Solar producing less than expected? Weather, a fault, or the estimate

Three comparisons to make before you call anyone

Owners usually spot a problem by setting this month's total against last month's, or against a twelfth of the annual estimate. Both comparisons mislead. In the PVWatts run used throughout this page, the same 8 kW roof is modelled at 768 kWh in December and 1,277 kWh in June, so a winter month looks like a 40 percent collapse against summer without anything being wrong. A twelfth of the annual estimate (1,063 kWh here) is too high for every month from October to February and too low for every month from March to September.

The first fair comparison is the same calendar month a year earlier, once the system has a year of history. The Department of Energy's guide for buyers of homes with solar treats a year-over-year production drop of more than 10 percent as a sign that the system could have maintenance issues; a smaller difference is within what weather and normal ageing can explain. The second is the estimate's own monthly figure, which most installer proposals and every PVWatts run provide. The third is a cross-check of the meter itself: confirm that the number you are reading comes from the inverter or a production meter, not from an app that estimates generation from what it sees at the main panel.

• Same month, prior year: under 10 percent apart is usually weather and ageing; more than 10 percent deserves a look at the shape of the data.
• Same month, modelled: compare against the estimate's monthly line, never against one twelfth of the annual total.
• Same number, two sources: the inverter's lifetime kilowatt-hour counter, or a utility production meter if you have one, should agree with the app to within a few percent over a month.

Rebuild the expected number with PVWatts

If the installer's estimate is lost, or you suspect it was optimistic, rebuild it. PVWatts, run by the National Laboratory of the Rockies (formerly NREL), takes the system size, tilt, direction, a losses percentage and the address, and returns monthly output from typical-year weather. We ran its version 8.5.0 API on October 30, 2026 for a labelled example: 8 kW DC, standard modules, roof mounted, 20 degree tilt facing due south, the default 14 percent losses and a DC to AC ratio of 1.2, at Denver, Colorado (weather from the NSRDB typical meteorological year 2020 file). The model returned 12,760 kWh a year, about 1,595 kWh per kW. That figure describes this example roof only; your own run needs your tilt, azimuth and shade.

Two settings decide whether the rebuilt figure is a fair yardstick. The losses value of 14 percent is the model's default bundle: 2 percent soiling, 3 percent shading in the version 5 manual's table (the current calculator's help text describes 1 percent as little or no shade), 2 percent mismatch, 2 percent wiring, 0.5 percent connections, 1.5 percent light-induced degradation, 1 percent nameplate rating and 3 percent availability, combined multiplicatively rather than added. A roof with real tree shade needs a larger shading value, as the shaded-roof guide explains, or the model will promise production the roof never had. And the array type matters: roof mounted runs a little warmer and produces a little less than open rack.

Read the right-hand column before the totals. It is the shape of a normal year at this site, and it travels better than the totals do: a Denver owner whose December came in at 6 percent of the year's output has a December that matches the model's shape, whatever the total. Owners elsewhere should make their own run, because the shape changes with latitude and climate; a cloudy-winter site has a deeper winter dip than this one.

One day proves nothing: the spread inside a modelled month

The hourly version of the same PVWatts run shows why daily totals are the wrong unit for diagnosis. Inside the typical-year May, the model's lowest day produced 12.4 kWh and its highest 56.6 kWh, with a median of 42.6 kWh. July ran from 16.9 to 51.0 kWh, and December from 2.9 to 35.3 kWh. Every one of those days comes from a healthy, unshaded model system. A storm day at a quarter of a sunny day's output is weather, not a fault.

Typical-year weather is itself an average. The PVWatts calculator's results page prints an expected range for annual output, which it says is based on 30 years of historical weather at the location and is meant to indicate year-to-year variability. Use that range, not the single figure, as the band your first-year total should land in. A year at the bottom of the band is a cloudy year; a year below the band, with no weather explanation, is where the shapes below start to matter.

