Anticipatory action trigger design: from a forecast to a decision

An anticipatory action trigger is a pre-agreed rule that turns a forecast into a decision: when a named forecast shows a named hazard passing an agreed level, with enough probability and enough lead time, funds are released and planned actions start. Designing one means choosing the forecast, the thresholds and the lead time, testing them on past years to know the hit and false-alarm rates, and agreeing all of it with the people who will act before the season starts.

Why anticipatory action needs a trigger at all

Humanitarian money has traditionally moved after a disaster, once needs are assessed. Forecast-based financing turns that round: it matches threshold forecast probabilities with appropriate actions, releases funding when a threshold forecast is issued, and writes standard operating procedures that include a mandate to act [1]. The trigger is what makes that mandate objective. Without it, every forecast becomes a fresh negotiation, and the window for acting early closes while people decide.

That is also why the trigger has to be agreed in advance. A threshold chosen after seeing this season's forecast is no longer a trigger; it is a judgement call, and it loses the speed that justified the approach.

What a forecast-based financing trigger is made of

Most triggers combine five elements:

• The forecast or observation it reads: a river-flow forecast, a rainfall outlook, a drought index, a water level at a named station.
• A magnitude threshold: a danger level, or a return period such as the 1-in-5-year flood. The return period sets how often, on average, the trigger is expected to fire, and so how often money is spent.
• A probability threshold: for ensemble forecasts, the share of members that must pass the magnitude threshold.
• A lead time: how many days before the expected peak the forecast must show it.
• An area: where the hazard must be forecast for the trigger to count, and where the actions will happen.

The Red Cross Red Crescent Climate Centre describes triggers in these terms: forecasts must exceed pre-agreed danger levels and probability thresholds, and the forecast is chosen from an inventory of candidates by verification, lead time, type and reliability [2].

Readiness and action thresholds

A single trigger forces a hard choice between lead time and certainty. Many programmes split it in two. A readiness threshold, read from a longer-range forecast, starts low-cost steps: staff on standby, suppliers alerted, cash lists checked. An action (activation) threshold, read from a shorter-range and more certain forecast, releases the main funds and starts distribution.

The 2020 monsoon in Bangladesh is a well-documented example. On 4 July, a GloFAS 10-day probabilistic forecast gave the targeted areas a greater than 50 percent chance of a severe flood between 14 and 16 July, which met the pre-activation trigger, and the UN's Central Emergency Response Fund released $5.2 million to three agencies to prepare. Less than a week later the national Flood Forecasting and Warning Centre forecast the Bahadurabad water level crossing the government's danger level, which met the activation trigger. Aid ultimately reached about 200,000 people. Both forecasts had been checked against flood records back to 1987 so that activation would happen only for severe-impact events [3].

The early action protocol trigger

In the Red Cross Red Crescent Movement, the trigger is written into an Early Action Protocol, which sets out in advance the trigger, where action will happen and who does what [2]. The simplified version piloted from 2022 gives a sense of the scale and timing involved. A simplified EAP can draw up to CHF 200,000 from the IFRC's Disaster Response Emergency Fund for readiness, pre-positioning and early action; readiness and pre-positioning together may take at most 65 % of the budget; early-action funds are disbursed when the trigger is met; and once approved, the protocol is valid for two years [4]. The same figures still apply under the 2026 DREF guidelines, where the CHF 200,000 ceiling includes the IFRC's 6.5 % support charge; a full EAP can draw up to CHF 900,000 and runs for five years [5].

The design point for the trigger is that the money behind it is fixed and limited. A trigger that fires too often uses up the readiness budget and the trust of partners; one that fires too rarely leaves a protocol on the shelf for its whole two-year life.

Lead time against skill

Every forecast loses skill as lead time grows. The trigger designer's job is to find the longest lead time at which the forecast still separates real events from non-events well enough for the action planned. Cash transfers that take a week to deliver need a week of lead time; moving livestock to higher ground may need only two days. So the actions come first: list them, note how long each takes, and only then pick the forecast and lead time that can support them.

Hit rate and false alarm rate

Back-test the candidate trigger on every past season for which both the forecast (or a reanalysis standing in for it) and a record of what happened exist. Each season is a hit, a miss, a false alarm or a correct quiet season. Three measures come out of that table [6]:

• Hit rate (probability of detection): the share of real events the trigger caught.
• False alarm rate: the share of quiet seasons in which it fired anyway.
• False alarm ratio: the share of activations that were not followed by the event. Because events are rare, this can be high even when the false alarm rate looks low.

No threshold removes false alarms. The question to settle with partners is how many are acceptable, given what the early action costs and what it saves when the event does come. Agreeing that number before the season, in writing, protects the team that acts on a forecast that turns out wrong. Acting on a good trigger that then misses is part of the design, not a failure of it.

A record of what happened is as important as the forecast. River gauges, disaster databases, satellite flood maps and local knowledge all help; where the record is thin, say so, and treat the hit rate as uncertain.

Drought triggers

A drought trigger usually reads a season: a seasonal rainfall forecast, an accumulated rainfall deficit, a vegetation index or soil moisture, checked at fixed points in the season. Lead times are weeks to months rather than days, and the actions differ: drought-tolerant seed, water trucking, livestock feed, early cash. The same design steps apply: pick the actions, set the thresholds, back-test, agree.

Season totals can also carry flood signals. In KAI's open test in Spain's Guadalquivir basin, the October–April rainfall total ranked flood years as well as the river model did, because floods there follow wet winters. A season total marks a risky year, not the flood itself, so it suits seasonal or anticipatory finance better than event cover. That case is told in full in the Guadalquivir flood trigger case.

Running the trigger in the season

A trigger is only useful if someone watches it. In the season that means a named forecaster checking the agreed forecast on an agreed schedule, a clear message when the readiness or action threshold is met, and a log of each check. After the season, verify: did the trigger fire, did the event happen, how much lead time was there in practice? Those results feed the next review of the thresholds.

What KAI offers

KAI designs and back-tests flood and drought triggers on historical data, documents the hit and false-alarm rates, and runs the forecast that fires them in the season, so the protocol's trigger and the forecast are built by the same team. For flood triggers this is a KAI-Hydro study on the GloFAS v5 reanalysis from 1980 to today, with the trigger written four ways (fixed flow, percentile, return period, share of a reference flood) and back-tested year by year. KAI recommends calibrating on gauges, satellite flood maps or local records, because a global model alone carries too much basis risk. The Anticipatory Action plan adds several basins and partner systems, a duty forecaster in the rainy season and support for anticipatory-action protocols. Details are on the anticipatory action triggers page; daily river forecasts with Normal, Watch and Warning levels are on the flood early warning page.

For insurance-style triggers, see the guide to parametric flood trigger design.