Part 1 of three. Part 2 asks how much of a PR decline you are actually allowed to recover. Part 3 asks how often you should act on it.

Over the past months we sat down with Asset Managers responsible for more than 200 MWp across four very different contexts — industrial rooftop portfolios in Portugal, C&I self-consumption in Spain, mixed rooftop-ground-carport-floating assets in Northern Europe, and utility-scale in Southeast Asia.

We asked all of them the same question: is performance management at your organisation preventive or reactive?

Every one of them said reactive. Not because their tools are bad — most run competent monitoring platforms with well-configured alarms. They said reactive because of what those tools are built to catch.

One manager of a 65 MWp rooftop portfolio described the detection path precisely: monthly comparison of actual output against an irradiance-corrected expected production model, then two to three months to isolate the cause. Another, overseeing roughly 60 MWp of mixed assets, named soiling outright as the single most underestimated risk in his portfolio — and he already has an alarm-based monitoring platform.

That is the shape of the problem. Inverter faults have alarms. Communication failures have alarms. Tracker faults have alarms. A loss that arrives at 0.05% per day has nothing. It simply becomes the new normal, and the model quietly re-baselines around it.

Performance Ratio tells you something is wrong. It never tells you what.

Performance Ratio is the right metric for the question is this asset delivering? It is the wrong metric for the question why isn't it?

The IEC 61724-1 definition is explicit about this. PR expresses the relationship between actual yield and target yield, and it accounts for losses inside the modules caused by heat, soiling, degradation or operational issues — all of them, aggregated into a single number.

So a PR that has drifted from 82% to 78% is genuinely informative. It is also completely silent on whether you are looking at a thermal artefact of an unusually hot quarter, eighteen months of accumulated degradation, a string that has been offline since March, or a contamination layer that a single intervention would remove in an afternoon.

Those four scenarios have wildly different financial consequences. Three of them are unrecoverable or already spent. One of them is money sitting on the table. PR alone cannot distinguish between them — and neither can a monthly production report.

The two metrics almost nobody tracks

If you want to isolate the soiling component, you need two numbers that are not in most monitoring dashboards.

Soiling ratio is a state measurement. It is the ratio between the output of a contaminated surface and the output of an equivalent clean surface at the same moment, under the same irradiance. It answers: how much am I losing right now?

Soiling rate is a gradient. It is the rate at which the soiling ratio degrades over time — typically expressed as a percentage of output lost per day. It answers: how fast does the loss come back after I intervene?

The distinction matters more than it first appears, and it is where most business cases fall apart.

Soiling ratio tells you the size of today's problem. Soiling rate tells you the interval — how long a recovery holds before the loss rebuilds. Without the rate, you can justify a single intervention but you cannot design a schedule, and you certainly cannot defend one to a CFO. A site losing 0.02%/day and a site losing 0.15%/day may show an identical soiling ratio on the day you measure it, and require completely different operational answers.

This is why calendar-based schedules — once a year, or twice at best, because that is what the contract says — are almost always either wasteful or insufficient. They are set without either number.

Three ways to actually measure it

There is no single correct method. There are three, with honest trade-offs.

Clean reference pair. Maintain one string (or a defined subarray) in a permanently clean state and compare it continuously against an untreated equivalent under identical orientation and configuration. Cheapest by a wide margin, highest signal quality, and it produces a directly auditable number. The weakness is entirely operational: it only works if the reference is genuinely maintained. A reference that slips becomes a reference that lies.

Dedicated soiling station. Purpose-built instrumentation with paired reference cells, one cleaned automatically. Accurate, standardised, and defensible in front of an investor or a lender's technical advisor. Two limitations: it is capex, and it measures a single point. On a large site with meaningful microclimate variation — proximity to an access track, a cement facility, an agricultural boundary — one station can misrepresent the whole plant.

Regression from existing SCADA data. Model expected output from measured irradiance and module temperature, then attribute the residual. No hardware, uses data you already own. It is also the weakest of the three: the residual contains everything the model failed to capture, so the quality of your irradiance sensing and your temperature correction sets a hard ceiling on how much you can trust the result. Useful for direction of travel and portfolio triage. Not sufficient on its own to justify capital.

Our own view, having run field campaigns across arid, semi-arid, tropical and coastal conditions: the clean reference pair delivers the best evidence per euro on the overwhelming majority of sites. It fails for organisational reasons, not technical ones.

What our own data can and cannot claim

We are going to be specific about a limitation in our field evidence, because a methodology article that hides its own caveats is not worth reading.

In our long-duration test campaign with Green Energy Park and UM6P in Ben Guerir, Morocco, five PV strings with identical modules and identical electrical configuration were monitored side by side under different treatment strategies, using DC-side PR to remove inverter and BOS noise from the comparison. The reference against which the treated strings were compared was an uncleaned string. The comparator strings were maintained periodically — at defined intervals — not continuously.

That design is realistic. It reflects how plants are actually operated. But it has a consequence that cuts against us: because the comparators were themselves accumulating contamination between interventions, the measured deltas understate the true gap against a permanently clean surface.

In other words, our published gains are conservative. The real distance between a protected asset and an unmanaged one is larger than our own numbers show.

We flag this because it is the exact question a competent technical advisor should ask about anyone's field data — ours included — and because the answer determines whether a number is a marketing figure or an engineering one. When you evaluate soiling evidence from any supplier, ask what the reference was and how often it was maintained. If the answer is vague, the number is decorative.

What this changes on Monday morning

Nothing in this article requires a purchase decision. It requires a measurement decision, which is a different and much cheaper thing.

Three questions worth putting to your O&M provider before the next intervention cycle:

1. What is our soiling rate, per site, expressed in %/day? If the answer is a shrug or a seasonal average borrowed from a datasheet, that number does not exist for your portfolio.
2. What is our reference, and how is it maintained? Every credible soiling figure has a reference behind it. If nobody can name yours, you do not have a measurement — you have an estimate wearing a measurement's clothes.
3. When we last intervened, how much PR came back, and how long did it hold? The recovery magnitude and the decay interval are the two inputs to every schedule decision you will make for the rest of the asset's life. Most portfolios have never recorded either, because nobody thought to measure before and after.

If you want a first-pass estimate of what a soiling loss is worth on one of your sites before you invest in measuring it properly, our performance simulator will give you an order of magnitude in a few minutes: performance.chemiteksolar.com. It is a model, not a measurement — but it is enough to tell you whether the measurement is worth commissioning.

Part 2: your PR is falling — how much of it can you get back? We break a PR decline into its recoverable and irreversible components. Part 3 then calculates how often it is worth acting.