Solar Loss Waterfalls: Measured, or Just Assumed?
A loss waterfall is meant to explain where the energy went between what the sun offered and what the plant sold. Many do not. They fill the bars with typical percentages carried over from the design model — two percent for soiling, one for mismatch, one for availability — and put whatever is left into a bar called “other”. The chart always balances, and it tells you almost nothing. Ellume Vector builds the waterfall only from measurements: every lost kilowatt-hour is attributed to the cause that produced it, priced at your tariff, labelled as measured, derived or modelled, and anything that cannot be explained is shown as an honest residual.
Every asset manager has seen the monthly variance slide. Production came in under budget, and a tidy waterfall shows why: a little soiling, a little temperature, a little downtime, a sliver of “other”. It looks rigorous. The question nobody asks is how each bar was calculated. Too often the honest answer is that most of them were not calculated at all. They were assumed, and the only figure on the chart that was actually measured is the shortfall at the bottom.
A waterfall built from typical percentages always balances. That is exactly the problem.
Who this matters to
Asset managers explaining variance to owners, O&M teams deciding where the next repair budget goes, and investors comparing operators across a portfolio. If a loss figure is going to drive a decision or appear in a report, you need to know whether it was measured or assumed.
Where typical loss percentages come from
Before a plant is built, an energy yield model has to estimate its losses without any operating data. So it uses assumed loss factors — for soiling, module mismatch, wiring, availability and the rest — based on industry experience and the site's design. For financing a plant that does not exist yet, that is the right approach. The trouble starts when those same factors are carried into operations. Once a plant is running, it produces data every few minutes. An operating waterfall that still says “soiling: 2%” is not reporting what happened at your site. It is repeating what the model expected to happen at a typical one.
Why an assumed waterfall always balances
Build a waterfall by applying fixed percentages, and the last bar is simply whatever is needed to reach the metered total. Call it “other”, “unexplained” or “weather and other” — it is a plug. Because the plug absorbs any difference, the chart always adds up, however wrong the other bars are. That creates two quiet failures. A real fault — a dead string, a derated inverter, an unrecorded curtailment — disappears into the plug instead of appearing as its own bar. And the assumed bars send money in the wrong direction: a site budgets for cleaning because the chart says soiling, when the real loss was an inverter.
What a measured waterfall looks like
- •Measurements only. Ellume Vector accounts for every kilowatt-hour between what the sun offered and what you delivered, using measurements only. No industry-standard “typical loss percentages” are used anywhere in the platform. If a loss cannot be measured, it is not claimed.
- •Each loss measured on its own terms. Downtime, clipping, string mismatch, grid curtailment, soiling and high cell temperature each get their own figure, built from the plant's own data. Curtailment — output your grid operator instructed you not to produce — is separated out as evidence for a compensation claim.
- •Attributed to a cause. Losses are tied to the fault that caused them, which is what turns a chart into a decision: “You lost 4,100 kWh last month, and 2,600 of it was this one inverter.”
- •Priced at your tariff. Every loss carries a money figure at your tariff, because repair decisions are made in currency. If the tariff has been assumed rather than configured, Vector says so, so a revenue figure never reaches a board without its basis being clear.
- •Labelled, balanced and honest about the gap. Every figure states whether it is measured, derived or modelled, so you always know whether you are looking at a fact or an estimate. The breakdown is required to add up to the whole, and the gap that nothing accounts for is shown as an unexplained residual against the physics model rather than swept into another bar.
Same month, two waterfalls: an illustrative case
Numbers illustrative. A solar site's physics model, driven by the month's measured irradiance and temperature, expected 4,000 MWh. The meter recorded 3,640 MWh, a shortfall of 360 MWh. Here is the same shortfall explained two ways.
| Loss | Typical-percentage waterfall | Measured waterfall |
|---|---|---|
| High cell temperature | 120 MWh (3% assumed) | 95 MWh |
| Soiling | 80 MWh (2% assumed) | 30 MWh |
| String mismatch | 40 MWh (1% assumed) | 25 MWh |
| Downtime | 40 MWh (1% assumed) | 140 MWh — 110 MWh from one inverter |
| Clipping | 20 MWh (0.5% assumed) | 22 MWh |
| Grid curtailment | Not recorded | 30 MWh |
| Other / unexplained | 60 MWh (plug) | 18 MWh unexplained residual |
| Total | 360 MWh | 360 MWh |
Both charts reach the same total. Only one is useful. The typical-percentage version points the team at soiling and a cleaning crew. The measured version shows that soiling was minor, that a single inverter was responsible for almost a third of the shortfall, and that 30 MWh of curtailment was never recorded and may be recoverable. At an illustrative tariff of $60 per MWh, that one inverter was worth $6,600 that month. The 18 MWh residual is small and visible, which tells you how much of the month is still unexplained instead of hiding it.
Four questions to ask of any loss waterfall
Was each bar measured at this site, or assumed from a model? Which equipment or event caused each loss, and could you send someone to fix it? Is each loss priced at the real tariff, and is that tariff configured or assumed? And how big is the unexplained bar — and does it change from month to month, or is it whatever was needed to make the chart add up? A waterfall that cannot answer these questions is describing a typical plant, not yours.
Frequently asked questions
- What is a loss waterfall?
- A chart that starts from the energy a plant could theoretically have produced and steps down, one loss at a time, to the energy it actually delivered. Each step shows how much one cause cost.
- Are design-model loss assumptions wrong?
- No. They are the right tool before a plant exists. They become the wrong tool once the plant is running, because measured data can replace every assumption with what actually happened at your site.
- Should the unexplained residual be zero?
- Not necessarily. A small, visible residual is honest: it tells you how much remains unexplained. A waterfall that always balances perfectly because a plug bar absorbs the difference is far more misleading than one that shows its gap.
- Does Ellume Vector use any default loss percentages?
- No. Every loss in the Vector breakdown is built from the plant's own measurements, and every figure is labelled as measured, derived or modelled. If a loss cannot be measured, it is not claimed.
