Hail Damage Calibration — v1.8 → v1.9

Loading...

Summary

Event Ground Truth N Bldgs MAE Before MAE After Δ MAE Spatial r

Physics Fixes Applied

Parameterv1.8v1.9Basis
theta — unrated asphalt46mm35mmUL 2218 / RICOWI: 3-tab fails 32–38mm
Supercell D_l multiplier1.15×1.30×Blair et al. (2017, BAMS)
Slope exponent1/32/3Hertzian contact theory

Per-Event Detail

Calibration Signal Limitations

No xBD/DINS equivalent for hail. Hail damage (granule loss, small punctures) is sub-visual from satellite imagery — no federal per-building hail damage survey program exists. Roofing permits are the best publicly available proxy for physical damage: they are not insurance data, they are geocoded, and they indicate actual roof repair activity.

The calibration signal is therefore probabilistic: P(permit filed | MESH) is compared to model P(damage | MESH). Permits systematically undercount minor damage (unpermitted repairs) and may overcount in marginal cases (pre-storm wear). These biases are noted in the spatial r values below — lower r may reflect permit noise, not model error.

Reference: Schmid et al. (2024, NHESS) — Swiss open-source hail damage model using C-band MESHS. V_half ≈ 74mm for residential buildings. Parameters not directly transferable (Swiss C-band vs US S-band, different building stock) but validates sigmoidal MDR vs MESH approach.