How much energy do solar panels lose to snow? NREL's city table by roof pitch, in years of payback
Snow takes 0 to 15.9 percent of a fixed rooftop array's annual energy in the continental United States, by NREL's national model at a 20 degree tilt, with a median of 2.5 percent. On a 12-year payback that adds 0 to 2.3 years, and 167 of the 201 continental stations in the table add less than one. For the rest of the winter story, cold, short days and what to do about snow, see do solar panels work in winter.
The numbers come from Appendix A, Table A-1 of the 2017 NREL report by David Ryberg and Janine Freeman, which simulated every winter from 1961 through 1990 at each station. Minneapolis averages 10.7 percent at a 20 degree tilt and 9.5 percent at a tilt equal to its latitude of 44.9 degrees. Phoenix and Los Angeles print 0.0 at both tilts, and Atlanta prints 0.1. Boulder loses 5.5 percent at 20 degrees and 3.9 at 40, which is 0.21 years of payback at the 12-year reference. Denver itself is not a station.
Because simple payback is price divided by yearly value, a loss L stretches a payback P to P divided by (1 minus L), which adds P x L / (1 - L) years. The DOE Q1 2025 benchmark prices an 8 kW system at $2.95 per watt, $23,600 in 2024 dollars, with no federal tax credit in the figure. A 12-year payback on that price means $1,967 of production value a year, and Minneapolis's 10.7 percent takes $210 of it. Every loss figure on this page rests on weather recorded 36 to 65 years before this check, so the table sets a starting point for a roof-specific model and does not replace one.
What Table A-1 actually measures
Table A-1 is a model output, not a field measurement. Ryberg and Freeman (NREL/TP-6A20-68705, August 2017) put the snow coverage model of Marion and co-authors (2013) into NREL's System Advisor Model and ran it on the hourly weather of the National Solar Radiation Database for 1961 through 1990. That data set covers 239 locations and includes a daily snow depth record. For each location and each of the 30 years, the authors simulated a photovoltaic system twice, once without the snow model and once with it, and took the difference normalized by the snow-free result. Table A-1 prints the average of the 30 yearly percentages and their standard deviation.
Two system designs ran at every station. One tilts the array at the station's latitude, the angle thought to maximize a year's sunshine and one the authors expected to shed snow faster in the north. The other holds the tilt at 20 degrees, which they call common because the design is simple and the hardware widely available. Both use 18 modules in a single row, 3 modules tall and 6 wide, and leave every other input at the System Advisor Model defaults for the detailed photovoltaic, no financial model option (version 2015.6.30). The table prints no latitude column. The latitudes on this page come from the station list in the 1995 TMY2 user's manual by William Marion and Ken Urban, which covers the same 239 stations.
The 2017 report replaced an earlier edition, NREL/TP-6A20-64260 from September 2015. The 2017 preface says: "The original implementation of the snow model described in this paper in the System Advisor Model (SAM) had a bug that, when fixed, changed the results described herein." The copy of the 2015 file on the NLR documents server now shows only a notice that the document has been updated, with a button to the newest version. A snow map or blog table built from 2015 numbers may not match Table A-1, so check which edition a source used.
Table A-1 prints 223 rows, not 239. Sixteen stations on the TMY2 list have no row: Long Beach CA, Eagle CO, Athens GA, Louisville KY, Cut Bank MT, Kalispell MT, Miles City MT, Tucumcari NM, Minot ND, Burns OR, Guam, Charleston SC, Midland TX, Lynchburg VA, Elkins WV, La Crosse WI. The report does not say why. Several sit in snowy country, so a reader near Minot, Kalispell or La Crosse has to borrow a neighboring row, such as Bismarck, Great Falls or Eau Claire, and treat the result as an approximation.
| Item | What the report states |
|---|---|
| Weather data | NSRDB 1961-1990, hourly, with daily snow depth; 239 locations; 30 years each |
| Designs simulated | Tilt equal to latitude, and a constant 20 degree tilt |
| Array | 18 modules in one row, 3 tall by 6 wide; other inputs at SAM defaults (detailed photovoltaic, no financial model, version 2015.6.30) |
| Value published | Mean and standard deviation of 30 annual loss percentages, per station and design |
| Snow removal modeled | Sliding only; no melting; wind removal only for flat fixed-tilt systems |
| Rows in Table A-1 | 223: 17 in Alaska, 4 in Hawaii, 1 in Puerto Rico and 201 in the continental states |
| Range stated in the text | 0% to 16% for fixed-tilt systems in the continental states; Alaska sites up to about 40% |
Where the losses fall across stations
At a 20 degree tilt, 79 of the 223 stations lose less than 1 percent and 84 lose 5 percent or more. Restricted to the 201 continental stations, the report's own scope, 69 lose 5 percent or more (34 percent of them) and 15 lose 10 percent or more. At a tilt equal to latitude, the counts at 5 percent or more fall to 54 and the counts at 10 percent or more to 8. The median continental station loses 2.5 percent at 20 degrees and 1.8 at latitude tilt.
