Madagascar’s cyclones: the record, the outlook to 2100, and what it means on the ground
The destructive capability bearing on Madagascar’s people — storm force × the population in its path — has risen about 4.6× since the 1980s and is on course for roughly 8× by the 2050s. Not because there are more cyclones; because they are stronger, and far more people now stand where they strike. This brief sets out the evidence and translates it into homes, schools and water supply.
~4.6×rise in destruction capability on the population since the 1980s
~8×the 1980s level by the 2050s (index 794, forecast)
6→25%share reaching Category 3+, 1980s vs recent
279cyclones within 300 km of Madagascar, 1975–2026
Total destructive capability on the population
The number that matters is not how many storms come, but what they can destroy when they arrive. Multiply the force of the storms by the people standing in their path: Madagascar’s population has trebled since the 1980s, so the destruction capability bearing on its people is up about 4.6× — and on the current build of population and warming it reaches roughly 8× the 1980s level by the 2050s. This index only climbs: there is no 2000s-style peak to fall back from, because exposure compounds every decade.
Destruction-capability index: PDI × population, 1980s = 100, with a 30-year forecast. Hazard IBTrACS; population UN WPP 2024. Forecast: PDI scaled by the cube of the IPCC wind factor × projected population (SSP2-4.5 median; whisker = scenario range).
The hazard component: destructive potential of the storms themselves
Damage scales with the cube of wind speed, so the standard physical measure is the Power Dissipation Index, PDI = Σ wind³. Stronger storms have roughly doubled the PDI reaching Madagascar since the 1980s.
Power Dissipation Index reaching Madagascar (106 kt³), by decade, with a 30-year forecast. IBTrACS; PDI after Emanuel (2005). Trend: centred 3-decade moving average. Forecast: the observed 2020s value scaled by the cube of the IPCC wind factor (SSP2-4.5 median; whisker = scenario range) — a slow climb to about 29–31 by the 2050s, well below the exceptional 2000s.
The force involved
The numbers are hard to feel, so here is the scale. A mature tropical cyclone is one of the most energetic events on Earth.
~2,000Hiroshima bombs of wind energy released per day (NOAA)
~800,000Hiroshima bombs of total energy (heat) released per day (NOAA)
200×the world’s entire electricity output, in heat (NOAA)
~5×the destructive power of a Category 3 over a Category 1, same place (wind³)
A single Hiroshima bomb released about 6×10¹³ joules. A mature cyclone dissipates roughly that much in the wind alone every forty seconds, and hundreds of thousands of times that in heat each day (NOAA AOML). Because destructive power rises with the cube of wind speed, an intensifying storm does not do a little more damage — a Category 3 hits with about five times the force of a Category 1, and a 10% rise in peak wind is a third more destructive power. This is why the shift toward more intense storms matters even when the count does not change.
Not more storms: the count is flat
Cyclones passing within 300 km of Madagascar each season, 1975–2026. The trend lines cross: the total count is slowly falling (−0.4 per decade), while the number arriving severe is rising (+0.4 per decade, from under one a season in the late 1970s to nearly three now). Fewer storms, more dangerous ones.
All cyclones near Madagascar · of which severe (Cat 1+). Dashed straight lines: linear trends, slope per decade. Part of the early-record severe rise reflects improving satellite intensity estimation. Source: IBTrACS (NOAA/WMO).
Stronger storms: the intensity is rising
The share of Madagascar’s cyclones reaching Category 3 or above rose from 6% in the 1980s to roughly a quarter, and mean peak intensity climbed with it.
Mean peak intensity of the season’s strongest storm (knots), by decade, with a 30-year forecast. IBTrACS. Trend: centred 3-decade moving average. Forecast basis as above: a slow climb of 1–4 kt by the 2050s depending on emissions path, from the observed 101 kt.
