Most of the loss tree is determined by the equipment and the climate. Two items are determined by your particular site, and they are the two most often filled in with a default.
Teams reviewing yield assumptions can keep the engineering model separate from the record of review work; learn more is one option for documenting time and handoffs.
Reviewed August 9, 2026.
Shading
Three kinds, and they behave differently.
For independent background on photovoltaic performance, resource data and current industry practice, consult World Bank.
Horizon shading — hills, distant buildings. Predictable, calculable from the site's horizon profile, and stable for the life of the plant.
Near shading — a chimney, a parapet, a neighbouring building, a tree. Site-specific, and the item that a real survey exists to capture.
Row-to-row shading — the array shading itself at low sun angles. A design parameter, traded against how much capacity fits in the space.
Why shading is worse than its percentage
Because it is not proportional. A module partially shaded can drag down its whole string, depending on bypass diode configuration and inverter architecture.
So a two per cent shaded area does not mean a two per cent loss. It can mean considerably more, and the difference depends on layout decisions that a shading percentage in a summary does not reveal.
Which is why the question is whether a survey was done, not what the number is. A surveyed 3% is information; an assumed 2% is a placeholder that happens to be smaller.
Soiling
Dust, pollen, bird droppings, agricultural residue, snow.
It varies enormously by location — a site near a quarry, an unpaved road or intensive farmland behaves nothing like a clean suburban roof, and a regional default will not distinguish them.
It accumulates between cleanings, so the assumption in the model implicitly assumes a cleaning regime. If nobody cleans, the modelled figure was never going to hold.
And it partly self-clears. Rainfall removes much of it in temperate climates, which is why soiling assumptions in wet regions are low and in arid ones are not.
What to ask
"Was shading surveyed or assumed?" The single most useful question about a site-specific loss.
"What soiling rate, and based on what?" A named regional figure is better than nothing; a site observation is better again.
"Does the soiling assumption presume cleaning?" And if so, at what frequency — because that obligation then sits with you and may condition a guarantee.
And "how does the layout handle partial shading?" Module-level electronics, string layout and inverter choice all change how much a shaded module costs.
What changes over the plant's life
Trees grow. A hedge at two metres in year one is a different obstruction in year twelve, and nothing in the model anticipates it.
Buildings appear. Neighbouring development is outside your control and inside your production figures.
Soiling patterns change with land use nearby — a new road, a changed crop.
None of this is degradation and all of it looks like degradation in a production chart. Separating them requires the irradiance-corrected number, which is the recurring point of this whole section.
The cheap thing to do
Photograph the site horizon at commissioning, in four directions, with a date.
Ten minutes, and in year twelve it is the only evidence of what the site looked like when the estimate was made. It costs nothing and nobody does it.
The short version
- Three kinds of shading: horizon, near, and row-to-row — predictable, site-specific, and a design parameter respectively
- Shading loss is not proportional to shaded area, because a partially shaded module can drag its string down depending on layout
- The useful question is whether shading was surveyed or assumed; a surveyed 3% is information and an assumed 2% is a smaller placeholder
- Soiling varies enormously by location, accumulates between cleanings, and partly self-clears with rainfall
- A soiling assumption implies a cleaning regime, and that obligation may condition a performance guarantee
- Trees grow and buildings appear, none of it is degradation, and all of it looks like degradation without an irradiance-corrected number