Validation of hail forecast data

Overview

In the current solar industry climate, hail has emerged as one of the primary drivers of insured losses and operational downtime, particularly across North America's "Hail Alley." As PV modules transition toward larger formats with thinner glass, the vulnerability of assets to kinetic impact has increased. Consequently, high-fidelity hail forecasting is no longer a luxury but a critical operational requirement for risk mitigation.

Solargis is one of the few providers offering a specialized hail forecast derived from the High-Resolution Rapid Refresh (HRRR) weather model. Our unique approach integrates advanced post-processing of HRRR data to provide probabilistic risk signals for specific hailstone sizes (e.g.,  ≥ 25 mm or ≥ 50 mm). The primary benefit of this approach is its early-warning capability. By identifying atmospheric risk factors several hours in advance, we provide PV operators with a crucial window to monitor conditions closely and prepare for defensive action. While local volatility makes pinpointing a specific "stow time" difficult, this advanced notice highlights when the risk of glass breakage and micro-cracking enters a critical zone, allowing for heightened readiness before the threat arrives.

Solargis Hail Forecast data accuracy

Hail events are difficult to validate comprehensively, as many significant occurrences go unobserved and unreported. NOAA’s Storm Prediction Center (SPC) collects and archives hail observations across the United States, forming the primary reference dataset for observed hail occurrence and size.

Hail observations are naturally incomplete, with reporting density varying across space and time and biased toward populated areas, cities, and major transportation routes. Some regions provide much more detailed reporting than others. Despite these limitations, SPC records offer reliable confirmation that a hail event occurred and document the reported hailstone size.

Figure 1: SPC records on observed hail events. Density of reported observations correlate with patterns of urban areas and transportation corridors

We compared the Solargis hail forecast with official SPC hail reports from 2021–2023, during which the SPC database recorded 20,905 hail events with reported sizes of at least 1 inch. The HRRR weather model detected approximately 94% of these events in at least one forecast run within a ±3-hour window and a 50 km radius, meaning that nearly all damaging hail events would have triggered an early warning in the Solargis forecast. Given the inherent complexity of hail formation and prediction, this level of performance is considered very good and provides PV operators with valuable time to implement mitigation measures. Around 3.5% of events were predicted as smaller, non-damaging hail, while approximately 2.5% were missed entirely.

The comparison of forecasted and observed hail sizes shows a more pronounced spread. For moderate hail events (1–3 inches), the densest areas of the 2D histogram lie mostly above the line of equality, indicating that the HRRR model tends to slightly overpredict hail size relative to SPC reports, resulting in a conservative estimate. For very large hail events (greater than 3 inches), the pattern reverses, with the model generally underestimating hail size compared to observations. These extreme events are relatively rare, which is reflected in the lower counts observed in the histogram.

Figure 2:  Comparison of SPC events with hail size larger than 1''  in the period 2021-2023 to HRRR model outputs, postprocessed by Solargis method, presented in the density plot.

Insights gained

On 23 June 2023, a violent storm approached Scottsbluff in western Nebraska, where a 5.2 MWp PV power plant with single-axis trackers is located.

Around 9 PM local time (02:00 UTC on 24 June), hailstones up to 4 inches in diameter struck the site, causing extensive damage to the majority of PV modules.

Figure 3: PV power plant in Scottsbluff, western Nebraska after damaging hail event. Source: renewableenergyworld.com

The maps and chart below illustrate how the Solargis hail forecast would have represented the Scottsbluff event several hours in advance. The forecast signal began rising about three hours before the peak impact, reaching a 95% probability for 25 mm (~1 inch) hail and a 35% probability for 50 mm (~2 inch) hail at the time of maximum impact. Maximum model probabilities reached 100% for 25 mm and 58% for 50 mm hail in an area approximately 20 km northeast of the plant.

This case demonstrates that access to short-term hail risk information could provide actionable lead time for PV operators to adjust tracker positions and reduce potential damage. According to news reports, the affected PV modules were later replaced, and the plant returned to service in early 2024.

Figure 7: Reconstructed Solargis hail forecast for 23–24 June 2023. The Scottsbluff PV power plant location is marked on the maps. Forecast potential for hail exceeding 25 mm (left) and 50 mm (right) at 9 PM local time (02:00 UTC). The chart shows the time series of the hail forecast signal at the plant location.

Conclusion

The validation data presented in this document confirms that the Solargis hail forecast is a robust tool for site-specific protection. Key findings include:

• High Detection Rate: The model successfully detected 94% of significant hail events (size ≥ 1 inch) within a ± 3-hour window and 50 km radius.

• Conservative Estimation: For moderate hail (1-3 inches), the model tends to slightly overpredict size, providing a "safety first" conservative estimate for operators.

• Actionable Intelligence: As demonstrated by the Scottsbluff case study, the forecast signal can reach high probability thresholds three hours before peak impact, offering a vital window for automated or manual mitigation.