I have lived in New England my entire life. You may be unaware of this fact, but it snows here sometimes.

Snow can be wonderful. As a child, I shoveled my fair share of driveways. I delivered newspapers in the snow. And many times, I walked through a neighbor’s yard and climbed through a hole in the fence to get to a nearby school, which had a perfect hill for sledding.

Notwithstanding those idyllic images, snow creates unique challenges for solar generation forecasting.
There are two broad categories of solar forecasting. The first category is behind-the-meter (BTM) solar generation, which typically consists of rooftop solar panels. BTM has its own set of challenges including lack of visibility into the data itself. In most cases, the amount of energy that is generated, consumed, and/or sent back to the grid is not directly available to grid operators (at the Independent System Operator level) and it may or may not be available to utility distribution system operators. Because the data is often unavailable, the danger is that grid operators may over-commit generation resources on sunny days (when the BTM is generating a lot of power) and under-commit generation resources on cloud days (when the BTM is not generating much power at all).

The second category consists of grid-connected solar plants. In the area where I live, these facilities are often along the sides of highways. The panels tend to be fixed (i.e., they do not track the sun) and they face southward for maximum exposure to the sun as it traverses the southern sky from east to west.

In some cases, the historical generation for these grid-connect solar plants is available to us, along with the system’s total capacity. And, we have access to the useful weather concepts, such as:

  • Drybulb temperature
  • The ambient temperature
  • Cloud cover – the percent of the sky that is covered by clouds, ranging from 0 to 100, where 0 means a totally clear sky and 100 indicates a totally cloudy sky.
  • Solar irradiance – the amount of energy striking a surface on the earth, measured in watts-per-square-meter. The concepts are often Global Horizontal Irradiance (GHI) or Plane of Array (POA). A related concept is the maximum solar irradiance, which can be calculated from the latitude and longitude of a location. This does not account for the presence or absence of clouds, but merely provides the maximum value at the location and time based on a totally clear sky.

We tend to not have data on snow accumulation. Even if we did have that data, I am not sure that it would be especially useful. If you do not live in a snowy region, this may come as a surprise to you – the sky is typically cloudy when snow is falling. That is useful information because the forecast models would perceive a cloudy day and the generation forecast would be lower.

Eventually (and thankfully), the snow stops falling, at which point the sky may become totally clear. The forecast models may then predict a much higher generation forecast. Unfortunately, that forecast is likely to be wrong because of the accumulated snow covering the panels.

I visited a nearby solar array and I took the following photograph a full 24-hours after the snow stopped falling. The sky is essentially clear, but I estimate that these panels are producing exactly 0 kWh of electricity because they are entirely covered with snow. What are the factors that could influence whether the snow melts? Both increasing temperature and clear skies would certainly contribute.

It seems to me that the owners and operators of such facilities would benefit from low-tech solutions to clear the snow. There may be an entirely reasonable explanation as to why these companies do not proactively clear the panels. Maybe there are issues regarding potential damage to the panels? That seems plausible. Maybe it is too expensive and not economically worthwhile? But that seems doubtful.

I submit to you the following figure, which depicts the historical data from a solar plant in New England. I have clipped the energy units from the Y-axis to keep the data anonymous. The figure shows data from Nov. 1, 2020 through Jan. 31, 2021. The most salient point is that the observations in January 2021 barely exceed 0 for weeks, which means they are generating and selling roughly 0 kWh. This is entirely attributable to snow cover.

The days after snow falls create an inherently difficult forecasting problem. We do not know how much snow fell at the actual site of the panels. In fact, we may only have weather data for a station that is miles away. We do not know if the facility clears the panels. Even if the temperature increases sufficiently to melt the snow, we do not know how long that process may take. The temperature may increase sufficiently and then decreases again, thus freezing the partially melted snow.

These are challenging and intractable issues from the perspective of the forecaster. What can we do about this? First, we must adjust our expectations. In snowy regions, we cannot possibly expect to be as accurate in the winter as we are in the summer. Second, we could attempt a ‘persistence model’ which utilizes lagged loads. This will only be useful when real-time data is available, and the accuracy will certainly degrade as we extend into the forecast horizon. Third, we could attempt to code-up some logic regarding the temperature after a snowfall to account for snowmelt, but those relationships are not likely to be sufficiently robust or consistent to provide a useful signal for the model to discern.

Here is the coda to this tale. I visited the same solar array 3-days after the snow stopped falling. I estimate that the panels are roughly 5% clear. In other words, the array looks substantially the same as when I visited 2-days prior.

Snow is to solar panels what kryptonite is to Superman – it takes away all the power.

Rich Simons
Principal Forecast Consultant - Itron
Since joining Itron in 2000, Mr. Simons has developed, implemented and supported numerous day-ahead and real-time forecasting systems for Independent System Operators (ISOs), retailers, distribution companies, cooperatives and wholesale generators, including NYISO, IESO, TVA, Consolidated Edison, NRG Energy, PSEG and Vectren.

Mr. Simons has implemented systems to support budget & long-term forecasting, weather-normalization, and unbilled-energy estimation for municipal utilities, electric cooperatives and investor-owned utilities, including Ameren, Entergy and FirstEnergy. Mr. Simons has developed forecasting and analysis solutions for municipal water utilities and has developed several customized applications and models for forecasting revenues, managing bills, weather-normalizing sales and estimating unbilled energy. Mr. Simons has reconfigured, streamlined and deployed load research systems at multiple utilities including United Illuminating, Indianapolis Power & Light, TECO Energy, NVEnergy, Colorado Springs Utilities and Lincoln Electric. Mr. Simons has implemented real-time natural gas forecasting systems to support operations at Vectren Energy and Consolidated Edison. In 2019 and 2020, Mr. Simons was a key team-member on a well-publicized report for NYISO to analyze long-term weather trends across the New York state.