1. Compare weeks or months, never single days, unless one day is zero in full sun.
2. Check a sunny, cloud-free day against the best days in the same month of last year: peak power on clear days is the most stable number a system produces.
3. Note the date the shortfall began. Almost every fault on the table below has a start date; weather does not.

Read the shape of the shortfall

Monitoring apps show daily totals, a daily power curve and, on systems with microinverters or optimizers, a per-panel view. Those three views are enough to sort nearly every shortfall into one of seven shapes. The table is the diagnostic core of this page: find the row whose description matches your data, make the check in the fourth column, and route the problem to the person in the last.

Two shapes often overlap. A system that was low from switch-on can also slide, and a sudden drop can hide inside a cloudy month. Work through them in the order of the table, because the early rows are the ones a warranty claim depends on: the date a fault began and evidence that the system once produced more.

Low since switch-on: usually the estimate, sometimes the build

A system that has never met its estimate is most often meeting a different estimate than the one in the folder. The commonest gaps are shade the proposal understated, a tilt or azimuth that differs from the plane actually used, and a losses value trimmed to make the figure look better. Rebuild the number with the as-built facts: count the panels on the roof against the contract, read the wattage off the module label or the permit plan, and enter the real direction. If the rebuilt PVWatts monthly figures match your production, the system is working and the original estimate was wrong, which is a contract question, not a repair.

If the rebuilt figures still sit well above production, the build is the suspect. A string that was never landed at the inverter, a mis-set inverter, or a panel group wired so a shaded module drags its neighbours can all keep a new system permanently low. On microinverter or optimizer systems, the per-panel view usually shows the problem in the first clear week. On a string inverter system, the inverter display typically reports the voltage and current of each input; a string producing nothing or half its neighbour's current is the clue.

First-year owners have one modelled loss on their side: light-induced degradation, which PVWatts carries at 1.5 percent in its default losses for the first months of operation. It is already inside the estimate if the installer used the default bundle, so it cannot explain a gap of 10 percent or more.

A sudden drop: something changed on a date

A step change is the easiest shortfall to diagnose because it has a date. Scroll the daily view back to the first low day, then ask what happened that day: a storm, a power cut, electrical work in the house, a roofer or satellite installer on the roof, or a new building or fence next door. Then check the inverter for an error message, and the main panel and any solar disconnect for a tripped breaker. If the message names an isolation or insulation fault and the system starts late only after wet nights, the isolation fault guide explains what the inverter tests at sunrise and when that pattern needs the installer. A tripped breaker is the one fault an owner can safely clear; anything that trips twice is an installer call.

The size of the step points to its scale. On a system with two equal strings, losing one string roughly halves output on clear days. On a microinverter system, one failed unit removes one panel, which on a 20-panel array is about a twentieth of output; the per-panel view will show a single blank tile. The Department of Energy's inverter overview makes the architectural point: on a string inverter, an issue with one panel reduces the whole string's production, while with microinverters, shading or damage on one panel does not affect what the others deliver. The inverter replacement guide covers what happens when the fault is the inverter itself, including whether repair or replacement is the better spend.

Do not overlook the monitoring path. A new router, a changed Wi-Fi password, or a failed communications card can make the app show zero or stale data while the panels keep producing. If the inverter's own display shows power, or the utility meter's received-energy register is still climbing, the system is fine and only the reporting is down.

Reading a June shortfall, step by step

A Denver owner with a system matching the PVWatts example sees 1,050 kWh for June, against the model's 1,277 kWh: 82 percent of the monthly line. Last June, the system made 1,240 kWh, so the year-over-year drop is about 15 percent, past the Department of Energy's 10 percent marker.

The daily view shows normal output until June 9, then a step down that persists through the clear days after. The per-panel view is not available, but the inverter display shows two inputs, one carrying normal current and one carrying none. That is the sudden-drop row with a string-level cause. The owner checks the breakers (none tripped), photographs the inverter screen, and sends the date, the comparison and the photograph to the installer. Every figure here other than the PVWatts model output is an illustration with its inputs stated, not data from a real system.

