What is catastrophe modeling?

Catastrophe modeling is the practice of estimating the financial losses a portfolio of insured properties would suffer from low-probability, high-severity events such as hurricanes, floods, earthquakes, and wildfires. A cat model simulates thousands of possible events, applies each to the exposure at every location, and produces a distribution of losses.

What catastrophe modeling is

Catastrophe modeling emerged in the late 1980s, after Hurricane Andrew in 1992 showed that historical claims alone could not price rare events. AIR (now Verisk) and RMS (now Moody’s) built the first commercial cat models.

A cat model has four modules. The hazard module generates a catalog of synthetic events with their footprints and intensities. The exposure module holds the insured locations with their attributes and values.

The vulnerability module translates intensity at a location into a damage ratio using functions specific to construction type and occupancy. The financial module applies policy terms to convert damage into insured loss.

The defining property is that the output is probabilistic. Instead of asking what a specific storm would cost, catastrophe modeling asks what the distribution of losses looks like across every plausible event.

Insurers, reinsurers, brokers, and rating agencies all run cat models. The Oasis Loss Modelling Framework provides an open platform, and regulators in many markets require model-based capital assessment.

What catastrophe modeling does

Catastrophe modeling puts a number on tail risk so that capital, pricing, and reinsurance can be set against it.

A primary insurer runs its portfolio through hurricane and flood models to calculate how much reinsurance to buy and at what attachment point.

A reinsurer prices a treaty by modeling the cedent’s exposure against its own view of the hazard.

Underwriters check a new commercial account’s contribution to portfolio loss before binding it, avoiding accumulation in a single wind zone.

Rating agencies and regulators read a carrier’s modeled 1-in-100 and 1-in-250 year losses to judge solvency.

After an event, the same model provides an early loss estimate from the actual footprint, days before claims arrive.

Every one of these runs starts from the exposure data, and its quality bounds the answer.

How catastrophe modeling works

A cat model run is a sequence of spatial joins followed by arithmetic.

Module	Input	Output
Hazard	Event catalog of 10,000 to 100,000 synthetic years	Intensity footprint per event: wind speed, flood depth, shaking, flame length
Exposure	Locations with coordinates, construction, occupancy, year built, replacement value	Geocoded exposure table
Vulnerability	Damage functions by construction class	Mean damage ratio and uncertainty per location per event
Financial	Deductibles, limits, reinsurance layers	Insured loss per event, per location, per policy

The exposure table is joined to each event footprint by location, so a million-location portfolio against a 50,000-event catalog is 50 billion location-event pairs in principle, pruned to those where intensity is nonzero.

Two outputs summarize the result. Average annual loss (AAL) is the mean loss across all simulated years. The exceedance probability (EP) curve gives the loss at each return period, such as the 1-in-100 year loss.

Data quality drives everything. Geocoding precision, correct construction class, and up-to-date replacement values change modeled loss by tens of percent.

Hazard layers are where geospatial data enters. Flood depth grids, wildfire hazard rasters, storm surge grids, and soil amplification maps are all rasters that must be sampled at every exposure location.

The model’s output is a loss table by event and location, which analysts then aggregate by region, peril, line of business, or reinsurance layer.

Catastrophe modeling in Wherobots

WherobotsDB handles the spatial half of catastrophe modeling: geocoding, enriching, and joining exposure to hazard at portfolio scale, before the vulnerability and financial arithmetic runs.

Millions of policy locations join to hazard footprints, flood zones, and administrative boundaries in one SQL statement. WherobotsDB partitions both sides spatially, so a portfolio-wide join completes on a single runtime.

Hazard rasters join the same way. RS_Value samples flood depth, wildfire hazard, or surge height at each location, and RS_ZonalStats returns the maximum inside each building footprint, per return period, in one query.

Exposure enrichment draws on the Wherobots Data Hub. Overture Maps supplies 785 million building footprints, ESA WorldCover supplies land cover around each site, and NOAA severe weather records supply event history. RasterFlow detects roofs and solar panels from imagery to correct construction attributes.

Getis-Ord Gi* hotspot analysis in WherobotsDB finds statistically significant accumulations of insured value before the next event does.

Wherobots documents these patterns in its Insurance & Risk use case, including a California coastal flood exposure notebook that intersects parcels with FEMA zones and sums assessed value at risk.

Data stays in the customer’s cloud storage or Iceberg catalog. Nothing is copied into a modeling appliance. The Wherobots MCP Server exposes the same tables and functions to Claude Code and other agents, so an analyst can ask for exposure by peril in plain language.

-- Exposure sampled against a 100-year flood depth grid, summed by county
SELECT c.county_name,
       COUNT(*)                                   AS locations,
       SUM(e.replacement_value)                   AS tiv,
       SUM(CASE WHEN RS_Value(d.rast, e.geometry) > 0
                THEN e.replacement_value END)     AS tiv_in_100yr
FROM exposure e
JOIN counties c ON ST_Intersects(c.geometry, e.geometry)
LEFT JOIN flood_depth_100yr d ON RS_Intersects(d.rast, e.geometry)
GROUP BY c.county_name;

Related terms

• Flood risk data: the hazard module input for flood perils
• Property risk data: the exposure attributes a cat model needs per location
• Wildfire risk map: the hazard layer for wildfire cat models
• Parcel data: how exposure locations are tied to a footprint

Read more from Wherobots

• How to score every building in a state for catastrophe risk
• Insurance & Risk use case
• Financial services & insurance solutions
• Getis-Ord Gi* hotspot example

Join your exposure to hazard layers in WherobotsDB at cloud.wherobots.com.