Satellite-derived triggers for anticipatory humanitarian action

Anticipatory action frameworks release pre-positioned humanitarian funds before a disaster peaks, using objective satellite and model-derived triggers. This page explains the data inputs, trigger logic, and honest uncertainty limits that any government or humanitarian agency must understand before committing to the approach.

Sensors

Sentinel-1 IW (C-band SAR): 5 x 20 m ground range resolution in Interferometric Wide swath mode, 250 km swath, 6-day repeat at the equator (3-day with both satellites). Penetrates cloud and operates at night, making it the primary source for near-real-time flood inundation extent used by Copernicus Emergency Management Service rapid mapping.
CHIRPS (Climate Hazards Group InfraRed Precipitation with Station data): Quasi-global (50°S–50°N) daily and pentadal rainfall estimates at 0.05° (~5 km) resolution, blending USGS/UCSB thermal infrared cold-cloud-duration proxies with rain-gauge records. Latency is roughly 3 weeks for the final product; a preliminary version is available within 2 days. Primary input to drought indices and SPI calculations used in slow-onset triggers.
GPM IMERG (Integrated Multi-satellitE Retrievals for GPM): Half-hourly global precipitation at 0.1° (~10 km) with a late-run latency of roughly 12 hours and an early-run latency of ~4 hours. Combines passive microwave retrievals from the GPM constellation with infrared estimates. Used for rapid-onset flood triggers where CHIRPS lag is too long.
GloFAS (Global Flood Awareness System, Copernicus EMS): ECMWF ensemble river discharge forecasts at 0.1° resolution, issued daily with a 30-day probabilistic horizon. Exceedance probabilities at return-period thresholds (2-year, 5-year, 20-year) are the direct trigger variable in most WFP and IFRC anticipatory action protocols. Not a satellite product itself, but driven by satellite-derived precipitation and reanalysis.
ECMWF ENS (Ensemble Prediction System): 51-member global ensemble at ~18 km resolution, providing tropical cyclone track and intensity probabilities up to 15 days ahead. Strike-probability maps at 48- and 72-hour horizons are used as cyclone anticipatory action triggers by several national Red Cross societies.

Why the trigger must fire before the peak

Anticipatory action is a specific funding mechanism, not a general preparedness concept. The logic is contractual: a government or humanitarian agency pre-commits funds and pre-positions supplies against a defined trigger condition. When satellite or model data cross that threshold, the money moves automatically, without a declaration of emergency. The evidence base, summarised in evaluations of the WFP CREW programme and the IFRC Forecast-based Financing scheme, consistently shows that pre-crisis cash transfers and pre-positioned stocks reduce harm more cost-effectively than equivalent post-event spending.

The satellite data problem is timing. A flood's peak discharge may last 24 to 72 hours. Supply chains into flood-prone areas close days before the peak. So a trigger based on observed inundation extent, which is what SAR flood mapping delivers, is almost always too late to move supplies. The trigger must instead fire on a forecast probability, accepting that the disaster may not materialise at the predicted severity. This is the fundamental design tension the data architecture must resolve.

The GloFAS–ECMWF trigger chain

The dominant operational approach for river-flood triggers combines ECMWF ensemble precipitation forecasts with GloFAS hydrological routing. GloFAS converts probabilistic rainfall into probabilistic river discharge, then compares that discharge to historical return-period thresholds derived from ERA5 reanalysis. A typical trigger condition reads: at least 30% of ensemble members forecast discharge exceeding the 1-in-5-year return period at a named gauge location, sustained over two consecutive model runs 24 hours apart. The two-run persistence requirement reduces false triggers caused by single-run outliers.

GloFAS version 4, operational since 2023, runs at 0.05° resolution globally and has demonstrated useful skill out to roughly 15 days for large river basins. Skill degrades sharply for small catchments (under roughly 1,000 km²) and in data-sparse regions where the reanalysis climatology is poorly constrained. Any government adopting GloFAS-based triggers for a small or ungauged river should validate the return-period thresholds against local gauge records before signing a trigger protocol.

CHIRPS and slow-onset drought triggers

Drought triggers operate on a different timescale. The standard approach uses the Standardised Precipitation Index (SPI) or the Standardised Precipitation-Evapotranspiration Index (SPEI) calculated from CHIRPS accumulations over 1-, 3-, or 6-month windows. A trigger might specify: SPI-3 below minus 1.5 across more than 40% of a defined livelihood zone for two consecutive dekads. The threshold percentages and zone boundaries are negotiated between the humanitarian agency and the government before the season begins, using historical CHIRPS data to calibrate the exceedance probability.

CHIRPS has a known wet bias over complex terrain and tends to underestimate orographic rainfall in the Andes, Ethiopian Highlands, and similar environments. Its 0.05° grid also smooths sub-grid convective events. For pastoral drought monitoring in the Horn of Africa, where the CHIRPS record back to 1981 is the primary asset, these biases are reasonably well characterised. For new geographies, independent validation against station data is not optional.

What Sentinel-1 contributes once the trigger fires

Sentinel-1 SAR flood products do not typically drive the trigger itself, but they perform two critical post-trigger functions. First, they validate whether the forecast event materialised, which is essential for learning and for recalibrating trigger thresholds over time. Second, they support operational targeting: once funds are released, knowing which specific villages are inundated, and which roads remain passable, determines where cash-transfer agents can physically reach beneficiaries.