The highest continental values at 20 degrees are International Falls, Minnesota (15.9), Sault Ste. Marie, Michigan (15.1), Duluth, Minnesota (15.1), Caribou, Maine (14.4), Houghton, Michigan (12.6), Fargo, North Dakota (12.2) and Bradford, Pennsylvania (11.9). The report describes the belt as starting in the Northeast, spanning the Great Lakes and continuing into the Midwest and the northern Rocky Mountains, with pockets of higher ground showing more loss than their neighbors. Its example is Flagstaff, Arizona, modeled at 4.7 percent at 20 degrees and 3.0 at latitude tilt, against Phoenix at zero.
A zero in the table is a rounded value to one decimal. 38 stations print 0.0 at a 20 degree tilt, 33 of them in the continental states, and the whole lowest band is worth little in payback terms: at a 12-year reference even a full 1.0 percent adds only 0.12 years. The 15 continental stations at 10 percent or more at 20 degrees are in Minnesota (International Falls, Duluth, Rochester, Saint Cloud, Minneapolis), Michigan (Sault Ste. Marie, Houghton, Alpena, Traverse City), Maine (Caribou), North Dakota (Fargo), Pennsylvania (Bradford), Wisconsin (Eau Claire), New York (Massena) and Montana (Lewistown). The standard deviations are the larger story in that group, and they get their own section after the city lookup.
| Loss band | All 223 rows, 20 degree tilt | All 223 rows, latitude tilt | 201 continental rows, 20 degree tilt | 201 continental rows, latitude tilt |
|---|---|---|---|---|
| Under 1 percent | 79 | 87 | 74 | 81 |
| 1 to under 2 percent | 20 | 23 | 19 | 23 |
| 2 to under 5 percent | 40 | 45 | 39 | 43 |
| 5 to under 10 percent | 59 | 51 | 54 | 46 |
| 10 percent or more | 25 | 17 | 15 | 8 |
| Total stations | 223 | 223 | 201 | 201 |
How the model clears a covered array
The Marion model, as the report describes it, checks at the start of each day for snowfall using the snow depth reading (the System Advisor Model version checks at the start of every time step), and when snow has fallen it treats the whole array as covered. The array stays covered until an hour arrives in which sliding is allowed. Sliding is allowed only while the ambient temperature in degrees Celsius exceeds the plane-of-array irradiance divided by a constant of minus 80 watts per square meter per degree. At 200 watts per square meter that means anything warmer than -2.5 degrees C; at 400 it means warmer than -5; at 800 it means warmer than -10. Under heavy cloud or at dawn the irradiance term is small and the threshold sits near freezing, and in a cold clear spell it sits much lower, which is why a sunny day after a storm clears a roof faster than a gray one.
In an hour when sliding is allowed, the exposed part of the array grows by 1.97 times the sine of the tilt, in tenths of the row's height. The report says Marion measured the 1.97 coefficient on roof-mounted systems and reports a coefficient of 6.0 for ground mounts; only the roof-mount value is built into the System Advisor Model. The authors explain the gap by snow accumulating at the lower edge of a roof array and preventing snow from leaving the lower modules. Any string with even one covered module is set to zero output, so the model gives up a whole string's output before the last strip of snow leaves.
The table works out what the formula means for a fully covered row. A 45 degree array needs about 7.2 hours of sliding-eligible time to clear, a 20 degree array about 14.8, and a 10 degree array about 29. Those are hours that qualify, not clock hours; a cold gray afternoon may supply none of them. In the national study the tilt effect is weaker than the clearing speeds alone would suggest. Across the 108 continental stations that lose 2 percent or more at 20 degrees, the latitude-tilt loss runs from 60 to 97 percent of the 20 degree loss, averaging 80 percent. The report's own explanation of what drives the loss is snow depth and how long it persists: its Figure 7 shows annual loss climbing with the yearly sum of hourly snow depth, "more of an exponential or polynomial relationship than a linear one," and notes that if snow cover coincides with low-irradiance hours, less energy is lost.
