StormClaim Score Methodology: How Hail Strike Calculates Property-Level Damage Likelihood

Introducing the StormClaim Score

When a hailstorm sweeps through a metropolitan area, thousands of properties may be affected. Roofing contractors need to know which doors to knock on. Insurance adjusters need to know which claims are most likely legitimate. Homeowners need to know whether it is worth filing a claim. Everyone is asking the same fundamental question: "How likely is it that this specific property sustained hail damage, and how severe is it?"

That is exactly what the StormClaim Score is designed to answer.

Developed by Hail Strike's data science team, the StormClaim Score is a proprietary metric that synthesizes multiple data sources -- NEXRAD radar hail estimates, satellite imagery analysis, property characteristics, environmental factors, and historical patterns -- into a single number ranging from 0 to 100 for each affected property.

This article explains the methodology behind the StormClaim Score in detail: what data feeds into it, how the score is calculated, how it is validated, and how professionals should interpret and use it in their work.

The Problem the StormClaim Score Solves

Before the StormClaim Score, storm damage assessment involved piecing together multiple incomplete data sources:

• Radar data tells you hail fell in a general area, but not which specific properties were damaged.
• Satellite imagery can reveal damage patterns, but coverage gaps and resolution limits mean not every property can be assessed visually.
• Storm reports provide ground-truth confirmation, but they are sparse -- a single report might represent an entire county.
• Property records tell you what kind of roof a home has, but not whether it was damaged.

Each source tells part of the story. The StormClaim Score tells the whole story by combining them through our AI-powered fusion model that weighs each data source based on its quality and relevance for the specific property in question.

Score Components: What Feeds Into the StormClaim Score

The StormClaim Score is built from four primary component scores, each derived from a different category of data. Understanding these components helps users interpret the final score with appropriate nuance.

Component 1: Hail Severity Score (0-100)

The Hail Severity Score quantifies the intensity of the hail event at the property's specific location based on radar data processed through our data pipeline. Key inputs include:

• Maximum estimated hail size at the property location, derived from Hail Strike's ML-enhanced hail sizing algorithm that improves upon raw radar estimates by incorporating environmental sounding data and dual-pol radar signatures.
• Hail duration: How long the property was exposed to hail, estimated from storm cell tracking and temporal interpolation between radar volume scans that occur every 4 to 6 minutes.
• Hail kinetic energy: A physics-based estimate of the total impact energy delivered to the roof, calculated from hail size distribution, estimated stone density, terminal velocity, and wind-driven horizontal velocity.
• Wind speed during hail: Higher wind speeds increase the horizontal component of hail impact, amplifying damage to windward-facing roof slopes and increasing the effective kinetic energy of each stone.
• Confidence level: The reliability of the radar estimate, which decreases with distance from the nearest radar station, at lower beam elevations, and in complex terrain where beam blockage occurs.

The Hail Severity Score uses a non-linear mapping that reflects the well-documented relationship between hail size and damage probability. Below 1 inch, damage to most roofing materials is rare. Between 1 and 1.5 inches, damage probability increases steeply for asphalt shingles. Above 2 inches, damage is highly likely regardless of roof type.

Component 2: Visual Damage Score (0-100)

When post-storm satellite or aerial imagery is available, Hail Strike's computer vision models analyze the property's roof and generate a Visual Damage Score. This component provides the closest thing to direct observation without physically inspecting the roof. The score reflects:

• Change detection magnitude: How much the roof's appearance changed between pre-storm and post-storm imagery, measured through pixel-level comparison of normalized imagery.
• Damage classification: The AI model's classification of observed changes into severity categories -- none, minor, moderate, severe, or catastrophic -- based on patterns learned from tens of thousands of labeled training examples.
• Damage extent: The estimated percentage of roof area showing visible damage patterns, calculated from the spatial distribution of detected changes.
• Model confidence: The neural network's softmax confidence in its classification, which varies based on image resolution, cloud shadow interference, roof visibility, and the clarity of the observed patterns.

If no post-storm imagery is available for a property (due to cloud cover, timing gaps, or geographic coverage limitations), the Visual Damage Score is marked as unavailable and the fusion model relies more heavily on the other three components. The system never penalizes a property for lack of imagery -- absence of visual data is treated as missing information, not as evidence of no damage.

