Solar · Sizing an array
Four numbers decide how many panels a property needs. What it uses in a day, what it draws while nobody is home, how much sun the site gets in its worst month, and how much is lost between the panel and the outlet. Get the third one wrong and the array comes up short every winter.
Start here
Daily watt-hours. Not your monthly bill, not a guess from the appliance labels, but the actual energy the property consumes in twenty-four hours. Every step after this multiplies that figure, so an error here scales all the way through.
Two ways to get it honestly. Measure it with a meter over a full week, which the household power use guide walks through. Or build it from a device list, which the power needs calculator does for you.
Measured beats calculated. A device list captures what you remember owning. A meter captures what the house actually does, including the things nobody thinks about.
The average US household uses about 10,500 kilowatt-hours a year[1], which works out near 29 kilowatt-hours a day. Sizing an off-grid array for that is expensive enough that most off-grid households reduce their load first. Cutting consumption is almost always cheaper per watt-hour than generating more.
The load nobody counts
An inverter left on idles. A charge controller runs its own electronics. Monitoring, routers, and anything with a standby light all pull continuously. None of it appears on a device list built by walking through the house switching things on.
A load only matters as watts multiplied by hours. Something drawing 45 watts around the clock uses more in a day than a 1,000-watt appliance run for one hour.
Standing draw is divided by worst-month sun like every other watt-hour. In a two peak sun hour December, one watt of constant draw needs roughly fifteen watts of panel to cover it.
Search mode on an inverter, switched outlets on standby devices, and turning off monitoring you do not read all cut it. This is the cheapest watt-hour on the property.
Peak sun hours
A peak sun hour is one hour of sunlight at 1,000 watts per square metre, which is the standard condition panels are rated at. It measures energy arriving, not time elapsed.
A December day at a northern site can hold nine hours of daylight and deliver two peak sun hours. The sun sits low, its light crosses more atmosphere, and much of the day is delivering a fraction of full intensity. Counting daylight instead of energy is the most common sizing error there is.
Latitude alone will not tell you the figure. Sites at the same latitude can differ sharply because of cloud cover, coastal fog, and elevation. Pull your own number for your own coordinates.
NREL's PVWatts calculator returns modelled output by month for any US location, tilt, and orientation[2]. Run your address, then read the monthly table rather than the annual total.
The month you want is the lowest one, which for most of the northern hemisphere is December. Write that number down. It is the one the rest of this page uses.
The method
The whole calculation is one division. Total daily watt-hours, divided by worst-month peak sun hours multiplied by system efficiency. What follows is where each of those comes from.
Array watts = daily watt-hours ÷ (worst-month peak sun hours × system efficiency)
NREL's PVWatts uses a default total system loss of 14 percent, and its manual is explicit that the total is the product of each individual loss rather than their sum[3]. Soiling, shading, mismatch, wiring, connections and nameplate tolerance all compound.
That 14 percent models a grid-connected system. An off-grid array also pays battery round-trip losses and charge controller conversion on most of what it makes. Using the grid-tied figure alone will undersize the array.
Take 0.86 for array and wiring losses, multiply by roughly 0.95 for a lithium battery round trip, and the result is near 0.82. Rounding down to 0.80 leaves a little honest margin.
Panels come in fixed sizes. Divide the array figure by the wattage of the panel you can actually buy, then round up. Rounding down is borrowing from February.
Worked example
A northern property near 43 degrees latitude. The peak sun hour figures below are illustrative, chosen to show the method. Yours will differ, and the point of the example is the sequence rather than the answer.
Measured over a week: 8,000 watt-hours a day. Well under the national average, because the household changed how it uses power before it bought panels.
Inverter idle, controller, and monitoring measured at 45 watts continuous. Across a full day that is 45 × 24 = 1,080 watt-hours. Total daily need is 9,080, rounded to 9,100 watt-hours.
PVWatts returns 2.0 peak sun hours for December at this site, against an annual average near 4.0 and a June figure around 5.5. December is the month that has to work.
9,100 ÷ (2.0 × 0.80) = 5,688 watts of array. At 400 watts per panel that is 14.2 panels, so fifteen panels, 6,000 watts.
6,000 × 2.0 × 0.80 = 9,600 watt-hours on an average December day, against a need of 9,100. The margin is thin, which is what worst-month sizing is supposed to produce.
Run the same property on the 4.0 annual average instead. 9,100 ÷ (4.0 × 0.80) = 2,844 watts, or eight panels at 3,200 watts. Roughly half the array, and half the cost.
In December that array makes 3,200 × 2.0 × 0.80 = 5,120 watt-hours against a need of 9,100. It delivers 56 percent of what the property uses, every day, for the months it matters most. That gap is the entire argument for worst-month sizing.
The same fifteen panels in June produce 6,000 × 5.5 × 0.80 = 26,400 watt-hours, close to three times what the property uses. Worst-month sizing always creates a summer surplus. What to do with it is a real design question, and it is covered on the pages about diversion loads and expansion.
Limits
The method above gets a property to a defensible panel count. It is not a site assessment, and four things can move the answer enough to matter.
A modelled figure assumes a clear horizon. Trees, outbuildings and terrain to the south change the answer, and winter shadows are far longer than the ones you see in July.
A fixed array pointed away from true south, or set at a shallow pitch, collects less winter sun than the model assumes. Mounting choices are covered on their own page.
A covered panel produces nothing. In snow country the worst month is not merely dim, it can include days at zero, which is a storage problem more than a panel-count problem.
Panels lose output slowly over decades. An array sized with no margin is sized for its first winter rather than its fifteenth.
Common mistakes
The single most expensive error, and the one worked out above. It produces an array that is fine for eight months and short for four.
Nine hours of winter daylight is not nine peak sun hours. This mistake can overstate winter output by a factor of four.
A device list captures what you switch on. It misses the load that never switches off, which in the example above was more than a tenth of daily use.
A 400-watt panel is rated under laboratory conditions. Real output is lower, which is what the efficiency figure in the division accounts for.
Losses belong on one side of the calculation. Reducing the demand figure and then also derating the panel rating charges the same loss twice and inflates the array.
Next
These three guides all start from the numbers on this page. They are in progress and will link here as they publish.
Turning daily watt-hours into days of autonomy, and what chemistry does to usable capacity.
Sizing the conversion hardware from the array figure and the bank voltage.
Where fifteen panels physically go, and why adding panels usually beats adding moving parts.
Back to Solar at Property Scale for the rest of the section.
Sources
Peak sun hour figures in the worked example are illustrative and chosen to demonstrate the method. Run your own site through PVWatts before sizing anything.