Parametric cover is fast and it is transparent. What it is not is a claim adjuster. It pays on an index, and the index comes out of a hazard model. Everything the hazard model cannot see becomes basis risk, and the amount it cannot see is a function of grid spacing.

US$91.9 Million Against US$12.2 Billion

Hurricane Melissa made landfall in western Jamaica on 28 October 2025 with sustained winds near 185 mph. CCRIF SPC paid US$70.8 million under Jamaica's tropical cyclone policy, the largest single payout in the facility's history, followed by US$21.1 million under the excess rainfall policy. Both moved within days, which is what parametric cover is for and what an indemnity product would not have managed.

The Planning Institute of Jamaica put total damage and loss at J$1.952 trillion, about US$12.2 billion, equal to 56.7 per cent of the country's 2024 GDP, and projected three to five years to return to pre-hurricane output.

Most of that gap is a coverage decision. Jamaica bought a certain amount of parametric protection and that is what it received; nobody expected a sovereign risk pool to cover half a year's national output. But a portion of the gap is not about limits at all. It is about how well the index tracked the damage, and that is where the physics comes in.

The Square Law

Wind does not load a structure in proportion to its speed. It loads it in proportion to dynamic pressure:

q = ½ ρ V²

With warm, moist tropical air at about 1.15 kg m−3, a 185 mph sustained wind gives 3.93 kPa. A 110 mph wind gives 1.39 kPa. The stronger storm is 68 per cent faster and applies 2.83 times the load.

For underwriting, the important consequence is the reverse direction. A hazard field that is 10 per cent wrong on wind speed is 21 per cent wrong on load. Twenty per cent becomes 44 per cent. Thirty per cent becomes 69 per cent. Model error does not pass through to loss estimate unchanged; it is amplified on the way.

Wind load scales with the square of wind speedA plot of dynamic pressure against wind speed, showing that the load at 185 miles per hour is 2.8 times the load at 110 miles per hour, and a diagram of the pressure coefficients acting on a building, with roof corners carrying the largest suction. Because load goes as the square of speed, a 20 per cent error in the predicted wind becomes a 44 per cent error in the predicted load.Why a 20% wind error is a 44% load errorDynamic pressure q = ½ρV², with ρ = 1.15 kg m⁻³ for warm, moist tropical air. Coefficients follow the usual building-code zones.ALoad against wind speed01234kPa04080120160200sustained wind (mph)Cat 1Cat 2Cat 3Cat 4Cat 5110 mph 1.39 kPa185 mph 3.93 kPa2.83× the load of Cat 2Melissa at Jamaican landfall, 28 Oct 2025BWhere the load actswindward GCₚ +0.83.1 kPaleeward −0.52.0 kParoof field −1.084.2 kPacorner −2.188.6 kPa425 kN of uplift on a 100 m² roof at 185 mphabout 43 tonnes trying to lift the roof off the walls. Corners let go first.Because q ∝ V², forecast error is amplified before it reaches the structure:a 10% error in wind becomes 21% in load, a 20% error becomes 44%, a 30% error becomes 69%. This is why resolving the wind field over terrain matters more than it looks.
Figure 1. Dynamic pressure and the square law. q = ½ρV² with ρ = 1.15 kg m−3 for warm, moist tropical air. At the 185 mph sustained wind recorded at Hurricane Melissa's Jamaican landfall on 28 October 2025, q is 3.93 kPa, which is 2.83 times the load at 110 mph. The pressure coefficients are the standard building-code zones: the roof-field value works out to 425 kN, about 43 tonnes-force, on a 100 m² roof, and corners carry roughly twice that per unit area. Because load goes as the square of speed, a 20 per cent error in the predicted wind is a 44 per cent error in the predicted load.

Now add terrain. Air accelerating over a ridge of moderate slope gains roughly 40 per cent in speed at the crest, which by the square law is close to double the load. A house on a ridge and an identical house on the plain a kilometre below it are exposed to materially different storms. A wind index that reports one speed for a 2,500 km² cell has described neither of them, and it will under-pay one insured while over-paying the other in the same event.

Surge Is About the Shelf

The same problem appears in water, and it is less well understood on the buy side.

Storm surge is often discussed as though it scales with wind speed. It does not, at least not primarily. Steady wind setup across a continental shelf follows

Δη = τL / (ρw g h), with τ = ρa CD V²

which grows with fetch L and falls with water depth h. The pressure contribution, the inverse barometer effect, adds about one centimetre of sea-level rise per hectopascal of central pressure deficit, so Melissa's minimum central pressure contributed a little over a metre before the wind did anything at all.

Run the same storm over two different coasts. Over a shelf 20 m deep and 10 km wide, which is broadly what Jamaica's south coast looks like, the setup is about half a metre and the total still-water rise is 1.7 m. Over a shelf 10 m deep and 100 km wide it is more than ten metres of setup and 11.5 m in total.