A different lens: intensity up, moderated only slightly by fewer storms
The two trends above pull against each other, and it is honest to read them together. Each storm is stronger; there are slightly fewer of them. The net physical product — the Power Dissipation Index, which integrates count and intensity³ — is up about 50% on the 1980s and sits well below its exceptional 2000s peak. Basin-wide studies go further: Tu et al. (2024) find total destructive potential across the whole South Indian Ocean declining since the mid-1990s, driven by falling storm numbers and duration.
×4rise in the share arriving Category 3+ (6→23%)
−0.4storms per season, per decade — the moderating trend
×1.5net rise in PDI reaching Madagascar since the 1980s
×4.6capability once population is counted — the moderation does not survive exposure
Two things stop the moderation being comfort. First, damage lives in the intense tail: destructive power goes with wind cubed, so one Category 3 outweighs several Category 1s, and a season’s harm now concentrates into single landfalls — Batsirai, Gamane, Gezani — rather than spreading across many weak ones. Fewer, stronger storms is not a milder regime; it is a lumpier, more destructive one. Second, the moderation applies to the hazard, not the risk: population growth multiplies whatever the storms deliver, which is why the capability index rises 4.6× while the hazard rises 1.5×. Our decadal model already banks the moderating side — frequency is held flat rather than increased (assumption A1), so if storm numbers keep falling our projections overstate in proportion, and we accept that direction of error.
Projected PDI reaching Madagascar to 2100
Two bases, shown together. Physical scaling applies the IPCC finding that peak winds rise about 5% per 2 °C to three warming trajectories; because PDI depends on wind cubed, the rise is modest but real. Historical trend extrapolates based on the cyclone intensification already experienced over the last 50 years, and is shown as an upper reference.
PDI index reaching Madagascar, projected to 2100 (2020s = 100). Every line starts at today’s level (100, black dot) and is marked at each decade. Solid lines: IPCC AR6 physical scaling under three warming paths. Dashed coral: extrapolation of the cyclone intensification already experienced over the last 50 years (1975–2026), an upper reference.
The same forecast, decade by decade
Year
SSP1-2.6
SSP2-4.5
SSP5-8.5
Historical ref.
2030
101
102
103
107
2040
102
104
106
125
2050
104
106
109
145
2060
104
107
112
166
2070
104
109
116
190
2080
105
110
119
217
2090
105
111
122
245
2100
105
112
126
276
PDI index, 2020s = 100. The physical paths diverge slowly because PDI responds to the cube of a wind change of only a few percent per degree; the historical reference assumes the recent intensification continues linearly, which no physical argument supports beyond the near term.
On physical scaling the destructive potential rises modestly by 2100; combined with a population heading for 54 million, the destruction capability roughly doubles under a middle emissions path and more under a high one.
Decade by decade: the modelled impact on the ground
This is our own model. We fitted the storm process to the 51-year observed record (a Poisson frequency of 3.67 storms a season near Madagascar and the observed distribution of 191 storm intensities, its tail cross-checked with an extreme-value fit), shifted the winds by the IPCC's +2.5% per degree, and ran 6,000 Monte Carlo simulations of every decade to 2100 under three warming paths. Damage scales with wind cubed; exposure grows with the UN population projection. The model is calibrated so its 2020s reproduces the observed 2023–26 damage register: 579,304 people affected over four seasons on the coast.
People affected by cyclones per decade, 2030s–2090s. Shaded band: 10th–90th percentile of the Monte Carlo under SSP2-4.5; lines: scenario medians. The red dot is the observed 2023–26 rate carried to a full decade (1.4M). The bands overlap between scenarios — the difference climate policy makes to Madagascar's cyclone damage emerges slowly and mostly after mid-century; population growth dominates first.
