Any programme that hardens water infrastructure against cyclones has to answer one question before it can prioritise anything else. Not “which communities look important?” but “how often is a genuinely damaging cyclone expected to strike here, and how many people would it affect?”

SaniTap’s coastal cyclone-resistant safe drinking water (CRSDW) programme runs that calculation on every fokontany in coastal Madagascar. This article sets out how the model works. It is deliberately mechanical: parameters stated in the open, every source named, every filter separately visible, and no discretion applied to the ranking itself. That constraint matters because the programme’s finance route (the Fund for Responding to Loss and Damage and the Adaptation Benefits Mechanism) requires that site selection be reproducible from a public dataset, not proprietary judgement.

The question the model answers

The model answers one question, for every coastal community in Madagascar: how often is a genuinely damaging cyclone expected to strike here, and how many people would it affect? It ranks communities on that basis and works down the list in phases. The same rule is applied to every coast, with no assumption that one side of the island matters more than another.

That last point is not decorative. Programme sequencing driven by donor familiarity, incumbent NGO presence or historical political weighting produces predictably biased outcomes. A model that starts from a country-wide exposure ranking and then applies a small number of transparent filters produces a ranking that survives scrutiny from the Government of Madagascar, from multilateral funders, and from the accredited validator that certifies adaptation benefits under the Adaptation Benefits Mechanism.

The predictor — 46 years of IBTrACS

The model does not attempt to forecast individual future storms. No method can do that reliably beyond about two weeks.^[1] What can be predicted is frequency, and frequency is what matters for infrastructure that has to last decades.

The prediction rests on IBTrACS — the International Best Track Archive for Climate Stewardship — maintained by NOAA and the World Meteorological Organization. It is the standard global record of cyclone tracks and intensities.^[2] SaniTap’s model uses the South-West Indian Ocean basin, every tropical cyclone from 1980 to 2026.

For each of Madagascar’s roughly 17,500 fokontany the model counts the number of distinct severe cyclones whose track passed within 100 km. Severe means the storm reached at least 64 knots of sustained wind — the threshold for a Category-1 hurricane and the point at which cyclones do serious structural damage to water systems.

The logic of using 46 years of history as a forecast is straightforward. The regions where cyclones form, and the steering winds that carry them onto Madagascar, are stable features of the climate. They do not move from decade to decade. A fokontany struck by six severe cyclones in 46 years sits in a corridor with a severe-cyclone return period of roughly one in eight years, and that is the best available estimate of its future exposure.

Track density in the past is the most reliable predictor of track density in the future. Two things reinforce this. First, SaniTap’s own record of 2023-26 damage falls on exactly the historical hotspots — evidence that the climatology is still current. Second, the climate signal runs one way: warmer seas raise the share of cyclones that reach severe intensity, so the historical count is best read as a floor, not a central estimate.^[3]

Five filters

Cyclone frequency alone does not site a water system. Five filters shape the working list.

1. Coastal scope. A community must lie within 50 km of the nearest shore. Cyclone damage to water is a coastal effect: storm surge, salt contamination of wells, wind damage, and the flooding that follows landfall. The 50 km rule is applied to every coast equally, with no side of the island favoured.

2. Repeat-strike threshold. A community enters the ranking only if it has been hit by at least three severe cyclones since 1980, a return period under about fifteen years. This restricts the programme to places genuinely inside the repeat-strike zone rather than places that saw one unusual event decades ago.

3. Population as need. Priority is population multiplied by exposure, using the 2018 national census at fokontany level. A high-strike corridor with more people at risk ranks above an equally exposed but emptier one. This turns a physical exposure measure into a human need measure.

4. City-core exclusion. City cores with municipal supply are excluded from the start. The programme does not compete with existing utility networks.

5. One system per fokontany. The programme places one cyclone-resistant system per community. This constrains the total system count and forces the ranking to work across the coast rather than doubling up in the highest-exposure clusters.

After the five filters, the field of eligible fokontany is 3,383. That is the pool from which the ranked list is drawn.

Phase 1 is different — loss-and-damage vs adaptation

The programme is sequenced so that the first phase answers for damage that has happened and the later phases answer for damage expected to happen. The two are different claims and are funded on different grounds.