A slow slide: separate ageing from dirt and trees

Every system loses a little each year. Jordan and Kurtz's NREL review of nearly 2,000 measured degradation rates found a median of 0.5 percent a year, with a mean of 0.8 percent because some systems fade faster, and 78 percent of the reported rates below 1 percent a year. Applied to the example's 12,760 kWh, a median rate leaves about 12,507 kWh in year five and the mean rate about 12,357 kWh. The guide to how long solar panels last runs that arithmetic across the warranty period. The diagnostic point is simpler: a slide of a few percent over several years is ageing; a slide of 10 percent in one year is something else.

Soiling is the usual something else, and it varies more by place than owners expect. PVWatts carries 2 percent for it by default and notes that it depends on location and weather. An NREL study of five utility-scale plants in four North Carolina counties, published in December 2023, found performance down as much as 15 percent at the peak of the pollen season, and estimated annual losses on the order of 10 percent without planned cleaning; rain alone did not restore the panels, and wet brush cleaning lifted performance by 5 to 11 percent. Those were ground plants left uncleaned for more than seven years, so the figures describe a ceiling, not a typical roof. They still show that a rainy climate is no guarantee against soiling.

Trees are the third cause, and the slowest. A tree that cleared the array at installation can grow into its morning or afternoon sky over several seasons. Because the loss arrives a few percent a year, it looks exactly like degradation in monthly totals; the difference shows up in the daily power curve, where shade cuts a notch at the same hour every clear day.

The flat-topped curve: clipping is usually designed in

A power curve that runs flat across the middle of a sunny day alarms owners more than any other shape, and it is usually the least worrying. Most systems are built with more panel capacity than inverter capacity, which PVWatts expresses as the DC to AC ratio; its default is 1.2. On the brightest hours, the panels can deliver more than the inverter is rated to pass, and the inverter holds its output at its rating. That is clipping.

We ran the same Denver example twice to size the effect. At a DC to AC ratio of 1.2 (an AC rating of about 6.67 kW), the hourly output came within 1 percent of the ceiling in 25 hours of the typical year. At 1.4 (about 5.71 kW), it sat within 1 percent of the ceiling for 434 hours, and on the model's best June day the curve ran flat at 5.71 kW for four straight hours, from 10 a.m. to 2 p.m. standard time. Yet annual output fell only from 12,760 to 12,641 kWh, a difference of 119 kWh or about 0.9 percent. The flat top looks dramatic and costs little, because it trims only the brightest hours, while the larger panel array adds output on every cloudy morning and winter afternoon.

Clipping becomes a problem in two cases. If the flat level sits well below the inverter's nameplate AC rating, something else is holding output down: a power or export limit configured during interconnection, an inverter reducing output because it is hot, or a setting left wrong after commissioning. And if the flat top appears on mild days rather than only the brightest ones, the system may be more oversized than designed. Both are installer questions, and the inverter's datasheet rating is the number to quote.

Seasonal shortfalls: snow, low sun and pollen

A system that meets its estimate in summer and misses badly in winter usually has a seasonal cause the estimate ignored. PVWatts assumes zero snow loss by default, on the basis that snow either never covers the array or is kept clear; on a snowy roof that assumption is generous, and a week of covered panels in January takes most of that week's output. Low winter sun lengthens shadows too, so a chimney, a neighbour's roofline or an evergreen that never touches the array in June can cover part of it every December morning.

Spring shortfalls in pollen country follow the North Carolina pattern above, and dry-season shortfalls in dusty climates follow the same logic with dust in place of pollen. In each case, compare the affected months with the model's monthly line rather than the annual total, and note whether a heavy rain or a cleaning brings production back. If it does, the shortfall was soiling; if it does not, it was shade or something electrical.

