Validation of the snow loss model

Overview

Solargis’ Snow loss model has been validated using ground-based PV production data from 27 sites in the USA and Europe, confirming its ability to categorize and quantify snow-related energy losses across diverse climates and system configurations. PV power production datasets were flagged to identify different categories of snow events impacting the performance.

Note: During the validation, snow events were categorized into three categories by experts through visual inspection of daily PV production profiles.

Snow event categorization

Snow events are categorized into three types based on their impact on PV power production:

• Partial losses: Some solar energy is absorbed by the snow, but not all. This typically occurs when a thin layer of snow covers the modules, allowing some irradiation to pass through.

• Total losses: All solar energy is absorbed or reflected by the snow, resulting in no power production. This occurs when the snow cover is thick enough to block all sunlight.

• Snow slide events: Snow slides off the PV modules rapidly, often triggered by rising temperatures. This results in a sudden increase in power production as the modules are cleared of snow.

Figure 1: Illustrates typical daily patterns representing partial loss, total loss, and snow slide events.

The expert-labeled data was then used to calibrate the model parameters. The monthly discrepancy in power loss estimated by experts and automatic flagging using snow models is presented in Figure 2. For some sites the deviation can be very large in snowy months.

Figure 2: Monthly error distribution between expert-flagged events and snow loss model across all sites.

However, the number of sites remains limited and does not encompass all climatic conditions or possible PV system configurations, highlighting the need for further validation.

Global extrapolation

The model is extrapolated globally using ERA5 data, allowing for the creation of a global map illustrating expected long-term average monthly snow pollution losses. However, this approach carries risks due to local variations in climate and topography. A comprehensive approach involving diverse geographical data and regional models is necessary for accurate predictions.

Figure 2: Shows a global monthly map of expected monthly snow losses (long-term average).

Year-to-year variability

Long-term average maps alone are insufficient for predicting snow losses due to significant year-to-year variability. To highlight areas at risk of substantial losses, Figure 3 presents a map showing the minimum (P10), maximum (P90), and standard deviation of projected snow losses in January from 1994 to 2022. This approach provides a clearer picture of potential snow impacts.

Figure 3: Global maps depicting fundamental statistical parameters of projected monthly snow losses for January (1994–2022).

Effective tilt angle effect

The impact of snow on photovoltaic systems also varies significantly based on the effective tilt angle of PV modules. Modules with steeper tilts accumulate and shed snow differently compared to those with shallower tilts, affecting how snow interacts with the panels and impacting energy production. Figure 4 illustrates how different effective tilt angles influence the expected average monthly snow losses in December from 1994 to 2022. Extrapolating the model to various tilt angles can lead to larger errors, similar to geographical extrapolations, highlighting the need for precise modeling to account for these variations.

Figure 4: Global maps of average expected monthly snow losses in December 1994-2022 in PV systems with different tilts: 10, 30, 50, and 60 degrees.

Conclusion

• The validation of the Solargis snow loss model demonstrates its capability to estimate snow-related photovoltaic (PV) energy losses with reasonable accuracy across diverse climates and system configurations. Using ground-based observations from 27 sites (11 in the USA, 16 in Europe), the model was tested against expert-labeled PV production data, which classified snow events into partial losses, total losses, and snow slide events. Visual inspection of daily production profiles, as illustrated in Figure 1 of the document, enabled precise categorization and calibration of model parameters.

• Despite these strengths, the validation process revealed several important limitations. Monthly error analysis (Figure 2) shows that while the model aligns well with expert assessments in many cases, significant deviations persist, especially during months with heavy snowfall. These discrepancies are attributed to the limited number of validation sites and the diversity of PV system layouts and local climatic conditions, which are not fully captured by the global ERA5/ERA5-Land meteorological datasets.

• Global extrapolation of the model, as visualized in the global maps (Figures 2–4), enables the estimation of long-term average monthly snow losses and their variability. However, these maps also highlight the risks of applying a globally calibrated model to local contexts, given the pronounced year-to-year variability (Figure 3) and the strong influence of PV module tilt angle on snow accumulation and shedding (Figure 4). Extrapolating to different tilt angles or regions without sufficient local data can introduce substantial errors.

• In summary, the Solargis snow loss model provides a robust framework for quantifying snow-related PV losses, supporting better project planning and risk assessment in snowy regions. However, the findings underscore the necessity for ongoing model refinement, expanded validation across more diverse sites, and careful consideration of local system and climatic factors. Continued integration of high-quality ground observations and site-specific calibration will be essential for improving predictive accuracy and ensuring reliable energy yield estimates for stakeholders in the solar industry.