- What the estimate was promising
- The plus or minus 10 percent band, and where it stops
- An order that keeps the sky ahead of the hardware
- What the portal number actually is
- The record, and the conductors you do not open
- Frequently asked questions
- Is one bad week enough to claim the system is underproducing?
- Does falling outside plus or minus 10 percent prove an inverter fault?
A solar system that "produces less than expected" is a comparison with a missing half. The portal number is real. The expectation is a model, a sales sheet, or a memory of what the neighbor's app showed last July. Until those two are the same month, the same meter, and the same weather assumption, the gap is not yet an inverter failure. The PNNL solar photovoltaic O&M best practice tells the sites it covers to check production against the installer's estimated monthly values, or against a model such as PVWatts if that estimate was never left behind. Start there. A replacement quote that starts anywhere else is skipping the only reference the system was sold against.
What the estimate was promising
A production estimate is a stack of assumptions: the array size, the orientation and tilt, the shade the model could see, a weather file for a typical year at a chosen station, and losses for soiling, wiring, and inverter efficiency. Change any one of them and the annual number moves. What a solar production estimate needs is the longer account of that stack. This page uses it as the thing you are allowed to compare against, not as a guarantee the roof will hit.
Monthly is the grain that matches the PNNL guidance. An hour can be a cloud. A day can be a storm. A month still moves with the season, but it is long enough that one dark afternoon does not indict the inverter. If the installer left a month-by-month table, use that table. If they left only a yearly total, you cannot yet say March is "behind" by splitting the year into twelve equal pieces. Winter and summer are not equal on most roofs. PVWatts, which the PNNL page names as the fallback when no estimate was left, can rebuild a monthly shape. The rebuilt shape is still a model. Label it as one.
The plus or minus 10 percent band, and where it stops
The same PNNL page says solar insolation can vary about plus or minus 10 percent from average, and that troubleshooting starts when production falls outside those expected ranges. Read the scope with the sentence. It is guidance for the systems that guide is written for. It is not a permit rule, not a warranty test, and not a promise that every roof will land inside 10 percent of a sales proposal. A proposal that already used an optimistic shade file can miss by more than weather, while the array is healthy. A year that is actually sunny can beat the model. The band is a reason to look, not a verdict.
Suppose, as hypothetical arithmetic only, the installer's estimate for this month is 900 kWh and the meter records 760 kWh. The shortfall is about 16 percent, which sits outside a plus or minus 10 percent weather band around 900. That arithmetic says "start the checks below." It does not say the inverter has failed, and 900 and 760 are not a national month. Discard them when you have your own estimate and your own meter. A 940 kWh month against the same 900 would sit inside the band and would not, by this guidance, be the moment to open a fault investigation.
An order that keeps the sky ahead of the hardware
| Check, in this order | What a miss can mean | Where the longer account is |
|---|---|---|
| This month against the monthly estimate | Weather inside or outside the band above | what a production estimate needs |
| Soiling on the glass | The biggest O&M issue in that PNNL guide, often hard to see | dirt or damage on a panel |
| Shade the model did not include | A new tree, a vent, or a string pulled down by part of the array | how shade affects PV output |
| Clipping on bright middays | The inverter holding AC at its rating while DC is higher | inverter clipping |
| A fault that remains after the weather returns | The inverter or the array, not the sky | signs an inverter is failing |
Soiling is first among hardware suspects because the PNNL page calls it the biggest operations and maintenance issue and notes that it is often hard to see. Wash frequency is a site fact on that page, not a calendar you can copy. It depends on rainfall, tilt, and local pollutants. An arid, low-tilt array may need a monthly wash in summer. A rainy roof at a tilt near latitude may never need one. Follow the module manual. A pressure washer and a household detergent are not a default method.
Shade and clipping can both flatten a day for opposite reasons. Shade removes irradiance from cells and can drag a series string down with the shaded part, which is why the shade article owns that mechanism. Clipping leaves the irradiance in place and lets the inverter refuse the surplus above its AC rating. A system that was designed to clip on clear middays will "miss" a proposal that assumed every DC watt became an AC watt. That miss is in the design ratio, not in a failing transistor. Weather shutdowns that look like either of those, and are neither, are separated in solar inverter faults and the weather.
What the portal number actually is
The app figure is whichever meter the installer pointed at: inverter AC output, a production CT, or the utility's net meter. Those three disagree on purpose. Inverter AC output does not include energy the house used on the way to the utility meter, and it does not subtract nighttime standby in the same way a net bill does. A production CT on one conductor can be installed backward or on the wrong leg. The utility net meter shows imports minus exports, which falls when the house uses more, even if the array did not.
Compare the portal with the estimate only after you know they share a boundary. A proposal written as "annual AC kilowatt-hours at the inverter" and a bill that shows net kilowatt-hours are not the same series. What self-consumption changes is the difference between solar used on site and solar exported. Using more of your own solar can make the utility bill look better while the inverter total stays flat. That is not a production shortfall.
The record, and the conductors you do not open
The PNNL page asks that a service and alarm log be kept. The useful entries are the month, the estimate, the portal total, the alarm text, and a note about dirt, new shade, or a hot afternoon. A qualified person uses that log. A reset that clears the screen and deletes the alarm text is described, with its limits, in resetting a solar inverter.
The same guide states that the National Electrical Code, as that page reads it, requires main AC and DC disconnects within sight of the inverter. That is a statement about the code text the guide is using. It is not a measured distance you can pace off, and it is not a procedure for opening those disconnects yourself. Modules stay live in daylight. Pulling connectors to "test a string" puts you on a live DC circuit. The homeowner's job ends at the log, the photograph of the glass, and the call.
Frequently asked questions
Is one bad week enough to claim the system is underproducing?
Not against the PNNL guidance used here. Troubleshooting starts when production falls outside the expected range, and the expected values that page names are monthly. A dark week inside an otherwise ordinary month can still match the estimate. Keep the week in the log. Judge the month.
Does falling outside plus or minus 10 percent prove an inverter fault?
It proves that the comparison deserves the checks in the table, on the sites that guidance covers. Soiling, shade, a mismatched meter, and a proposal that ignored clipping all sit outside the inverter and can move the month by more than weather. The fault check comes after those.