The middle path, decade by decade (SSP2-4.5)
Decade
People affected
Cat 3+ storms near MG
Homes destroyed
School strikes
Water points out
2030s
1.77M (1.15M–2.50M)
8
93,559
2,818
5,607
2040s
2.06M (1.33M–2.90M)
9
108,820
3,278
6,522
2050s
2.24M (1.46M–3.18M)
9
118,474
3,569
7,100
2060s
2.40M (1.57M–3.38M)
9
126,711
3,817
7,594
2070s
2.53M (1.65M–3.58M)
9
133,835
4,032
8,021
2080s
2.54M (1.65M–3.63M)
9
134,287
4,045
8,048
2090s
2.61M (1.69M–3.65M)
9
137,853
4,153
8,262
Medians of 6,000 simulated decades; range in brackets. Homes, schools and water points are converted from people affected at the observed 2023–26 rates (assumption A6). “Water points out” includes contamination requiring disinfection, the dominant mode after flooding (Gamane 2024: ~700 in one storm).
Loss of access to safe water — and what the programme removes
The same model expressed in people-years without safe water: each affected person is assumed to lose safe access for three months on average (contamination is resolved in weeks; destroyed infrastructure takes months). The teal share is what a rolling CRSDW programme removes: 80 systems completed every two years for 30 years (2027–2057, 1,200 systems), each protecting its community, working down the ranked site list.
People-years without safe water per decade, SSP2-4.5 median. Full bar height (dashed top): loss with no programme. Teal: averted by completed CRSDW systems (90% effectiveness assumed); coral: remaining loss. “% covered” is the share of the identified exposed-coast population (4.5M, the sum of all ranked candidate sites) living in a protected community. Coverage reaches 2.66M people (59%) when the build completes in 2057.
4.0Mpeople-years without safe water, 2030–2100, if nothing is built
1.9Mpeople-years averted by the rolling CRSDW programme
1,200systems: 80 every 2 years, 2027–2057
2.66Mpeople living behind a cyclone-resistant supply by 2057
What each scope of the programme takes on
The same century of storms (2030–2100, middle path), seen from the communities each scope protects. Impacts are distributed in proportion to population across the identified exposed coast, so the 81 Phase 1 FRLD communities (8.8% of the exposed-coast population) carry 8.8% of every impact: this is the burden the FRLD investment stands in front of, and the water columns show how much of it the programme removes.
Scope
Systems
People protected
People affected (century)
Water-point outages
People-years without safe water, no programme
Averted by CRSDW
Remaining with CRSDW
Phase 1 (FRLD)
81
398,859
1.4M
4,485
354k
−319k
35k
Phases 1-3
241
927,686
3.3M
10,431
823k
−732k
91k
Rolling programme
1,200
2,663,788
9.5M
29,953
2364k
−1883k
482k
Century totals 2030–2100 on the SSP2-4.5 median path. The three water columns read left to right: the loss those communities suffer if nothing is built, the part averted by that scope of CRSDW (90% effectiveness; build at 80 systems per 2 years from 2027: Phase 1 complete 2029, Phases 1-3 by 2033, the rolling programme by 2057), and the part that remains even with the programme.
Start from one system
One CRSDW system protects one community — on the Phase 1 list, an average of 4,924 people. On the middle path a community on this coast has about a 39% chance per decade of being cyclone-affected, and each event costs its people around three months of safe water. That is 479 people-years of safe water lost per community per decade. A cyclone-resistant system removes 90% of that loss: 431 people-years preserved, per system, per decade.
4,924people protected by one system (Phase 1 average)
479people-years of safe water its community loses per decade without it
431people-years preserved per system per decade (90% of the loss)
Multiply up
People-years of safe water preserved per decade, once each scope is fully built. Each bar is simply its number of systems × the per-system rate. Build timing then decides when the rate is reached: Phase 1 delivers its full 34.9k from 2029, inside the FRLD window; Phases 1-3 from 2033; the rolling programme reaches its full 233k only in 2057. Century totals in the table above account for that timing.
The water figure deserves a pause. Madagascar's national water-point register holds about 79,800 water points today. The model expects roughly 51,155 water-point outages over the century on the middle path — the equivalent of putting two-thirds of today's entire national stock out of action, with many points hit again and again. Every one of those outages is a period when a community drinks unsafe water. A water point that survives the storm, or is back within days rather than months, removes its share of that number outright. That is the programme's case in a single statistic.