Component 3: Vulnerability Score (0-100)

Not all roofs respond to hail equally. The Vulnerability Score captures property-specific factors that influence damage susceptibility:

• Roof material: Asphalt shingles, particularly aged three-tab shingles, are far more susceptible to hail damage than impact-resistant architectural shingles, standing seam metal roofing, or concrete tiles. Material type is the single strongest vulnerability predictor in the model.
• Roof age: Older roofs with brittle, weathered shingles sustain significantly more damage from the same size hail than newer roofs with pliable, well-adhered materials. Our model uses a non-linear deterioration curve that shows accelerating vulnerability after 15 to 20 years for standard asphalt products.
• Roof slope and orientation: Steeper slopes receive more direct hail impact at certain angles, while the orientation relative to the storm's wind direction determines which faces bear the brunt of windward-driven hail.
• Previous damage history: Properties with prior hail claims may have pre-existing compromised areas -- cracked shingles, loosened granules, degraded sealant strips -- that are more susceptible to further damage from subsequent events.
• Geographic factors: Altitude, typical humidity levels, UV exposure intensity, and temperature cycling patterns affect shingle condition and baseline vulnerability independent of any specific storm event.

The Vulnerability Score is calculated even when no storm has occurred, providing a baseline assessment of each property's inherent susceptibility that updates as property characteristics change over time.

Component 4: Contextual Score (0-100)

The Contextual Score incorporates environmental and statistical factors that affect damage probability beyond the direct physical interaction of hail with the roof:

• Nearby confirmed damage: If neighboring properties have confirmed damage from inspections, insurance claims, or high-confidence visual analysis, the probability increases for the subject property. Hail swaths have strong spatial correlation over short distances.
• Storm report proximity: The distance to the nearest official NWS hail report and its reported hail size. Official reports, while sparse, provide reliable ground-truth calibration points.
• Historical claim rates: For the property's geographic area and roof type, what percentage of similar properties filed successful hail damage claims after similar-sized hail events in the past? This historical base rate provides powerful Bayesian prior information.
• Environmental conditions: Temperature and humidity at the time of hail impact influence damage patterns. Cold, dry hail tends to be harder and more damaging per unit size, while warm-season hail falling on hot, pliable shingles may cause less visible damage initially but create latent weaknesses.

The Contextual Score acts as a calibration layer, adjusting the other components based on broader patterns that emerge from Hail Strike's extensive historical database of storms and verified outcomes.

See your properties' StormClaim Scores after the next hailstorm. Sign up for Hail Strike to get property-level damage scores, hail swath maps, and real-time storm alerts delivered straight to your dashboard.

Score Fusion: How Components Combine

The four component scores are combined using a learned weighted fusion model rather than a simple average or fixed formula. The fusion model -- a calibrated gradient-boosted ensemble trained on historical claim outcomes -- determines the optimal weight for each component based on data availability, quality, and contextual relevance.

Dynamic Weighting

The weights assigned to each component are not fixed. They shift dynamically based on the data landscape for each specific property and storm event:

• When high-quality imagery is available: The Visual Damage Score receives elevated weight because it provides the most direct evidence of actual damage.
• When no imagery is available: The Hail Severity Score and Contextual Score receive higher weight to compensate for the missing visual channel.
• For radar-distant properties: Where radar data quality is degraded due to distance from the nearest NEXRAD station, the model reduces the Hail Severity Score weight and increases reliance on imagery and contextual data.
• For unusual roof materials: When the model has less training data for a specific material type (e.g., slate, wood shake, clay tile), it widens confidence intervals and adjusts weights toward data sources that are less material-dependent.

Calibration

The final StormClaim Score is calibrated so that the numeric value approximates the actual probability of actionable hail damage at the property. A score of 75 means that approximately 75 percent of properties with similar scores in our validation data had confirmed hail damage upon professional inspection. This calibration is re-validated after each major storm season using fresh ground-truth data collected from our contractor partner network.