Phase 1 — the loss-and-damage phase

Two conditions govern a Phase 1 slot. First, the community must lie in a district with documented damage from the 2026 cyclones — the two Category-3 landfalls, Fytia (31 January 2026, north-west) and Gezani (10 February 2026, Toamasina) — because this phase responds to the loss and damage of the 2026 season specifically. Eleven districts qualify.

The wider damage register — 51 districts across 2023-26, assembled from BNGRC, UN OCHA and ReliefWeb, the IFRC and Malagasy Red Cross, UNICEF, ACAPS and ECHO, with a source URL and a report date behind every figure — stays attached to every site as evidence but is not itself a Phase 1 selector.

Second, the community must lie inside one of the Government of Madagascar’s five priority regions for this programme: Atsinanana, Analanjirofo, Boeny, Analamanga and Melaky, as communicated by MEAH (Ministère de l’Eau, de l’Assainissement et de l’Hygiène). In practice every 2026-damaged coastal district already sits inside them.

Loss-and-damage finance answers for harm that has already occurred, so the first phase is evidenced by harm rather than by modelling, and it is built where the Government has directed.

A high-ranking community that fails either condition is not excluded from the programme; it simply cannot occupy a Phase 1 slot, so it moves to Phase 2 and the next qualifying community moves up. On the current ranking, 65 of the top-81 sites moved out of Phase 1 on this basis: Fenerive Est (14), Mahanoro (12), Antalaha (11), Sambava (9), Mananjary (6), Nosy-Varika (4), Vohemar (3), Vavatenina (3), and one each in Ambilobe, Soanierana Ivongo and Andapa.

Phase 1 then holds 81 systems serving 214,657 people on the two 2026 landfall coasts: Toamasina II (44 systems), Brickaville (26), Mahajanga II (4), Mitsinjo (4) and Soalala (3).

Phases 2 and above — the forward-looking phases

Phases 2 and 3 are taken in strict rank order on predicted exposure. Whether or not a community was hit in the last three years is irrelevant to these phases; the question is only how likely is it to be hit next? That is the adaptation case rather than the loss-and-damage case, which is why it sits behind Phase 1 in sequence.

The predictive basis is the same 46 years of IBTrACS track density, not a forecast of individual storms.

Refining the raw ranking — mWater layers and operator filter

Two live datasets from the national mWater inventory refine the raw ranking, and a third is used as evidence.^[4]

Water availability

A layer built from 79,817 mapped water points lifts under-served communities and lowers well-served ones. It measures the improved-water points available per person. This is coverage, not working status.

Operators already present, and the backfill

Where a utility or a private operator already runs piped water, a cyclone-resistant system is not needed, so those sites are removed. SaniTap identifies them two ways: from the mWater register of 3,040 piped schemes, and from a public review of named operators (Jirama and the private and NGO operators). Twenty-three programme sites were removed on this basis, mostly Jirama town supplies at Mananjary and Morondava.

Crucially, the ranking then closes up. Removing six sites from Phase 1 pulls the next six up from Phase 2, which pulls six up from Phase 3, and so on down the list. The shortfall accumulates as it descends, so 6, then 10, 11, 16, 18 and finally 23 sites move up by Phase 6, and 23 new communities enter the tail. Every phase stays at its full size, and no priority is lost.

The raw ranking before the operator filter is kept for reference.

Confirmed broken water, as evidence

A GPS-level field survey records the state each water point was actually found in. The current export holds 3,381 surveyed points with a functional status and coordinates: 2,145 functional, and 1,236 not working (1,051 not functional, 79 partially functional, 106 no longer exist).

Coverage is concentrated in Atsinanana (2,552 points) with a further block in Analamanga (788), so it cannot be used to rank the country without biasing the result toward the surveyed regions. SaniTap therefore uses it strictly as corroborating evidence, shown on the map as its own layer. Of the current 241 programme sites, 113 lie in surveyed regions; 33 have a confirmed broken or partial water point in their own fokontany and 55 in their commune. For those sites the case is no longer that few water points are mapped nearby, but that the standpipes there are confirmed dead.

Everywhere else, an absence of survey data means the place was not surveyed, not that its water works.

Sites the Government has already financed

MEAH holds a list of post-cyclone rehabilitation projects that are already financed: thirteen schemes, of which nine sit inside the five priority regions (the other four, in Ambatondrazaka, Alaotra Mangoro, are outside the programme scope and are neither shown nor excluding). The nine do not need FRLD funding, so they are shown on the map as their own layer (MEAH — already financed, not included) and every candidate community within 5 km of one is removed from the ranking — fifteen candidates dropped on the current list. The ranking closes up behind them in rank order, so every phase stays at its full size.