At the observed 2023–26 death rate this century of storms would carry roughly 1,868 deaths on the middle path. We state that figure once and do not build on it: deaths depend on warning, shelter and response far more than the other impacts, and Madagascar's early-warning coverage is improving. It could fall well below this even as material damage rises.
What we assumed, stated plainly
Frequency held flat at the observed 5-plus storms a season near Madagascar (AR6: 'same or fewer'). If frequency falls, our totals are overstated in proportion; intensity, not count, drives the trend either way.
Peak winds shift by +2.5% per degree of warming above today (AR6 central: +5% per ~2C; the published range is +1% to +10%).
Ground damage scales with the cube of peak wind above gale threshold (64 kt), the same physics as the Power Dissipation Index. Rainfall flooding (+7% rain per degree) is NOT separately modelled, so flood-driven damage makes these figures conservative.
The model is calibrated so its simulated 2020s reproduces the observed 2023-26 coastal damage register (579,306 people affected over four seasons). It therefore inherits that register's scope: the 51 coastal districts, not the whole country.
Exposure grows with national population (UN WPP 2024 medium variant) and its coastal share is held constant. No allowance for adaptation, better housing, or protection -- these figures are what happens if nothing changes but the climate and the population.
Conversion rates (homes, schools, health centres, water points, deaths per person affected) are fixed at the 2023-26 observed rates. Deaths especially depend on warning and response and could fall even as exposure rises.
Storm winds are drawn from the observed 1975-2026 distribution, so the model cannot invent storm types never seen; it only shifts the observed distribution as the IPCC prescribes.
Our case: the uncertainty that matters for Madagascar is not whether the burden grows — every path shows it growing — but how fast. Population growth roughly doubles the people in harm's way by the 2050s regardless of emissions; the climate signal then compounds it. Infrastructure built this decade will spend its whole working life inside these curves.
What stronger storms mean on the ground
Intensity is not an abstraction. A 10% rise in peak wind is roughly a 33% rise in physical damage (wind cubed), and it lands on a larger population. Anchoring on Cyclone Gezani (2026) — Category 3, the costliest cyclone in Madagascar’s history — and scaling to a +10% peak wind (Category 4) on the mid-century population (about +43% vs today); physical damage scales with wind^3 (+33%), multiplied by exposure:
Impact
Cyclone Gezani (2026)
A stronger future storm*
People affected
478,000
906,997
Homes destroyed
17,980
34,117
Homes damaged
37,253
70,687
Homes damaged or destroyed
102,000
193,543
Schools hit
761
1,444
Health centres hit
30
57
*Illustrative: +10% peak wind (Category 4) on the mid-century population (about +43% vs today); physical damage scales with wind^3 (+33%), multiplied by exposure, a factor of about 1.9×. Gezani figures: UNICEF/BNGRC (2026). Deaths and injuries are not scaled mechanically here: they depend on warning and response as much as on wind, and better preparedness can cut them even as exposure rises.
Water supply is the most exposed and the slowest to recover. Cyclones do not just damage water points, they contaminate them: storm surge and flooding push saltwater and sewage into wells, wash out standpipes, and cut piped supply. After Cyclone Gezani (2026), Toamasina II left without supply; wells and latrines contaminated by flooding. After Cyclone Gamane (2024), some 700 water points needed disinfection. A community can lose safe water for weeks after the wind has passed, which is the gap a cyclone-resistant supply is built to close.
A rising share of intense storms, meeting a population approaching 54 million, means more homes down, more schools hit, and more communities cut off from safe water — each storm reaching further than the last. That is the case for building water supply that survives the storm.
What the science projects
The IPCC Sixth Assessment (2021) and the South-West Indian Ocean literature agree on the shape: not more storms, but a rising share of the most intense and wettest ones.
Total cyclone frequency
▼ down / unchanged
Global TC numbers projected to stay the same or decrease; SWIO studies project a decrease in the number reaching the region.
