Read the loss as a waterfall
The clearest way to think about a system’s output is a waterfall: start from what it should have produced against a weather-normalised expectation, then subtract each loss bucket in turn — soiling, faults, downtime, degradation, shading, clipping — until you reach what it actually produced. Each step is a different problem with a different owner and a different fix. Lumped together they are just “underperformance”; separated, they become a to-do list ranked by size.
| Loss bucket | Typical magnitude | Recoverable? | How it’s spotted |
|---|---|---|---|
| Soiling (dust) | ≈5% typical; 5–20% in long dry spells | Yes — clean | Gradual decline that recovers after rain or a clean |
| String / equipment faults | Now the #1 loss category (Raptor Maps) | Yes — repair | One string or inverter below its peers |
| Downtime / availability | ≈3% default; ≈0.5% with active monitoring | Yes — faster response | Windows of zero or missing output |
| Degradation | ≈0.5–1% per year, cumulative | No — mostly permanent | Slow, steady multi-year decline |
| Shading | Site-specific | Partly — trim, or permanent | A repeatable dip at the same time each day/season |
| Clipping / temperature | By design / physics | No — expected | Flat-topped midday output; heat-related dips |
The recoverable buckets — where the money is
Soiling is the everyday one: dust builds up, output drifts down, a clean brings it back. In Pakistan’s climate it sits at the high end of the range and can spike during long rain-free spells. String and equipment faults are now the single largest observed loss category — a dead or weak string, a failed optimiser, a derating inverter — and they are fully recoverable once found. Downtime is the quiet one: every hour a system is offline is output gone, and active monitoring is worth roughly two to two-and-a-half points of availability on its own.
These three — soiling, faults, downtime — are where a decomposition earns its keep, because they are both large and fixable.
The buckets you don’t chase
Not every loss is a fault to fix. Degradation is a slow, mostly permanent decline of roughly half a percent to one percent a year — real, but not something a truck visit recovers. Clipping (when the array briefly produces more than the inverter can pass, so the peak is shaved) and temperature losses (panels produce less when hot) are design and physics, not failures. Mistaking these for faults is how teams waste call-outs on systems that are behaving exactly as designed — which is the whole reason attribution matters.
How SPC decomposes the loss
Solar Performance Cloud measures each string against a weather-normalised expectation and attributes any shortfall to a cause — soiling, a fault, downtime, shading — rather than reporting one blended “low” number. AI grades every string every day; solar engineers verify what matters. Reading every major inverter brand into one independent view, it turns a vague sense that output is “a bit low” into a ranked, itemised waterfall you can act on, biggest recoverable bucket first.