Copernicus Emergency Management Service can deliver a Sentinel-1-derived flood extent map within 12 to 24 hours of a tasking request, at 20 m resolution. The product distinguishes open water from likely flooded vegetation using a change-detection approach relative to a reference SAR acquisition. It does not detect inundation under dense forest canopy, and it struggles in urban areas where double-bounce returns from buildings mimic dry-land backscatter. These are well-documented limits, not caveats to be buried.

False triggers: the cost of acting on probability

A false trigger, where funds are released but the disaster does not reach the forecast severity, is not a system failure. It is an expected statistical outcome of acting on probabilistic forecasts. At a 30% ensemble threshold, roughly 70% of triggers should, by design, not be followed by an event exceeding the return period. The humanitarian community has largely accepted this framing, treating false-trigger costs as the price of early action. But governments signing trigger protocols need to understand this explicitly, in writing, before the first activation.

The more dangerous failure mode is a missed trigger: the disaster occurs but the forecast did not cross the threshold. GloFAS missed triggers are most common for flash floods driven by localised convective rainfall that the 0.1° ensemble grid cannot resolve, and for cyclone-induced surge events where the hydrological model does not include coastal backwater effects. Cyclone triggers based on ECMWF strike-probability maps address the wind and rainfall hazard but not the surge component, which requires a separate coastal inundation model. Designing a trigger protocol without acknowledging these gaps produces false confidence.

Building a trigger protocol that holds up

A trigger protocol is a legal document as much as a technical one. The satellite and model products are inputs to a threshold definition, and that definition must be specific enough to be unambiguous on the day it activates. Vague language about 'significant flood risk' is not a trigger; a named GloFAS station, a specific return-period exceedance probability, and a persistence rule are.

Satellize supports governments developing these frameworks by running historical back-tests across the CHIRPS archive and GloFAS reforecast datasets, quantifying how often a proposed trigger would have fired historically and whether those activations correspond to documented disaster events. The same analytical infrastructure underpins the Tonga crop-estimation programme, where satellite-derived indices are validated against ground truth before any operational decision depends on them. The principle is identical: the threshold must be calibrated, not assumed. Agencies wanting to commission a back-test for a specific river basin or livelihood zone can request a scoping call with Satellize's analytics team.

Typical figures

Sentinel-1 IW spatial resolution	5 × 20 m (range × azimuth); flood products typically resampled to 20 m
Sentinel-1 revisit (dual-satellite)	6 days at equator; 1–3 days at mid-latitudes due to orbit overlap
GloFAS forecast horizon	Up to 30 days probabilistic; useful skill ~15 days for large basins
GloFAS spatial resolution	0.05° (~5 km) for version 4; discharge routed on river network
CHIRPS resolution and latency	0.05° (~5 km); preliminary product ~2 days; final product ~3 weeks
GPM IMERG latency	Early run ~4 hours; late run ~12 hours; final run ~3.5 months; 0.1° resolution
ECMWF ENS ensemble size	51 members; ~18 km resolution; tropical cyclone track guidance to 15 days
Copernicus EMS flood map delivery	12–24 hours from tasking activation; Sentinel-1 based; GeoTIFF and vector
CHIRPS archive depth	1981 to present; sufficient for SPI climatology at 30+ year baseline
Minimum detectable inundation (Sentinel-1)	Open water patches above ~1 ha reliably detected; sub-canopy and urban flood not detected

Analytics Satellize can run

GloFAS trigger monitoring dashboard	Automated daily ingestion of GloFAS ensemble discharge forecasts; exceedance probability computed against user-defined return-period thresholds at named gauge locations	Daily alert feed with activation/no-activation status per trigger zone; JSON and email notification
CHIRPS-based SPI/SPEI drought index	Rolling SPI-1, SPI-3, SPI-6 calculated from CHIRPS dekadal accumulations over defined livelihood zones; SPEI adds temperature-driven evapotranspiration via ERA5	Dekadal GIS layer and PDF bulletin with zone-level index values and historical percentile context
Trigger back-test report	Historical simulation of proposed trigger logic against CHIRPS archive (1981–present) and GloFAS reforecast dataset; confusion matrix of activations vs. documented disaster events	Written report with activation frequency, false-trigger rate, missed-trigger rate, and recommended threshold adjustments
Sentinel-1 post-trigger flood extent validation	Change detection between reference and event-date Sentinel-1 IW acquisitions; thresholding on backscatter reduction; open-water and flooded-vegetation classes	GeoTIFF flood extent map with confidence classification; comparison table against GloFAS forecast extent
Cyclone strike-probability trigger monitor	Automated ingestion of ECMWF ENS tropical cyclone track ensemble; strike probability computed for user-defined administrative boundaries at 48- and 72-hour horizons	Twice-daily alert with probability values and ensemble spread visualisation; activates protocol notification if threshold exceeded
GPM IMERG rapid rainfall accumulation alert	72-hour rolling accumulation from GPM IMERG late run; comparison against CHIRPS climatological percentiles for the same season and location	Near-real-time alert (sub-24-hour latency) flagging accumulations exceeding the 90th or 95th historical percentile; GIS polygon and API push

Sources

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.