Storm surge depends on the shelf, not only the windTwo coasts under the same 60 metre per second wind and the same 120 hectopascal pressure deficit, drawn at the same horizontal scale. Over a shelf 20 metres deep and 10 kilometres wide the wind setup is half a metre. Over a shelf 10 metres deep and 100 kilometres wide the same wind piles up more than ten metres, because setup grows with fetch and falls with depth.The same storm, two coasts, two very different surgesWind setup Δη = τL / (ρ_w g h), with τ = ρ_a C_D V². Both panels: V = 60 m s⁻¹, central pressure deficit 120 hPa, same horizontal scale.ANarrow, steep shelf · 20 m deep, 10 km wideJamaica's south coastshelf breakstill water levelV = 60 m s⁻¹1.7 m020406080100distance offshore (km)pressure term, inverse barometer at 1 cm/hPa1.19 mwind setup τL/(ρ_w g h) across 10 km of 20 m water0.51 mtotal still-water rise at the shore1.71 mBWide, shallow shelf · 10 m deep, 100 km widenorthern Gulf of Mexico typeshelf breakstill water levelV = 60 m s⁻¹11.5 m020406080100distance offshore (km)pressure term, inverse barometer at 1 cm/hPa1.19 mwind setup τL/(ρ_w g h) across 100 km of 10 m water10.29 mtotal still-water rise at the shore11.49 mSetup rises with fetch and falls with depth, so bathymetry is the larger term. Vertical scale exaggerated about 30×.A model that resolves the storm but not the shelf gets the same storm right on one coast and wrong on the next one. That is the failure mode islands live with.
Figure 2. Surge and the shelf. Wind setup is Δη = τL/(ρwgh) with τ = ρaCDV², so it grows with fetch and falls with depth. The same 60 m s−1 wind and the same 120 hPa pressure deficit give 1.7 m of still-water rise over a shelf 20 m deep and 10 km wide, and 11.5 m over one 10 m deep and 100 km wide. The pressure term, about 1 cm per hectopascal, is identical in both. Steady-state setup only: a real surge also carries wave setup, tide phase and the storm's forward speed.

The same wind, the same pressure, a factor of seven in the water at the shoreline, decided by bathymetry. A parametric surge trigger calibrated on one coastal geometry and applied to another is carrying a large and mostly invisible source of basis risk.

Where the Index Loses the Detail

Catastrophe models for this region inherit their hazard fields, directly or indirectly, from climate and weather models that run at 25 to 50 km. At 50 km, between 12 and 20 degrees north and 64 and 56 degrees west, there is no land grid point at all: every island in the Lesser Antilles is ocean to the model, with the nearest land about 500 km away.

Jamaica fares better, occupying four to six cells, but loses its relief. Blue Mountain Peak reaches 2,256 m; a 50 km cell averages it to about 851 m along the line of the trade winds. Since Jamaican rainfall is produced by air being forced up that terrain, flattening the terrain flattens the rainfall. The north-eastern slopes take 3,000 to 5,000 mm a year and the southern coastal plains under 1,500 mm, and the model gives both the same figure.

For an excess-rainfall trigger, that is the whole ballgame. The index is computed from a field that does not contain the gradient the loss actually follows.

The practical form of basis risk. It is not a random error that averages out across a portfolio. It is a systematic error correlated with terrain: policyholders on ridges and in steep catchments are systematically under-indexed, and policyholders on open flat ground are systematically over-indexed. Portfolio diversification does not remove a bias that tracks topography, because topography does not diversify.

What Better Hazard Modelling Changes

Three things, in descending order of confidence.

Tighter triggers. A hazard field at one kilometre can distinguish a destroyed parish from an intact one. That allows a trigger defined on a smaller geographic unit, which mechanically narrows basis risk in both directions and makes it possible to price micro-level and meso-level products that are currently uneconomic because the index is too blunt.

Honest tail estimates. Extreme-value behaviour is where resolution bites hardest, because extremes are local. A coarse field smooths the tail by construction, so a coarse model tends to understate the frequency of severe local events and overstate the frequency of moderate widespread ones.

Affordable ensembles. This one matters commercially. Running a conventional model at one kilometre is a factor of 125,000 more expensive than at 50 km in two dimensions, because refining the grid forces a shorter timestep to satisfy the Courant condition. An AI emulator has no timestep to shorten, and a decomposed emulator evaluates subdomains independently, so the cost of a large ensemble stops being prohibitive. Ensembles are what turn a single scenario into a distribution, and a distribution is what an actuary can price.

Questions for the Buy Side

If you are buying, renewing or broking parametric cover in this region, four questions expose most of the hidden basis risk.

  1. At what native resolution is the hazard field computed? Not the resolution it is delivered at. Interpolating a 50 km field onto a 1 km grid adds pixels and no information.
  2. Does the model carry terrain into the wind field, or only into the exposure layer? If terrain enters only at the vulnerability step, topographic speed-up has been left out of the hazard entirely.
  3. How was the surge component calibrated, and on which shelf geometry? Ask specifically whether the bathymetry used is the local one.
  4. What is the modelled-to-actual ratio on the last three events, by parish or district? A portfolio-level ratio near one can hide large offsetting errors underneath it. Ask for the dispersion, not the mean.

Frequently Asked Questions

Does this mean parametric insurance does not work?

No. Speed is the product, and CCRIF moved money to Jamaica within two weeks of Melissa, which no indemnity process would have done. The point is that basis risk has a specific, physical and reducible component, and the industry has generally treated it as an unavoidable cost of the structure.

Would a finer model have increased Jamaica's Melissa payout?

Not by itself. Payout size is set by the policy limit and the attachment point, and those are coverage decisions. A finer hazard model changes how closely the payout tracks the distribution of damage within the covered area, and it changes what products can be written at all at the meso and micro level.

Who is doing this work in the region?

Sub-kilometre climate emulation for the Caribbean is an active research area at the Climate Studies Group Mona, in the Department of Physics at The University of the West Indies, Mona, where Caribbean climate science has been done since 1994. The work is in progress and there are no published results yet. The technical argument is set out here.

Is any of this in production today?

The physics in this article is standard and long established: the square law, the shelf-setup relation and the inverse barometer effect are in every coastal engineering text. What is not yet established is whether kilometre-scale AI emulation can deliver hazard fields at that resolution with the reliability an underwriter needs. Treat resolution claims from any vendor, including research groups, as claims to be tested rather than accepted.

Adrian Dunkley is a physicist and AI researcher in the Department of Physics at The University of the West Indies, Mona, and a member of the Climate Studies Group Mona. His UWI profile is here. This article is educational content and not insurance or financial advice.