95–98 % Performance After 20 Years — Why So Many Get This Wrong

Teaser: The data is unambiguous: the dominant problem in aging PV parks is not ageing. It is defects. Anyone reading string-level numbers and concluding that "the modules are tired" is confusing two very different measurements — and making systematically wrong decisions about replacement, reuse, and residual value.

The mistake

When a 15-year-old PV park sits twelve percent below its expected performance ratio, the reflex is almost always the same: the modules are old, time to repower. It sounds logical. In most cases, it is wrong.

The reflex conflates two distinct quantities: module degradation (the physical power loss of an individual module) and system degradation (the performance decline of the whole plant on the DC side). In practice they are treated as synonyms. They are not — and the entire economic margin of an aging asset lives in the gap between them.

What the data actually says

Pure module-level degradation for crystalline silicon sits at roughly 0.5 % per year in NREL long-term field data. DNV's current whitepaper uses the same order of magnitude — about 0.50 %/year module plus 0.14 %/year of additional system effects, for its P50 system-level standard of 0.64 %/year.

The arithmetic is simple: a defect-free crystalline module should still deliver 90 to 95 % of its rated power after 20 years. In well-aged cohorts with benign thermal histories, values of 95–98 % are common. The largest long-term field study to date — Prieto-Melo et al. (2024) with 1.25 million German PV systems totalling 34 GW — finds a mean of 0.59 %/year at system level, and explicitly notes that the often-cited 0.8 %/year from older literature overstates reality.

Operators of real aging parks nevertheless see numbers that look far worse than 0.5 %/year. That gap is the actual story.

Why the string ages faster than any of its modules

A series string is a classic weakest-link system. The current through the entire string is set by the module with the lowest short-circuit current. When modules degrade at different rates — which they always do in a real park — the operating points drift apart, the MPP tracker is forced into compromises, and string output falls faster than the arithmetic average of the individual modules.

It gets genuinely severe with defects. A module with broken cell interconnects, active PID, or a failed bypass diode becomes the limiting element for its 20 or 24 string neighbours. Once the bypass diode activates, current in that substring drops to zero and the surrounding modules operate off their MPP. PVsyst simulations put the typical penalty at roughly 1.3 modules' worth of output lost per defective module. A single failure costs more than it weighs.

DNV's 2024 whitepaper names these effects explicitly as system-level contributions that do not belong to module degradation: mismatch, resistance losses in connectors and junction boxes, corroded contact points. The IEA-PVPS Failure Fact Sheets (2025) list the same repeat offenders: PID, cell cracks, bypass diode failures, delamination, hotspots — all of them defects, not ageing.

Ageing is linear. Defects cluster.

The late-life acceleration visible in long field records — DNV's 35-year reference dataset on 18 modules (1982–2017) is the textbook example — is not a sudden onset of material fatigue. It is the accumulation of local faults: a small percentage of modules develop defects, those defects turn into disproportionately large string-level losses, and the MPP landscape becomes increasingly rugged. The park "ages" in a measurable way; the average module does not.

This distinction is operationally decisive because it flips the intervention logic. A fleet of modules with average degradation is not ready for replacement. A fleet with a few but effective defects is, by contrast, an extreme lever: surgically identifying and removing the worst 2–5 % of modules can deliver more yield uplift than any homogeneous refresh.

What this means for valuation, reuse, and business models

Valuation. The capital-market-relevant residual value of a 15-year-old park depends less on the average module age than on the distribution of individual module conditions. Without module-level diagnostics, that distribution is invisible — and any due diligence built on string-level SCADA and a clean linear degradation curve is buying or selling under systematic uncertainty.

Reuse. The second-life conversation suffers from a misunderstanding that keeps the market artificially small: the idea that used modules are leftovers. In reality, most modules pulled from a 20-year-old park sit very close to their datasheet — if you can find them. Without standardised, high-throughput testing, "reuse" becomes either guesswork or a relabelled export channel. With genuine module-level diagnostics, reuse becomes a measurable, qualified product class.

Business models. The binary logic of classical repowering — rip it all out, install everything new — systematically destroys value because it ignores the distribution of module conditions. Between "do nothing" and "replace everything" a distinct category is emerging: parks are diagnosed module by module, the defective ones are removed or repaired, the healthy ones are resorted to reduce mismatch. This decision layer — the ability to tell modules apart at scale — is exactly what the market does not yet deliver reliably.

Bottom line

The uncomfortable truth is that the number on the SCADA dashboard says very little about the technical condition of the modules beneath it. Module degradation is slow, predictable, and for healthy modules economically irrelevant after 20 years. The problem in aging parks is almost always mismatch, defects, and resistance losses — and those three problems become solvable the moment module-level data is on the table.

Treating old parks as a uniform ageing problem replaces 100 modules to neutralise the effect of three. Splitting the problem with data leaves the other 97 where they are.

Sources:

• Prieto-Melo et al., From Shine to Decline: Degradation of over 1 million solar photovoltaic systems in Germany (2024)
• DNV Whitepaper, Views on Long-Term Degradation of PV Systems (Hieslmair et al., 2024)
• Jordan & Kurtz, NREL Photovoltaic Degradation Rates — An Analytical Review
• IEA-PVPS T13-30 (2025), Degradation and Failure Modes in New Photovoltaic Cell and Module Technologies
• EPRI / Fregosi et al., Guidance on PV Module Replacement, IEEE PVSC 2020
• PVsyst, Mismatch Basic Principles (Bypassdioden- und Current-Limiting-Logik)