One panel or string well below its neighbours

Per-panel monitoring makes this shape visible on microinverter and optimizer systems; string systems reveal it only as a string-level imbalance at the inverter. When one panel produces well below the others on a clear day, look first for what changed on or above it: a new shadow, leaves, bird droppings or debris. A module is built from groups of cells protected by bypass diodes; NREL's shading test bed report notes that typical 60-cell modules carry three diodes, each protecting 20 cells. When one group is shaded or damaged, its diode routes current around it, and the panel loses roughly a third of its output at once. A panel stuck at about two thirds of its neighbours on clear days points to one cell group out, by shade, soiling or damage.

In the same NREL test, a microinverter system produced 3.7, 7.8 and 12.3 percent more than a string inverter on identical 8 kW arrays under light, moderate and heavy shade. That is why a shaded panel costs less on a microinverter system, and why a single low panel there is a local problem, while on a string system it can pull down a whole string. The guide to pricing a shaded roof has the full trade-off.

If a panel stays low after the obvious causes are ruled out, the candidates are a failing microinverter or optimizer, a loose or corroded connector, or module damage such as cracked cells or a hot spot. None of these is an owner repair. Photograph the per-panel view on a clear day, note the panel's position, and send both to the installer.

Zero output in daylight

Zero has a short list of causes, so work it quickly. First, the grid: grid-tied inverters disconnect and shut down when the utility line goes dead, a safety behaviour the Department of Energy describes in its inverter overview and the outage guide explains in full. If the neighbourhood is out, the system will restart on its own a few minutes after power returns. Second, switches: the solar breaker in the main panel, an outdoor AC disconnect and, on newer systems, a rapid shutdown initiator near the meter can all be switched off by a meter reader, an electrician or a curious child. Third, the inverter: a display showing an error, or no display at all, is an inverter call.

The fourth cause is not a fault at all: monitoring that has gone offline. Before calling for service, read the inverter's own screen or the utility meter. If either shows energy flowing, the system works, and the fix is a router or a communications card.

What the shortfall costs, priced in the site's units

A shortfall raises the cost of every kilowatt-hour the system delivers, which is the unit this site prices quotes in. At the DOE Q1 2025 residential benchmark of $2.95 per watt DC (2024 dollars, checked October 30, 2026), the example 8 kW system costs $23,600, a comparison line rather than a quote. Over the modelled 12,760 first-year kWh, that is about $1.85 per first-year kWh. If the system delivers only 82 percent of the model, the same price buys about 10,463 kWh, and the cost per first-year kWh rises to about $2.26. No federal credit enters this arithmetic; the residential credit ended for systems placed in service after 2025.

Put the same way, a persistent 18 percent shortfall is equivalent to having paid about 22 percent more per unit of energy. That is the figure to set against the price of a service visit or a tree-trimming bill when a gap stays unresolved, and the figure a production guarantee, where one exists, is meant to protect. The dollar value of the lost kilowatt-hours depends on your tariff and export rule; enter your own rate rather than a national average.

Building a case the installer cannot wave away

Installers hear about underperformance every week, and most reports arrive as a vague sense that the bill is higher than hoped. A report built on the shapes above gets a different response. Send the start date of the shortfall, the same-month comparison with last year or with the estimate's monthly line, a screenshot of a clear-day power curve, the per-panel or per-string view if you have one, and any error codes with their times. State which row of the diagnostic table the data matches and why.

If the contract includes a production guarantee, read its exact terms before writing: many measure a full year against a stated figure with a tolerance, and some adjust for weather. The production estimate guide reads that clause line by line. If the installer has closed, the inverter and panel makers' warranties still stand, and the guide to an installer that went out of business covers who to call instead.

Before you book a service visit

• Note the date the shortfall began, from the daily view.
• Compare against the same month last year and the estimate's monthly figure, not the annual total divided by twelve.
• Confirm the reading on the inverter display or utility meter, not only the app.
• Match the data to one shape in the diagnostic table.
• Reset a tripped breaker once; call the installer if it trips again.
• Rerun PVWatts with the as-built size, tilt and direction if the system was low from the start.
• Send the installer the date, the comparison, a clear-day curve and any error codes, in writing.