October 13, 2026 · Geosapiens
Geosapiens has launched the Geosapiens Hail Model, a hail risk engine that maps the probability and severity of hail at 1 km resolution across North America. Built on a bias-aware neural marked point-process model, it turns fragmented hail observations into location-level, probabilistic risk intelligence for insurers, MGAs and risk teams.
Hail is becoming one of North America's most financially disruptive convective hazards. According to the Insurance Bureau of Canada (IBC), the August 2024 Calgary hailstorm alone generated more than 130,000 claims and over $3 billion in insured losses, and Alberta has had a major hailstorm every year for the past decade, with damage totaling $8 billion. More recently, CatIQ estimated losses from the June 9–10, 2026 Prairie storm outbreak at $923 million.
For insurers, the challenge is not knowing where hail has been reported. It is understanding how often damaging hail is likely to occur, how large the stones may be, and how that risk changes from one location to the next. That is the problem the model is built to solve.
The model combines two decades of atmospheric data with high-resolution environmental information, quality-controlled hail reports and independent surface-station observations in a single probabilistic framework. It answers two questions for every 1 km location: how often does hail fall, and how big do the stones get?
To estimate how often hail falls, the model looks at the atmospheric conditions that produce severe storms, such as instability, wind shear, freezing level and moisture. The model's AI learns how those conditions translate into hail on any given day. When the atmosphere can't plausibly produce hail, the model switches hail off, avoiding false signals. A high-resolution layer adds local climate and terrain detail that broad weather datasets miss.
For size, the model doesn't sort hail into a few fixed categories. Instead, it predicts the full range of likely hailstone sizes. That lets insurers set their own damage threshold, for example 25 mm or 40 mm, and get results instantly, with no retraining.
Explainability analysis confirms that the model's predictions are consistent with the physics of hailstorms. Atmospheric instability, the fuel for strong updrafts, has the most influence. Next come local climate patterns, then freezing level and cloud-base height, then wind shear and rotation.
Validation was designed around genuine out-of-sample tests. The entire 2024 season was quarantined from training and model selection. On that unseen season, the model correctly ranked hail days above non-hail days at weather stations 93% of the time (AUC of 0.933). Its predicted hail geography matched observed reports with a correlation of 0.934 at the regional scale (1° aggregation).
The model provides:
The model is available through the Geosapiens Portal, via API and through Geosapiens data services.
The model gives insurers and MGAs a consistent, location-level view of hail risk across North America from a single calibrated engine. For carriers with property and auto exposure on the Prairies and in hail-prone US states, it supports location-level pricing, geographic risk segmentation and a clearer view of portfolio accumulation. For MGAs, it means faster, more consistent risk selection, with scores and size thresholds that plug directly into underwriting, rating and referral rules.
The key shift is simple: from mapping where hail was reported to modeling how often it occurs, how large it gets, and what happens where reports fall short.
For a demo or data access, contact sales@geosapiens.com.