface. A rigid life raft or vessel debris produces a detectable return in moderate sea states. A person in the water, wearing no retroreflective material and presenting a cross-section of perhaps 0.01 m², is at or below the detection threshold of C-band SAR in sea states above 1 to 1.5 m Hs. Corner reflectors and SART transponders dramatically improve detection probability, but survivors do not always have them.
VIIRS DNB adds a night-time search capability that SAR alone cannot provide. Distress flares, vessel lights, and activated EPIRBs with visible-light emission can appear in DNB imagery at 750 m resolution. The practical constraint is that a single flare burns for roughly 60 seconds; the probability of a VIIRS overpass coinciding with an active flare is low unless the survivor can be contacted and asked to fire on cue. For persistent light sources such as a vessel fire or a strobe, DNB is considerably more useful.
Optical imagery from Sentinel-2 or commercial sensors is cloud-limited and offers no advantage over SAR in overcast conditions, which are common in the high-latitude and tropical regions where many maritime distress events occur. SAR's all-weather capability is its primary operational advantage in this context.
Latency, coverage, and the operational clock
Search and rescue is acutely time-sensitive. A person in 15°C water has a median survival time of roughly 6 hours before incapacitation from hypothermia. The operational value of any satellite input depends entirely on how quickly it reaches the rescue coordination centre in usable form.
CMEMS ocean current analysis products are updated daily with a latency of around 5 hours from observation cutoff. Sentinel-1 Level-1 products are typically available within 1 hour of acquisition via the Copernicus Data Space. SAR wind field retrieval and current field extraction can be completed in under 30 minutes with pre-built processing pipelines. Spaceborne AIS data from Spire is available via API with latency measured in minutes for recent passes. The bottleneck in most operational settings is not data availability but the integration step: getting heterogeneous satellite products into the drift model's input format and into the hands of the duty officer. Satellize builds and maintains exactly these integration pipelines, drawing on the same open constellation data that coast guards can access independently. The Tonga crop-estimation programme gave us direct experience of low-latency operational delivery in a Pacific island context, where connectivity and data-handling capacity are genuinely constrained.
Revisit is the other operational limit. Sentinel-1's 6-day repeat means that any given location may not be imaged again for up to 6 days without commercial tasking. In a fast-moving search, a single SAR pass over the probability area may be the only imagery available. Commercial SAR constellations such as ICEYE or Capella can reduce this gap to hours, but at cost and with tasking lead times that vary by operator.
Honest limits and what they mean for programme design
Satellite data improves drift prediction; it does not make it precise. The Leeway model's output is a probability density, not a position fix. Even with excellent current and wind forcing, the 72-hour probability ellipse for a drifting life raft can span thousands of square kilometres in open ocean. Satellite imagery cannot search that area exhaustively in a single pass.
The practical design principle is to use satellite data to reduce the search area to something that aircraft and surface vessels can cover within the survival window, not to replace them. Current field resolution at 0.25 degrees (roughly 25 km at mid-latitudes) is adequate for large-scale drift but misses mesoscale eddies and coastal current jets that can deflect a drifting object by tens of miles. Higher-resolution ocean model products exist for some regions through CMEMS regional configurations, but global coverage at fine resolution remains limited. A maritime rescue coordination centre that understands these limits will use satellite data to prioritise search sectors, not to declare sectors clear.
Typical figures
| AIS last-known-position latency (Spire) | Typically minutes via API; open-ocean revisit median under 20 minutes |
| SAR spatial resolution (Sentinel-1 IW mode) | 10 m ground range, 250 km swath |
| SAR revisit (Sentinel-1 two-satellite) | 6 days at equator; shorter at higher latitudes due to orbit geometry |
| Ocean current field resolution (CMEMS altimetry-merged) | 0.25 degree (~25 km); daily update, ~5 h latency |
| SAR wind speed retrieval resolution | Effective 500 m to 1 km after CMOD5.n inversion of Sentinel-1 NRCS |
| VIIRS DNB resolution | 750 m; daily global coverage; detection threshold ~2 × 10⁻⁹ W cm⁻² sr⁻¹ |
| Significant wave height (altimetry) | Along-track ~7 km footprint; merged gridded product at 0.25 degree |
| Minimum detectable SAR target (C-band, moderate sea state) | Life raft or vessel debris: detectable above ~1 m Hs. Person in water: generally below detection threshold above 1.5 m Hs |
| Sentinel-1 archive depth | From October 2014 (Sentinel-1A launch); accessible via Copernicus Data Space |
| Delivery formats | GeoTIFF drift probability rasters, GeoJSON probability polygons, AIS track CSV, API-fed current/wind NetCDF |
Analytics Satellize can run
| Drift probability density map | Leeway model (USCG published coefficients) driven by CMEMS surface currents, SAR-derived winds, and altimetry wave height; Monte Carlo ensemble over initial position and leeway coefficient uncertainty | GeoTIFF raster and GeoJSON polygon set showing 50th, 75th, and 95th percentile search areas, updated at each new satellite forcing input |
| AIS last-known-position reconstruction | Spire spaceborne AIS track assembly with dead-reckoning forward projection to estimated distress time | CSV track file with timestamped positions, heading, speed, and confidence flag; ingests directly into SAROPS or equivalent RCC software |
| SAR-derived surface wind field | CMOD5.n geophysical model function applied to Sentinel-1 IW NRCS; validated against ECMWF ERA5 for bias correction | NetCDF wind speed and direction grid at 1 km resolution, timestamped to SAR acquisition, formatted for direct input to drift model |
| SAR probability-area search product | Sentinel-1 CFAR (Constant False Alarm Rate) ship and object detection applied within the 95th percentile probability polygon; detections cross-checked against AIS to suppress known-vessel clutter | GeoTIFF with candidate target overlays and confidence scores; anomalous returns flagged for analyst review within 90 minutes of SAR acquisition |
| VIIRS night-time light anomaly alert | VIIRS DNB radiance threshold detection within probability zone; background subtracted using multi-day composite to isolate transient sources | Alert message with pixel coordinates, radiance value, and UTC timestamp; issued within 2 hours of VIIRS overpass |
| Backward drift reconstruction | Time-reversed Leeway model run from a found object or debris field to estimate origin point and time; uses archived CMEMS current fields | Probability origin ellipse report with sensitivity analysis showing dependence on current field uncertainty; useful for post-event investigation and for redirecting search if a debris field is found before survivors |
Who does the work
We can get this done for you. Satellize runs its own analyst desk and a strong science team. You do not buy a data feed and work out what it means; our people source the imagery, run the analysis described on this page, and hand you the answer with its confidence limits stated. Discuss this requirement.