In some locations, snowfall is one of the most critical aspects of PV system performance and production modeling. Ontario, for example, not only has a high-value solar feed-in tariff program ($0.27–$0.38 per kilowatt-hour, depending on installation location and system capacity), but also has significant yearly snowfall averages (30–130 inches). The industry’s tendency to use increasingly higher dc-to-ac sizing ratios—up to 1.8 in some cases—amplifies snow losses by increasing the relative amount of energy produced in winter. Array oversizing results in inverter-power limiting during summer months, which means that project profitability is more dependent on wintertime yields.
A fortunate aspect of snow soiling is that these losses generally occur at periods of low insolation: PV modules are generally covered with snow when the days are shortest. Snowfall losses are unlikely to occur during on-peak pricing periods in time-of-use markets. The downside of snow soiling is that these losses can be very persistent and can lead to significant energy losses. This is especially true for commercial rooftop systems with a low tilt angle.
To properly account for snow soiling effects, you need to use monthly rather than annual loss factors. Assuming you can supply these monthly values, most performance modeling packages correctly model snow losses. While none of the standard packages currently integrates a model for predicting snow loses, Table 3 includes three published models—the Townsend, Marrion and Andrews models—that you can use, with varying degrees of difficulty, to predict snowfall losses. Note that the reported mean bias error (MBE) of these models is likely lower than their practical accuracy. To date, the industry has only tested the three published models against the data used to derive them, which does not account for differences in climate, mounting method or snow type.
Townsend model. The Townsend model is published in the 2011 Photovoltaic Specialists Conference (PVSC) proceedings (see Resources). BEW Engineering (now DNV GL) derived the model from data for two test sites. The model is easy to implement in Microsoft Excel and can provide a good initial estimate of snowfall losses. It predicts monthly snow soiling losses based on monthly snowfall amount, number of snow events, humidity, air temperature, insolation, array tilt angle, row length and drop height to roof or ground surface. Since the model was derived from locations receiving more than 200 inches of snowfall per year, it loses some sensitivity in regions with lower snowfall amounts.
Marrion model. The Marrion model is published in Solar Energy (see Resources). The model derives from data for six PV systems in Colorado and Wisconsin. It uses hourly air temperature and plane-of-array irradiance values to predict when and how far snow slides off an array, and uses these results to calculate the system’s fractional energy output. Because the Marrion model tracks many historical parameters, it is more complex to implement in Excel, likely requiring scripting.
Andrews model. The Andrews model is published in the 2012 PVSC proceedings (see Resources). It is a purely empirical model that you can tune using data previously collected from similar sites. Project developers can use data from fielded systems with