The authors list what the model leaves out: melting, wind removal except on flat fixed-tilt arrays, and time-delayed effects such as a cold night freezing the bottom layer of snow to the glass. It was built for tilts between 10 and 45 degrees, and the System Advisor Model version adds a check that clears the coverage when the snow depth reads zero, so that flat arrays do not hold snow forever. Validation covered three systems that played no part in building the model: a flat array in Washington, D.C., a 10 degree array in Golden, Colorado, and a one-axis tracker. Annual energy error fell from 9.7 to -2.2 percent for the Washington array and from 8.5 to 4.4 percent for the Golden array. Monthly results were poor; the Golden array's February error fell from 306.1 to 197.4 percent and was still far off. The report advises that results "should only be factored into annual considerations and not applied to monthly or shorter time periods." It adds that the model "will certainly not always produce reliable results for any PV system in any location."
| Tilt | Tenths of row height cleared per eligible hour | Eligible hours to clear a fully covered row (10 tenths) |
|---|---|---|
| 10 degrees | 0.342 | 29.2 |
| 20 degrees | 0.674 | 14.8 |
| 30 degrees | 0.985 | 10.2 |
| 40 degrees | 1.266 | 7.9 |
| 45 degrees | 1.393 | 7.2 |
Snow loss in 25 cities at two tilts
The 25 stations below span the snow regimes in Table A-1, from places where snow is a footnote to Fargo and Fairbanks. A 20 degree tilt is a pitch of 4.37:12, so a 4:12 or 5:12 roof sits within two or three degrees of the first column. The tilt-equals-latitude column is the loss at the roof pitch shown in the third column, for example 10.7:12 for Chicago. Standard deviations across the 30 winters are in parentheses. The names are weather stations and not municipalities. Boulder, for example, is the Colorado station nearest the Denver area, since Denver has no row.
Latitude does not predict the loss. Boulder, at 40.0 degrees north, averages 5.5 percent at 20 degrees; Salt Lake City, at 40.8, averages 3.5. Chicago sits at 41.8 and loses 6.1, while Minneapolis, three degrees north, loses 10.7, 1.75 times Chicago's figure. Buffalo's 8.6 is 5.4 times Philadelphia's 1.6. Portland, Maine, loses 5.6 percent, twice Boston's 2.8. The pattern is regional weather: how much snow falls, how long it lies, and whether the days that follow are cold and gray or cold and bright. The same logic applies to distance: a city that sits in a different snow belt from its nearest station has a different number.
Alaska partly reverses the tilt rule. Seven of the 17 Alaska stations lose more at latitude tilt than at 20 degrees; Fairbanks loses 12.0 percent at 20 degrees and 14.2 at its 64.8 degree latitude, and Barrow 33.0 and 38.3. The report's footnote offers a reason: at high latitudes the benefit of a low tilt in summer, measured against annual production, outweighs the winter penalty. The interpolation in the roof pitch section below leaves Alaska out for that reason.
| Station | Latitude (deg N) | Pitch equal to latitude | Loss at 20 degrees (about 4.4:12) | Loss at latitude tilt |
|---|---|---|---|---|
| Phoenix, AZ | 33.4 | 7.9:12 | 0.0 (0.0) | 0.0 (0.0) |
| Atlanta, GA | 33.6 | 8.0:12 | 0.1 (0.2) | 0.1 (0.2) |
| Los Angeles, CA | 33.9 | 8.1:12 | 0.0 (0.0) | 0.0 (0.0) |
| Seattle, WA | 47.5 | 13.1:12 | 0.3 (0.4) | 0.2 (0.3) |
| Philadelphia, PA | 39.9 | 10.0:12 | 1.6 (0.8) | 1.1 (0.6) |
| New York City, NY | 40.8 | 10.4:12 | 1.9 (0.9) | 1.2 (0.6) |
| Boston, MA | 42.4 | 10.9:12 | 2.8 (1.1) | 1.9 (0.8) |
| Salt Lake City, UT | 40.8 | 10.3:12 | 3.5 (1.4) | 2.3 (1.1) |
| Boise, ID | 43.6 | 11.4:12 | 2.0 (1.7) | 1.3 (1.0) |
| Flagstaff, AZ | 35.1 | 8.4:12 | 4.7 (1.5) | 3.0 (1.0) |
| Boulder, CO | 40.0 | 10.1:12 | 5.5 (1.4) | 3.9 (1.2) |
| Cheyenne, WY | 41.1 | 10.5:12 | 6.0 (1.7) | 4.6 (1.3) |
| Kansas City, MO | 39.3 | 9.8:12 | 3.7 (1.6) | 3.0 (1.4) |
| Indianapolis, IN | 39.7 | 10.0:12 | 4.3 (2.2) | 3.5 (2.0) |
| Chicago, IL | 41.8 | 10.7:12 | 6.1 (2.6) | 5.2 (2.6) |
| Cleveland, OH | 41.4 | 10.6:12 | 6.7 (1.8) | 5.8 (1.8) |
