Sentinel-1 flood monitoring: a defensible workflow for Colombia
How to use Sentinel-1 radar to prioritise flood verification, document limitations and avoid presenting a satellite alert as a damage assessment.
Sentinel-1 can help locate changes consistent with flooding when cloud blocks optical imagery, but it cannot by itself declare damage, water depth or an area safe. Its strongest operational role is to prioritise verification, combine evidence and make clear what was observed and what remains uncertain.
The Sentinel-1 mission uses C-band synthetic aperture radar and can acquire imagery by day, at night and through cloud. Open, smooth water commonly returns low backscatter, so comparing a pre-event observation with an event observation can flag candidate areas. Wind, vegetation, terrain geometry and urban surfaces alter that signal; a dark pixel is not always new water.
GeoSAT can design this workflow within GIS for risk management or satellite monitoring. Before automating it, review the data a municipality needs for GIS-based risk management.
Define the decision before processing imagery
The question should not be “can I produce a flood map?” It should be a decision such as prioritising road inspections, checking community reports, reviewing access to infrastructure or communicating a preliminary indicator to an incident team.
For each decision, set the area of interest, accountable owner, time window, audience and necessary level of evidence. A prioritisation alert can retain visible uncertainty; action affecting people or critical infrastructure requires additional verification and the procedures of the responsible authority.
Build a comparable baseline
An after-event scene without a reference can confuse permanent water, wet soils, radar shadow and cover change with flooding. Assemble in advance:
- Prior Sentinel-1 observations, with comparable acquisition geometry where possible.
- River network, permanent water bodies and wetlands, each with source and cutoff date.
- A digital elevation model and a mask of areas with known radar limitations due to relief.
- Exposure layers: roads, facilities, installations or relevant administrative units.
- Field reports, gauges, photographs and alerts, each with time, location, source and confidence level.
The aim is not to construct a single “truth.” It is to know what the signal was compared against and which evidence can confirm or refute it.
A workflow that leaves an audit trail
- Bound the event. Record date, area, working hypothesis and owners. Do not process an entire department when the decision concerns one catchment or corridor.
- Select comparable scenes. Document product, relative orbit, polarisation, date and acquisition conditions. Do not mix observations without retaining that difference.
- Prepare the data. Apply appropriate processing for comparable backscatter and document the software, parameters, elevation model and versions used.
- Generate candidates. Compare pre-event and event observations, remove permanent water where appropriate, and label output as candidate extent rather than final truth.
- Add context. Review slope, land cover, river network, available optical imagery and field reports. Optical data may confirm detail when skies are clear, but it is not required for radar to work through cloud.
- Validate and publish with limits. Separate confirmed, awaiting verification and not assessable. Deliver image date, processing time, method, owner and a link to the evidence.
Copernicus Data Space documentation provides an example that compares before-and-after observations to identify newly flooded areas. It is a technical reference, not a universal recipe for every Colombian river, land cover or city.
Put the limitations in the layer, not a hidden note
Open, calm water commonly has low backscatter. However, wind can increase the signal; vegetation can hide water beneath a canopy; and double-bounce scattering from urban structures can be hard to classify. On slopes, radar shadow and geometric distortion can prevent interpretation. The ESA Knowledge Hub summarises these constraints and points to combining sources and validation for the context at hand.
That is why a single threshold rule for the whole territory is fragile. Record non-assessable zones and confidence by area. If the system cannot observe a condition, stating that is more useful than colouring it as flood.
Design the operation, not only the algorithm
A useful workflow needs a trigger —for example, a hydrometeorological alert or local report— people assigned to review output, an escalation route and a policy for correcting the map when new evidence arrives. It also needs to preserve scenes, parameters, versions and review decisions so another person can reproduce the result.
Copernicus Global Flood Monitoring processes Sentinel-1 data using several flood-detection algorithms. It can provide context, but it does not remove the need to assess coverage, date, resolution, availability and relevance for a local decision.
Next step
Start with a historical event and a representative area. Compare output with known observations, identify where it fails and agree which label is used for each level of evidence. Then integrate the workflow into municipal or sector operations.
GeoSAT can structure the baseline, evidence model and prioritisation dashboard. The result should help decide where to verify first, without promising to predict flooding, measure damage or replace field assessment.