Score Ranges and Interpretation

The StormClaim Score uses the following interpretive ranges:

• 0-20 (Low): Minimal hail impact likely. Damage is unlikely, and most properties in this range will not require repair. Inspection is optional unless the homeowner has specific concerns.
• 21-40 (Below Average): Some hail exposure detected, but damage is uncertain. Properties with older or more vulnerable roofs may have minor damage worth investigating.
• 41-60 (Moderate): Meaningful hail exposure with a reasonable probability of damage. Professional inspection is recommended, especially for roofs with vulnerable materials or significant age.
• 61-80 (High): Strong indicators of hail damage. Most properties in this range will show damage upon inspection, ranging from minor to moderate severity. Inspection is strongly recommended.
• 81-100 (Very High): High-confidence assessment of significant hail damage. Physical inspection should be prioritized urgently. Most properties in this range will require some level of repair or replacement.

Validation Methodology

Hail Strike takes validation seriously. The StormClaim Score is evaluated through several rigorous processes designed to measure real-world predictive performance, not just statistical fit on training data.

Holdout Validation

For each model training cycle, 20 percent of the labeled data is held out as a test set that is never used during training or hyperparameter tuning. Model performance is evaluated on this held-out data using standard classification metrics:

• AUC (Area Under the ROC Curve): 0.91 -- The model correctly ranks damaged properties above undamaged ones 91 percent of the time.
• Precision at 80 percent Recall: 0.84 -- When the model is configured to catch 80 percent of truly damaged properties, 84 percent of its positive predictions are correct.
• Calibration Error (ECE): 0.04 -- The average difference between predicted probability and actual observed frequency of damage is only 4 percentage points across all score bins.

Temporal Validation

To ensure the model generalizes to new storms it has never encountered -- not just new properties from storms it has seen before -- we perform temporal validation. This involves training on data from storms before a cutoff date and testing exclusively on storms that occurred after that date. This is a more realistic and more demanding assessment of real-world performance, since each storm has unique characteristics.

Geographic Cross-Validation

Similarly, we test the model's ability to generalize across geographic regions by training on data from one set of states and testing on another. This guards against geographic overfitting -- the risk that the model memorizes regional patterns rather than learning universal damage indicators.

Field Validation Program

Hail Strike maintains partnerships with roofing contractors across the country who provide ground-truth inspection data for properties in their service areas. After each major storm, we compare StormClaim Scores against actual inspection results to identify any performance degradation, bias patterns, or edge cases. These results feed directly back into the training pipeline, creating a continuous improvement loop.

How Professionals Use the StormClaim Score

For Roofing Contractors

The StormClaim Score transforms how contractors respond to hailstorms, replacing guesswork with data-driven targeting:

1. Storm alert: Within hours of a significant hailstorm, Hail Strike generates StormClaim Scores for all affected properties in the contractor's service area.
2. Sort and prioritize: The contractor sorts properties by score, focusing first on the 80-100 range where damage is most likely and most severe.
3. Targeted canvassing: Instead of knocking every door in a neighborhood, the contractor visits high-score properties first, dramatically improving door-knock efficiency and conversion rates. Data from our contractor partners shows that canvassing efficiency improves by 3 to 5 times when guided by StormClaim Scores versus traditional blanket canvassing.
4. Homeowner education: Showing a homeowner their property's StormClaim Score -- along with the hail size data and imagery comparison that underlies it -- builds credibility and trust. It demonstrates that the contractor is using objective data, not just hoping to find damage.
5. Claim support: The StormClaim Score report, including radar data, imagery comparison, and component breakdowns, provides compelling documentation for insurance claim submissions and supplements.

For a deeper look at using hail data for business growth, see our roofing contractor lead generation guide.

For Insurance Adjusters

Adjusters use the StormClaim Score to streamline the claims evaluation process:

• Triage incoming claims based on damage probability, prioritizing high-score claims for faster processing
• Validate claim narratives against objective data -- does the claimed damage align with the storm's measured impact at that location?
• Identify potential fraud -- very low scores on claimed properties warrant closer scrutiny
• Prioritize field inspections for high-score properties where confirmation of damage is most likely

For Homeowners

Homeowners can look up their property's StormClaim Score after a storm to:

• Determine whether a professional inspection is warranted based on objective data
• Understand the severity of hail at their specific location versus the general area
• Have informed conversations with contractors and adjusters from a position of knowledge
• Make data-driven decisions about filing insurance claims rather than guessing

Limitations and Transparency

No model is perfect, and Hail Strike believes that transparency about limitations is essential to responsible use of the StormClaim Score.