| Detroit, MI | 42.4 | 11.0:12 | 5.6 (1.8) | 4.5 (1.7) |
| Milwaukee, WI | 43.0 | 11.2:12 | 7.4 (2.8) | 6.2 (2.7) |
| Minneapolis, MN | 44.9 | 12.0:12 | 10.7 (2.9) | 9.5 (2.6) |
| Fargo, ND | 46.9 | 12.8:12 | 12.2 (4.0) | 11.7 (3.9) |
| Buffalo, NY | 42.9 | 11.2:12 | 8.6 (2.0) | 7.4 (2.0) |
| Burlington, VT | 44.5 | 11.8:12 | 9.3 (2.9) | 8.0 (2.9) |
| Portland, ME | 43.6 | 11.4:12 | 5.6 (1.6) | 3.5 (1.1) |
| Anchorage, AK | 61.2 | 21.8:12 | 7.1 (2.6) | 6.6 (3.0) |
| Fairbanks, AK | 64.8 | 25.5:12 | 12.0 (3.9) | 14.2 (4.6) |
Finding your roof pitch in the table
Roof pitch is the rise per 12 inches of run, and the angle is the arctangent of rise over 12. A flush-mounted array takes the roof's angle, so 4:12 is 18.4 degrees, 5:12 is 22.6, 6:12 is 26.6, 7:12 is 30.3, 8:12 is 33.7, 10:12 is 39.8 and 12:12 is 45. Twenty degrees falls at 4.37:12. Going the other way, the pitch whose angle equals a station's latitude is 12 times the tangent of the latitude: 10.7:12 in Chicago, 12.0:12 in Minneapolis and 7.9:12 in Phoenix. The two columns of Table A-1 therefore span pitches from 4.4:12 up to about 8:12 at Phoenix's latitude and about 12:12 at Minneapolis's. A tilt-up frame on a flat roof or a ground mount follows the frame's angle and not the roof's, and the row spacing a tilted flat-roof frame needs sets how many kilowatts fit.
Table A-1 gives two points per station and nothing between them. The table below draws a straight line between each station's two points and reads off the pitches in between. That interpolation is this page's arithmetic and not an NREL result, and the report warns that the model's loss need not fall smoothly with tilt: because a partly covered string produces nothing and the model runs in discrete hours, the loss can rise a little with tilt and then drop in a jump when a new string clears an hour sooner. The report says this "is not a real-world phenomenon but rather a result of model assumptions and discrete time intervals." The straight line is a planning estimate whose error is unknown, because the report tabulates no intermediate tilt. A pitch at or below 20 degrees takes the 20 degree value, and a pitch steeper than the station's latitude tilt is outside what the report modeled.
Roofs flatter than 20 degrees are not in the table. The report notes in another context that lower tilt angles inhibit snow removal by sliding, and the sliding formula slows as the tilt falls, so the 20 degree column is best treated as the low end of the loss for a 2:12 or 3:12 roof, an inference from the formula and not a tabulated value. Roofs steeper than the latitude tilt are also unmodeled, though the formula and the 80 percent average above point toward a smaller loss. Snow weight on the roof structure is a separate matter from energy loss; the roof load guide covers it.
| Station | Pitch equal to latitude | 4:12 | 6:12 | 8:12 | 10:12 | 12:12 |
|---|---|---|---|---|---|---|
| Chicago, IL | 10.7:12 | 6.1 | 5.8 | 5.5 | 5.3 | outside |
| Minneapolis, MN | 12.0:12 | 10.7 | 10.4 | 10.0 | 9.7 | outside |
| Buffalo, NY | 11.2:12 | 8.6 | 8.3 | 7.9 | 7.6 | outside |
| Boulder, CO | 10.1:12 | 5.5 | 5.0 | 4.4 | 3.9 | outside |
| Cheyenne, WY | 10.5:12 | 6.0 | 5.6 | 5.1 | 4.7 | outside |
| Salt Lake City, UT | 10.3:12 | 3.5 | 3.1 | 2.7 | 2.4 | outside |
| Boston, MA | 10.9:12 | 2.8 | 2.5 | 2.2 | 2.0 | outside |
| Burlington, VT | 11.8:12 | 9.3 | 9.0 | 8.6 | 8.2 | outside |
A 7:12 roof in Minneapolis, worked through
A 7:12 roof is 30.3 degrees. Minneapolis is at 44.88 degrees north, so the roof sits 10.26 degrees above the 20 degree point on a line that runs 24.88 degrees to the latitude point, a fraction of 0.412. The loss at 20 degrees is 10.7 percent and at latitude tilt 9.5 percent, so the interpolated loss is 10.7 + 0.412 x (9.5 - 10.7) = 10.21 percent.
At a 12-year snow-free payback the added time is 12 x 10.21 / (100 - 10.21) = 1.36 years, against 1.44 years on the 20 degree point and 1.26 on the latitude point. In dollars, the $1,967 a year of value loses $201. These are inputs chosen for the illustration, with no incentive applied.
