|
A Model: Smooth Historical Data to Estimate
Underlying Hazard
 Data is county hailstorms per square mile since 1950.
Low population areas have systematically under-reported historical events. “If a
tree falls and no one hears, it doesn't get in the data”. We adjusted reported
experience from low population counties (<50,000) to reflect this underreporting.
We evaluated several geographic smoothing algorithms, and selected Inverse
Distance Weighting (IDW). In this approach, every data point contributes same
weight to the smoothed value for every other point. Nearby points add more
information than distant points. This applies a credibility approach and downplays
shock losses at a location. Other approaches (such as TIN or VIP) base the
smoothed on the most extreme points and would emphasize any shock losses.
We evaluated several possible exponents in the weighting, and selected 2.0 This
produced more consistent patterns, and in considering how loss from a single
moving event would randomly propagate, a square power rate would result.
Estimated Hail Events Per Square Mile
Specific Observations
Activity is more intense on the lee side of mountain ranges:
- Cascades
- Rockies
- Adirondacks
- Appalachians
River valleys seem to matter, but the pattern is complex. Lower
hazard on the lee-side? Higher hazard where three rivers meet?
Possibly over-corrected in lower population areas.
Tornado and Hail
|
Previous