Catastrophe modeling has spent the last 20 years getting more precise about where a catastrophe happened and what was affected. Parcel-level geocoding, building-level vulnerability, more accurate secondary modifiers and high-resolution terrain data are examples of this focus. Yet almost no progress has been about when events occurred during the catastrophe. Data resolution is no longer the binding constraint on understanding risk, cadence is.
In my years working with insurance buyers of remote sensing data, one refrain came up constantly, "this is great," they would say about high-resolution aerial or satellite imagery, "but I also need to know what is happening before, during, and after an event to fully understand my risk."
There is no clearer example of that temporal gap than wildfire. Wildfire conditions can be forecast, but unlike an impending hurricane there are no days of notice for when they start, just the standing risk that one could begin at any moment. And unlike most other perils, wildfires can be slowed or contained. We can do more than build better and warn people – with the right data at the right time, we can change the outcome of an event.
Unfortunately, there are data gaps in our existing systems that limit our ability to do that today. Geostationary weather satellites provide a snapshot every 20 minutes at two-kilometer infrared resolution, and the published detection floor for the GOES-R fire product is a large 60 x 60 meters. Low Earth Orbit sensors give far finer pixels but only have a couple of looks a day. Aerial sensors are even less frequent and are limited by weather. In situ sensors have limited geographic range and rarely see the fire itself. These systems provide excellent snapshots but are poor at telling the entire story of a wildfire from ignition to extinction.
Commercial GEO remote sensing systems are designed to close that gap. They are deployed farther out in orbit from the Earth, thereby matching the speed of the Earth's rotation, "holding still" over a large geographic region to provide continuous data.
There are obvious upsides to a livestream of data over a large geographical area, event response being the biggest. However, the most impactful are less about what is happening right now and more about a better temporal record of what happened over the life of the event. These include model validation, hours clauses, parametric triggers and mitigation credits.
Every catastrophe model is calibrated against outcomes: final footprint, depth and damage state. The process is inferred from the outcome rather than from observing the progression of the event, and inference is where model divergence lives. This is akin to skipping pages while reading a book. You know how the story ends, but you don't understand how you got there, which in the case of a catastrophe modeling methodology is critical.
Wildfire is the extreme version. We know where fires ended up; we know far less about ignition timestamp, first-hour spread rate, the hours when suppression held and the hour it stopped. A continuous infrared record turns each fire into a time series rather than a handful of perimeter snapshots. Two models can produce the same 40,000-acre burn for entirely different and entirely wrong reasons.
Hours clauses are the clearest case of our market pricing something it cannot measure well. The apparatus assumes we can say when an occurrence began, when it ended, and which losses belong inside it. In practice, it is reconstructed from incident reports, agency data, and argument.
With wildfire, separate ignitions become one event. A continuous, timestamped record of ignition, merge, and progression gives the cedant and reinsurer one clock to read from. That matters for aggregation disputes, retention attachment, and live Cat trading, where the market prices off news coverage and helicopter footage.
Wildfire triggers today lean on a burned area mapped from optical imagery, typically days after the event. Practitioners admit that trigger precision is the whole game, and that the underlying data determines how well a settlement tracks the loss.
Continuous observation moves the trigger off the residue and onto the event, which is why sensors work well. It timestamps onset, which makes duration triggers writable. A geosynchronous satellite does that across a huge geographic area, not just that covered by sensors.
Wind shows what a defensible credit requires. After Hurricane Sally, Alabama's regulator ran a claims data call and commissioned a peer-reviewed study. FORTIFIED construction reduced loss frequency by at least 55% to 74%. Which is why wind mitigation discounts survive a rate hearing.
That method works when the hazard leaves a readable aftermath. Wildfire does not. A destroyed structure tells you almost nothing about what the fire was doing when it arrived. Meanwhile, the regulatory clock is running. California introduced the first mandatory wildfire safety discounts in 2022 and now requires catastrophe models to reflect mitigation. Billions are going into defensible space, home hardening, and vegetation management, credited on engineering judgment and good intentions.
Continuous observation, matched against parcel-level mitigation records and claims outcomes, provides an observed distribution of results for hardened versus unhardened exposures under comparable fire behavior. Mitigation credit stops being an educated guess and becomes a defensible rating variable.
Closing the temporal gap changes what is knowable, giving us a record of a catastrophe as it happened, rather than a reconstruction. We need to start creating a record now for improved peril modeling that will become the standard for the coming decades.