Turning a loss percentage into years
Simple payback is the system price divided by the yearly value of its production. If snow removes a share L of the year's energy and each lost kilowatt-hour is worth the same as an average kilowatt-hour, the value falls to (1 - L) of what it was and the payback becomes P / (1 - L). The added years are P x L / (1 - L). With a snow-free payback of 12 years, each percentage point of loss adds about 0.12 years while the loss is small, and a 10 percent loss adds 1.33. The shortcut of multiplying P by L understates the added time slightly, because the lost energy is taken from a smaller total as the loss grows.
The price enters through the baseline. At the DOE benchmark of $2.95 per watt, an 8 kW system costs $23,600, and a 12-year payback implies $1,967 a year of production value. A 5 percent loss takes $98 of it, a 10 percent loss $197. A snow-free payback of 8 years implies $2,950 a year and a loss that takes 10 percent of it costs $295; one of 20 years implies $1,180 and the same loss costs $118. The table sets out the grid for a reader's own baseline. It uses no federal credit and no escalation of electricity prices, and degradation is left out so that the only moving part is snow.
Two limits come with the arithmetic. Winter kilowatt-hours are not always worth the average: under time-of-use rates the hours a snowy array misses may sit in cheaper or dearer periods, and a net metering credit pays differently by month, so a loss in December and January can cost more or less than its share of the year. Time-of-use rates and net metering are covered separately, and electricity prices themselves are outside this page. The second limit is that the percentage is a share of that design's own snow-free output, and the two designs do not have the same snow-free output, so the two columns of Table A-1 compare shares of different totals and are not a like-for-like ranking of tilt.
| Annual snow loss | Snow-free payback 8 years | 10 years | 12 years | 15 years | 20 years |
|---|---|---|---|---|---|
| 1 percent | 0.08 | 0.10 | 0.12 | 0.15 | 0.20 |
| 2 percent | 0.16 | 0.20 | 0.24 | 0.31 | 0.41 |
| 3 percent | 0.25 | 0.31 | 0.37 | 0.46 | 0.62 |
| 5 percent | 0.42 | 0.53 | 0.63 | 0.79 | 1.05 |
| 7.5 percent | 0.65 | 0.81 | 0.97 | 1.22 | 1.62 |
| 10 percent | 0.89 | 1.11 | 1.33 | 1.67 | 2.22 |
| 12.5 percent | 1.14 | 1.43 | 1.71 | 2.14 | 2.86 |
| 15 percent | 1.41 | 1.76 | 2.12 | 2.65 | 3.53 |
Years added to payback in 25 cities
At a 12-year snow-free payback, 10 of the 25 stations add less than half a year at a 20 degree tilt, and 20 add less than one year. Fargo adds 1.67 years and Fairbanks 1.64. Minneapolis adds 1.44, and a 7:12 roof there, by the interpolation, adds 1.36. Across all 201 continental stations 167 add less than a year and 120 add less than half a year.
The last column asks what price would cancel the loss. A roof that loses 10.7 percent to snow pays back as slowly as a snow-free roof priced at $2.95 divided by (1 - 0.107), which is $3.30 per watt; put the other way, the snow-free payback is restored only if the snow-country system costs $2,525 less, $2.63 per watt instead of $2.95. These are equivalences for comparing snow-country and sunbelt quotes against one benchmark. They do not say that a quote should be cut by that amount, because the benchmark is a national modeled price and local installed prices vary around it for reasons unrelated to snow. The cost per watt guide lays out that variation.
Tilt matters less than the loss column suggests. Moving from 20 degrees to the latitude angle saves at most 0.35 years at the 12-year reference at any continental station (Caribou, ME), with Lander at 0.34 and Houghton at 0.29, and the largest cut in the loss percentage is 2.4 points at Lander, WY. In the 25-city table the saving is between 0.00 and 0.27 years for every station outside Alaska. A change in pitch or racking angle that costs money has to be weighed against those few months, and also against its effect on snow-free output, which the loss percentage does not capture.