Why Anosy is not on the cyclone list

Anosy — the far south-east around Fort-Dauphin (Taolagnaro) — does not appear in the cyclone programme. This is a deliberate, evidenced result, not an oversight.

Anosy is not spared cyclones. On average 13.6 tropical systems have passed within 100 km of each Anosy community since 1980, as many as its neighbours further west. The difference is intensity. By the time storms curve down to Anosy’s latitude, around 25 degrees south, they have weakened. In 46 years of record, the strongest wind ever measured within 100 km of any Anosy community is 55 knots — below the 64-knot severe threshold. Neighbouring Androy catches the same storms while they are still at 64 to 80 knots, which is why Androy qualifies and Anosy does not.

The model is reading the physics correctly.

Anosy’s water problem is real, but it is a different problem. Anosy and Androy form the heart of the Grand Sud drought belt, the region of chronic water scarcity and recurrent kere (famine). The threat to water here is not a cyclone tearing out a standpipe. It is rain that does not come. The adaptation response is therefore different: deep boreholes into reliable aquifers, rainwater harvesting and storage, solar-powered pumping, and demand management. It does not need cyclone hardening.

The recommendation, made openly in the CRSDW methodology note, is to address Anosy and the wider Grand Sud through a separate drought-adaptation programme, not by relaxing the cyclone-list criteria to include it. Doing the latter would corrupt the ranking; doing the former honestly reflects that Anosy’s people need a different intervention type.

Auditability and what the model produces

The model produces a ranked list of coastal communities drawn from 3,383 eligible fokontany, taken in phases of 80 systems (Phase 1 carries 81). The current proposed programme is three phases: 241 systems serving 898,292 people.

Phase 1 (81 systems, 214,657 people) is the FRLD loss-and-damage phase, responding to the 2026 cyclones, restricted to districts damaged by Fytia and Gezani and inside the Government’s five priority regions. Phases 2 and 3 are ranked on predicted exposure and may lie outside those regions.

The ranked list continues beyond Phase 3, so the programme can extend in the same rank order as further funding arrives without any re-scoring.

Every figure in the model traces back to a named public source:

• IBTrACS for cyclone exposure.
• The 2018 national census for population, at fokontany level.
• mWater for water points and broken-water status.
• MEAH for the priority regions and the financed-project list.
• BNGRC, UN OCHA and ReliefWeb, IFRC and Malagasy Red Cross, UNICEF, ACAPS and ECHO for the 2023-26 damage register, with a source URL and a report date behind every figure.

The parameters — the 100 km radius, the 64-knot threshold, the 50 km coastal scope and the three-storm cut-off — are stated openly. They can be changed, at which point the list re-sorts. Nothing is hand-picked, and no site is added to or removed from the ranking except through a parameter change or an explicit filter (operator present, MEAH-financed within 5 km).

A note on map positions. A site marker is placed at the centre of its community as recorded in the census geography, then snapped toward the nearest cluster of existing water points where one lies close by. It marks the community to be served, not the exact spot where a system would be built; siting within the community is an implementation decision.

The ranked list, the parameters and the source layers are the auditable object that a validator or a funder can examine to confirm that the programme’s priority-setting is reproducible from the public data.

Footnotes

---

1. Forecast horizons for individual tropical cyclones extend usefully to about 5-7 days for track, less for intensity. Beyond that, only frequency and climatology remain predictable. See the WMO’s tropical-cyclone forecasting guidance at community.wmo.int/en/activity-areas/tropical-cyclone-programme. ↩︎

2. NOAA National Centers for Environmental Information, IBTrACS — International Best Track Archive for Climate Stewardship. Home: www.ncei.noaa.gov/products/international-best-track-archive. Basin used: South-West Indian Ocean, 1980-2026 full history. ↩︎

3. The trend in tropical-cyclone intensity in the South-West Indian Ocean and its implications for Madagascar are set out in the SaniTap Knowledge Hub report Madagascar’s cyclones — the record, the outlook to 2100. ↩︎

4. mWater is a WASH-sector data standard used by governments, NGOs and multilaterals to record water-point status at GPS resolution. SaniTap and MadAvance publish the operational Madagascar dashboard at portal.mwater.co. ↩︎