Not a Substitute for Physical Inspection

The StormClaim Score is a screening and prioritization tool, not a definitive damage determination. A high score indicates high probability of damage, but confirmation requires hands-on inspection by a qualified professional. Conversely, a low score does not guarantee the absence of damage -- localized conditions, unusual hail trajectories, microbursts, and unique property features can cause damage that the model does not fully capture.

Data Availability Variation

Score quality depends directly on data availability. Properties close to a NEXRAD radar station with clear post-storm aerial imagery and nearby storm reports will have the most reliable scores. Remote properties far from radar coverage, in areas where post-storm imagery was obscured by clouds, or in regions with sparse storm report networks will have wider confidence intervals and less reliable scores.

Material-Specific Accuracy

The model performs best on asphalt shingle roofs, which constitute approximately 75 percent of residential roofs in the United States and represent the majority of our training data. Accuracy for less common materials -- slate, wood shake, clay tile, flat membrane -- is somewhat lower due to limited training examples, though it continues to improve as our labeled dataset grows with each storm season.

Score Stability

StormClaim Scores may be updated as new data becomes available after a storm. Initial scores based on radar data alone may shift when satellite imagery is analyzed 24 to 72 hours later, or when field inspection reports are incorporated over the following weeks. The platform always displays the most current score along with a timestamp, a data source summary, and the score's revision history so users can track how the assessment evolved.

The Science Behind the Score: A Worked Example

To make the methodology concrete, consider a hypothetical property in suburban Dallas after a significant hailstorm passes through the metroplex:

1. Hail Severity Score: 78. NEXRAD radar data from the KFWS station (25 miles away) shows maximum estimated hail size of 1.75 inches at the property location with 45 mph sustained winds during the hail core. The ML-enhanced sizing algorithm has high confidence. Estimated hail duration at the property was approximately 8 minutes based on storm cell tracking.

2. Visual Damage Score: 72. Post-storm aerial imagery captured 3 days after the event shows texture changes consistent with granule loss on the south-facing and west-facing roof slopes. The convolutional neural network classifies the damage as "moderate" with 0.81 softmax confidence. North-facing slopes show less change, consistent with the storm's wind direction from the south-southwest.

3. Vulnerability Score: 65. County assessor records indicate the property has a 12-year-old three-tab asphalt shingle roof with a 6:12 pitch. The roof age places it in the accelerating-vulnerability portion of the deterioration curve, and three-tab shingles are inherently more susceptible than dimensional or impact-resistant products.

4. Contextual Score: 82. Three neighboring properties within 500 feet have confirmed damage from contractor inspections uploaded to the platform. An NWS Local Storm Report documented 1.75-inch hail 0.8 miles from the property. Historical claim rates for similar properties and hail sizes in this DFW subregion are 74 percent.

5. Fused StormClaim Score: 81. The fusion model weights all four components -- with slightly elevated weight on the Contextual Score due to strong nearby ground-truth evidence and slightly reduced weight on the Visual Damage Score due to moderate image resolution -- and outputs a calibrated score of 81, placing the property in the "Very High" range. The recommendation: priority inspection strongly recommended.

Continuous Improvement

The StormClaim Score is not a static product. Our data science team continuously refines the model through:

• Expanding training data with each new storm season, incorporating tens of thousands of new labeled property outcomes
• Improving radar processing algorithms as new research in dual-polarization radar meteorology advances the field
• Incorporating new data sources as higher-resolution satellite imagery, drone surveys, and IoT weather sensors become available
• Refining the vulnerability model as we accumulate more inspection data on less common roofing materials and configurations
• Collaborating with academic partners in atmospheric science and computer vision to stay at the frontier of damage detection methodology

Conclusion

The StormClaim Score represents the culmination of Hail Strike's entire technology stack -- from NEXRAD radar processing through satellite imagery analysis and AI-powered fusion. By combining multiple independent data sources into a single calibrated metric, it delivers what the hail damage industry has long needed: an objective, data-driven, property-level answer to the question of damage likelihood.

For roofing contractors, the StormClaim Score means more efficient canvassing, higher conversion rates, and more credible homeowner conversations. For insurance professionals, it means faster triage, more consistent evaluations, and reduced fraud risk. For homeowners, it means transparency and confidence in the damage assessment process.

Ready to see StormClaim Scores in action? Sign up for Hail Strike today and get access to property-level damage scores, real-time storm alerts, and the full power of our data-driven storm damage assessment platform.