| Station | Loss at 20 degrees (%) | Years added, 20 degrees | Years added, latitude tilt | Value lost per year at 20 degrees | Price per watt that restores the snow-free payback |
|---|---|---|---|---|---|
| Phoenix, AZ | 0.0 | 0.00 | 0.00 | $0 | $2.95 |
| Atlanta, GA | 0.1 | 0.01 | 0.01 | $2 | $2.95 |
| Los Angeles, CA | 0.0 | 0.00 | 0.00 | $0 | $2.95 |
| Seattle, WA | 0.3 | 0.04 | 0.02 | $6 | $2.94 |
| Philadelphia, PA | 1.6 | 0.20 | 0.13 | $31 | $2.90 |
| New York City, NY | 1.9 | 0.23 | 0.15 | $37 | $2.89 |
| Boston, MA | 2.8 | 0.35 | 0.23 | $55 | $2.87 |
| Salt Lake City, UT | 3.5 | 0.44 | 0.28 | $69 | $2.85 |
| Boise, ID | 2.0 | 0.24 | 0.16 | $39 | $2.89 |
| Flagstaff, AZ | 4.7 | 0.59 | 0.37 | $92 | $2.81 |
| Boulder, CO | 5.5 | 0.70 | 0.49 | $108 | $2.79 |
| Cheyenne, WY | 6.0 | 0.77 | 0.58 | $118 | $2.77 |
| Kansas City, MO | 3.7 | 0.46 | 0.37 | $73 | $2.84 |
| Indianapolis, IN | 4.3 | 0.54 | 0.44 | $85 | $2.82 |
| Chicago, IL | 6.1 | 0.78 | 0.66 | $120 | $2.77 |
| Cleveland, OH | 6.7 | 0.86 | 0.74 | $132 | $2.75 |
| Detroit, MI | 5.6 | 0.71 | 0.57 | $110 | $2.78 |
| Milwaukee, WI | 7.4 | 0.96 | 0.79 | $146 | $2.73 |
| Minneapolis, MN | 10.7 | 1.44 | 1.26 | $210 | $2.63 |
| Fargo, ND | 12.2 | 1.67 | 1.59 | $240 | $2.59 |
| Buffalo, NY | 8.6 | 1.13 | 0.96 | $169 | $2.70 |
| Burlington, VT | 9.3 | 1.23 | 1.04 | $183 | $2.68 |
| Portland, ME | 5.6 | 0.71 | 0.44 | $110 | $2.78 |
| Anchorage, AK | 7.1 | 0.92 | 0.85 | $140 | $2.74 |
| Fairbanks, AK | 12.0 | 1.64 | 1.99 | $236 | $2.60 |
Show the values behind the chart
| Station | Years added, 20 degrees | Years added, latitude tilt |
|---|---|---|
| Fargo, ND | 1.67 | 1.59 |
| Fairbanks, AK | 1.64 | 1.99 |
| Minneapolis, MN | 1.44 | 1.26 |
| Burlington, VT | 1.23 | 1.04 |
| Buffalo, NY | 1.13 | 0.96 |
| Milwaukee, WI | 0.96 | 0.79 |
| Anchorage, AK | 0.92 | 0.85 |
| Cleveland, OH | 0.86 | 0.74 |
| Chicago, IL | 0.78 | 0.66 |
| Cheyenne, WY | 0.77 | 0.58 |
| Detroit, MI | 0.71 | 0.57 |
| Portland, ME | 0.71 | 0.44 |
| Boulder, CO | 0.70 | 0.49 |
| Flagstaff, AZ | 0.59 | 0.37 |
| Indianapolis, IN | 0.54 | 0.44 |
| Kansas City, MO | 0.46 | 0.37 |
| Salt Lake City, UT | 0.44 | 0.28 |
| Boston, MA | 0.35 | 0.23 |
| Boise, ID | 0.24 | 0.16 |
| New York City, NY | 0.23 | 0.15 |
| Philadelphia, PA | 0.20 | 0.13 |
| Seattle, WA | 0.04 | 0.02 |
| Atlanta, GA | 0.01 | 0.01 |
| Phoenix, AZ | 0.00 | 0.00 |
| Los Angeles, CA | 0.00 | 0.00 |
A bad winter against the average
The standard deviation in Table A-1 is the spread of the 30 yearly loss percentages around their mean. In a snowy place it is large. Among the 69 continental stations that lose 5 percent or more at 20 degrees, the standard deviation runs from 15 to 49 percent of the mean, with a median of 30 percent. For Minneapolis, with a mean of 10.7 percent and a standard deviation of 2.9 points, one deviation either side spans 7.8 to 13.6 percent.
The table adds one standard deviation to the mean as a marker for a bad winter and recomputes the payback arithmetic. It is a marker and not a forecast, because 30 winters do not define a distribution. The effect on payback is modest because a payback spans decades of winters. A 12-year payback in Minneapolis moves from 1.44 added years at the mean to 1.89 if every winter were one standard deviation worse, a difference of 0.45 years.
The spread matters in two other places. A single winter's production can fall well short of the estimate without anything being wrong with the system, which is the situation the underperformance guide sorts into weather, fault or estimate. And an owner who uses the production figure to pay a loan or to forecast winter bills should expect the winter months to vary by far more than the annual loss implies, because the annual percentages average a December that may be almost fully covered with a March that is not. The report advises against using the model for monthly or shorter periods, so no month-by-month version is offered.
| Station | Mean loss (%) | Standard deviation (points) | Mean plus one deviation (%) | Years added at mean | Years added at mean plus one deviation | Value lost per year at mean plus one deviation |
|---|---|---|---|---|---|---|
| Chicago, IL | 6.1 | 2.6 | 8.7 | 0.78 | 1.14 | $171 |
| Milwaukee, WI | 7.4 | 2.8 | 10.2 | 0.96 | 1.36 | $201 |
| Minneapolis, MN | 10.7 | 2.9 | 13.6 | 1.44 | 1.89 | $267 |
| Fargo, ND | 12.2 | 4.0 | 16.2 | 1.67 | 2.32 | $319 |
| Buffalo, NY | 8.6 | 2.0 | 10.6 | 1.13 | 1.42 | $208 |
| Burlington, VT | 9.3 | 2.9 | 12.2 | 1.23 | 1.67 | $240 |
| Cheyenne, WY | 6.0 | 1.7 | 7.7 | 0.77 | 1.00 | $151 |
| Boulder, CO | 5.5 | 1.4 | 6.9 | 0.70 | 0.89 | $136 |
What 1961 to 1990 weather cannot show
The weather file ends in 1990 and the report was published in 2017, and the report does not compare its station winters with later ones. A reader in 2026 gets thirty winters that are at least thirty-six years old. The table says nothing about whether recent winters at a station were snowier, shorter or more variable, and it cannot be adjusted for that without recent snow depth records, which the report does not contain. A station's own recent record, or a roof-specific simulation using recent weather, is the check on this label.
On place, the report says: "these values may not be representative of any particular year nor do they account for any microclimates that might be present. Rather, these values are only meant to represent a starting point for estimating snow losses in a given area before a more specialized analysis can be performed." It also warns that variable terrain means variable snow cover, which is the Flagstaff and Phoenix contrast, repeated on a smaller scale at every hill and lake shore.
The array in the study is one row of 18 modules with the roof-mount sliding coefficient and one string layout. A roof with several rows, a dormer, a valley or an obstruction at the eave changes how snow piles and how it leaves, and the report's model treats the piling at the lower edge as a fixed coefficient and not a site property. It also does not state an array azimuth, so the table says nothing about roofs facing east or west. The earlier literature the report cites ranged from 0 percent (Andrews, Pollard and Pearce, 2013) to 25 percent (Powers, Newmiller and Townsend, 2010), so Table A-1 sits inside the published range but does not narrow it for a single roof.
For a production estimate the practical consequence is on the proposal side. The PVWatts Version 5 manual lists snow among the losses a user enters as a percentage and gives it a default of 0 percent in Table 6. An estimate run at that default has assumed no snow loss at all. The guide to reading a production estimate shows where the losses line sits in a proposal, and the maintenance guide covers what an owner can do about snow and treats clearing as something to do only where it is safe. The report's own worked example is a 20 degree system in Nebraska whose snow-free simulation gives 8,000 kWh: the map suggests a designer include a 4 to 7 percent loss, or 320 to 560 kWh. The Nebraska stations in Table A-1 run from 4.8 percent (North Platte) to 6.5 (Norfolk) at 20 degrees.
Using the table on a quote
Start with the nearest station that shares your snow regime, not simply the nearest. Hills, lake shores and a change in elevation can matter more than distance, so a home on high ground above a station should look at the next station up the slope. If the nearest station is one of the sixteen without a row, use two neighbors and keep the larger. Then take the 20 degree value for a 4:12 or 5:12 roof, the interpolated value for a steeper one, and the latitude value only for a roof close to latitude pitch.
Next, ask what snow loss the production estimate used. If the proposal came from a PVWatts run at default losses, the snow line was 0 percent, and the loss belongs on top. Subtract the percentage from the annual figure, divide the price by the reduced kilowatt-hours, and compare the resulting cost per delivered kilowatt-hour with the cost per watt benchmark. At 10.7 percent in Minneapolis, a quote that matches the $2.95 benchmark is still the equivalent of $3.30 per watt of snow-free output. A quote that already includes a snow adjustment should state the figure, and a figure far below the table for a snowy station deserves an explanation.
Then decide what is worth paying for. The table shows that raising the tilt buys months at most, and the 79 stations under 1 percent add less than 0.12 years each at the 12-year reference. In the middle band, 3 to 8 percent, the loss adds 0.37 to 1.04 years. Above 10 percent, which 15 continental stations reach, snow is a main line of the payback arithmetic and a roof-specific simulation with local snow records is worth asking for.
Solar panel snow loss questions
How much energy do solar panels lose to snow?
In NREL's national model the loss runs from 0 to 15.9 percent a year across the continental stations at a 20 degree tilt, with a median of 2.5 percent. 34 percent of the continental stations lose 5 percent or more. Alaska stations run higher, up to 38.3 percent at Barrow. The weather behind the numbers is 1961 to 1990.
Do steeper roofs lose less to snow?
In the study the loss at latitude tilt is never higher than at 20 degrees at a station outside Alaska, averaging about 80 percent of the 20 degree figure where the loss is 2 percent or more. In payback terms the difference reaches at most 0.35 years at a 12-year reference. The model also shows that the loss does not fall smoothly with tilt, so treat interpolated values as estimates.
How many years does snow add to solar payback?
The added time is the snow-free payback times L / (1 - L). At a 12-year payback, 1 percent adds 0.12 years, 5 percent adds 0.63 and 10 percent adds 1.33. At the DOE $2.95 per watt benchmark, 8 kW costs $23,600, and no federal credit is assumed.
Does PVWatts include snow loss?
The PVWatts Version 5 manual lists snow as one of the losses a user enters, with a default of 0 percent. An estimate run at defaults assumes no snow cover. Ask the installer what snow loss the production figure used, and compare it with the station values on this page.
Is the data current?
No. The weather is NSRDB 1961 to 1990, simulated in NREL's System Advisor Model and published in August 2017. The 2017 edition replaced a September 2015 edition. The table is a starting point and does not reflect recent winters at a station.
What if my city is not in the table?
Table A-1 has 223 rows although the weather data covers 239 stations; sixteen, including Minot, Kalispell and La Crosse, have none. Use the nearest station with a similar snow regime, take the higher of two neighbors when in doubt, and treat the result as an approximation.
A snow loss percentage is easy to quote and hard to apply, because it depends on a decades-old station record, one array layout and a pitch the table only brackets. Used with those limits stated, it turns a vague worry about winter into a number of months, and for most of the country that number is small.
Sources and verification
- D. S. Ryberg and J. Freeman, Integration, Validation, and Application of a PV Snow Coverage Model in SAM, NREL/TP-6A20-68705 (August 2017), Appendix A, Table A-1: OSTI record 1374128. Table A-1 prints 223 station rows with mean and standard deviation of annual snow loss at tilt 20 degrees and tilt equal to latitude (NSRDB 1961-1990, 30 years per station, 18-module single row, SAM 2015.6.30 defaults). Text: 239 locations; losses 0% to 16% in the continental states, up to about 40% in Alaska; Marion roof-mount sliding coefficient 1.97 and ground-mount 6.0; sliding condition with a constant of -80 W per square meter per degree C; validation errors 9.7 to -2.2 percent (Forrestal) and 8.5 to 4.4 percent (RSF2); monthly use not advised. Checked November 3, 2026.
- NREL/TP-6A20-64260 (September 2015 edition), docs.nlr.gov document notice: The file now shows a notice that the document has been updated and links to the most recent version. The 2017 report is the edition used here. Checked November 3, 2026.
- W. Marion and K. Urban, User's Manual for TMY2s, Derived from the 1961-1990 National Solar Radiation Data Base (NREL, June 1995), Table 2-1: Station list with latitude, longitude, elevation and class for 239 stations (216 class A, 23 class B); source of the latitudes used for the tilt-equals-latitude pitch and for the interpolation, and of the list of 16 stations absent from Table A-1. Checked November 3, 2026.
- B. Marion, R. Schaefer, H. Caine and G. Sanchez, Measured and modeled photovoltaic system energy losses from snow for Colorado and Wisconsin locations, Solar Energy 97 (2013), pp. 112-121: Bibliographic record (DOI 10.1016/j.solener.2013.07.029). The model coefficients and model design are taken from the description in NREL/TP-6A20-68705, which cites this paper; the paper itself is not reproduced here. Checked November 3, 2026.
- A. P. Dobos, PVWatts Version 5 Manual, NREL/TP-6A20-62641 (September 2014), Table 6: System losses entered as a percentage of DC energy include snow cover; default snow 0%, total default losses 14%. Checked November 3, 2026.
- U.S. Department of Energy, Solar Photovoltaic System Cost Benchmarks (Q1 2025): Table 2, PV-only residential system, 8 kW DC: modeled market price $2.95 per watt DC (minimum sustainable price $2.78), 2024 dollars, owner subsidies and tax credits excluded. The page lists the 2025Q1 tables and no later edition. Checked November 3, 2026.
Every loss figure is a modeled average for a weather station using 1961-1990 data and one array layout (18 modules in a row, roof-mount sliding coefficient, SAM 2015.6.30 defaults); it is not a measurement and not a forecast for a roof. Table A-1 prints 223 of the 239 stations; Alaska, Hawaii and Puerto Rico rows are excluded from the continental counts. Interpolated pitch values are this page's straight-line arithmetic between two NREL points. The 12-year payback is a reference value, the arithmetic treats every lost kilowatt-hour as worth the average kilowatt-hour, and degradation, electricity price changes and time-of-use or net metering effects are left out. The DOE $2.95 per watt line is a national modeled price, not a quote, and no federal residential solar tax credit is in any figure. No later revision of the national snow tables was found as of November 3, 2026. Nothing here is engineering or financial advice.