On Closer Inspection: Wind farm underperformance

On Closer Inspection examines claims about clean energy that generate more heat than light.

Some criticisms of renewable energy don't survive contact with the evidence. That solar panels leach toxins into the soil is simply false; "wind turbines kill more birds than any other human activity" isn't close to true. The claim that wind farms often fail to deliver the energy their own design documents promised is different. It deserves examination rather than dismissal, because it is sometimes true.

The claim here is different from complaints about wind's intermittency — output that varies with weather. That's a genuine issue too, though one that storage technologies are addressing increasingly well. Here it's not that output varies from day to day, but that total output over years may fall consistently short of the figure the project was built against. The same phrase — "wind farms underperform" — often covers both, which allows day-to-day variability to stand in for the more specific charge.

The claim though is typically presented as though it applied uniformly: as if underperformance relative to design targets were an inherent feature of wind technology. It's not. Some sites consistently beat their design targets — a fact that, unsurprisingly, gets far less attention. That both outcomes are real points to an underlying systemic problem. It turns out the explanation for both is the same.

The physics works. The forecasts don't always.

To understand the yardstick by which wind farms are assessed as "underperforming" you need to understand "capacity factor" (CF). CF is the standard metric that measures actual output over a period divided by what a turbine would produce running at full rated capacity the entire time.

WattClarity, the Australian energy-market analytics outlet, observed in a February 2026 analysis that the term is technically a misnomer — once curtailment is accounted for (output the turbine could have generated but didn't, for grid or market reasons), the number tracks production rather than capacity potential. "Production factor" would be more precise. The energy sector continues to use "capacity factor."

Stockyard Hill Wind Farm, in Victoria's central highlands about 35 kilometres west of Ballarat, was designed for a capacity factor of around 40.9%. In its early operating years, Rystad Energy data placed it consistently at the top of Australia's performance rankings; in some months its capacity factor approached 50%.

Macarthur Wind Farm, also in western Victoria, was designed for a capacity factor of around 35%. Since reaching full operation in early 2013, it has averaged 24.5% — never exceeding 28.6%. In the three years to the end of 2023, its average was below 20%. Macarthur is routinely cited in online arguments about wind farm underperformance. Stockyard Hill is rarely mentioned.

Both outcomes are evidence of the same phenomenon — inaccurate forecasting, not turbine physics.

The design number was always a forecast

Two CFs matter in any wind project: the design CF embedded in finance and contracting documents, and the measured CF once the turbines have been running. The second number is based on actual measurements when the farm is in operation. The first is a forecast — and forecasts are only as good as their inputs.

A 31-kilometre grid can't see a 90-metre forest

Estimates of wind strength and duration typically come from global atmospheric reanalysis models: historical weather datasets built from decades of observations. The most widely used, ERA5, is produced by the European Centre for Medium-Range Weather Forecasts and assigns a single wind value to each grid cell — at a resolution of roughly 31 kilometres square, everywhere on Earth.

That's a lot of averaging. A 2021 peer-reviewed study published in the journal Energies compared ERA5 output against local measurements taken at six diverse sites worldwide. For simple terrain, the model performed acceptably: bias within 1% at an offshore platform in the North Sea (about as simple "terrain" as you can get), within 7% at a flat Dutch site close to sea level. For more complex terrain and land cover, though, the picture changed sharply. ERA5 substantially under-predicted resource at a high-altitude site in the US Rockies and at a mountainous site in Iran (wind speeds at high altitudes tend to be higher than ERA5's grid cell average). At Wallaby Creek, Victoria, it substantially over-predicted resource. For sites with complex terrain or land cover, ERA5 simply isn't fine-grained enough to give an accurate forecast.

Wallaby Creek sits at around 720 metres on the southern edge of the Hume Plateau, within a forest of Mountain Ash — trees that can reach above 90 metres, tall enough to fall within a wind turbine's swept area. A research flux tower at the site gave researchers a direct comparison point. ERA5's 31-kilometre cell, containing a patchwork of crops, mixed farming, and dense forest, had no way to resolve the aerodynamic roughness of that canopy.

Better tools exist; they aren't yet universal

Obviously it would be better for developers to get accurate measures for a site before the first spade goes in the ground. The most accurate way is to install on-site meteorological masts or LIDAR systems and make recordings for at least a year — these produce site-specific wind profiles that no atmospheric model can match. Unfortunately they're also expensive and slow to deploy, which rules them out as a general-purpose first-pass screening tool.

Australia now has an intermediate option. BARRA-2, developed by the Bureau of Meteorology, is a regional reanalysis built specifically for the Australasian domain and nested within ERA5, but run at substantially finer resolution: 12 kilometres for the standard regional product, and as fine as 4.4 kilometres over the Australian continent. A Monash University study tested BARRA-R2 against actual generation from 54 Australian wind farms and found it more accurate than both ERA5 and MERRA-2. At finer resolution, more of the local terrain and land cover is visible to the model — which is precisely the detail ERA5 lacked at Wallaby Creek.

Turbulence is the second local factor the models miss

A further underappreciated factor in wind farm performance is turbulence — more formally defined as "short-timescale fluctuations in wind speed and direction" — which affects turbines in two connected ways. First, bursts of strong wind increase fatigue loads on blades and mechanical components. That increases long-term maintenance requirements, but more importantly here pushes control systems toward conservative operating modes. Second, it pulls real output below the idealised power curve used to estimate energy output in the first place. The curve is derived under smooth, steady-flow assumptions; it's a complicated relationship, but in turbulent flow, a turbine reliably produces less than the curve predicts for the same average wind speed.

Like terrain effects, turbulence is site-specific — driven by local topography, surface roughness, and the spacing and orientation of turbines within the farm. Also like terrain effects, it is largely opaque to a coarse model grid. A project that doesn't account for site-level turbulence in its resource estimates starts from an optimistic baseline. The turbines encounter what the model didn't.

Underperformance doesn't mean what they'd like you to think it means

There are typically two things wrong with the argument that wind farms underperform. The first is selective framing: detractors tend not to acknowledge that some farms beat their design targets. The second is more fundamental: the design target was always a forecast. Underperformance at a given site usually traces to how accurately the wind resource was characterised before a turbine went in the ground — not to something inherent in wind technology.

Improving that characterisation — through finer-resolution regional reanalyses, strategic on-site measurement, and greater transparency about design-CF assumptions when projects are assessed — is an active area of work. It's an argument for better forecasting. It's not an argument against turbines.

This article was researched and drafted in collaboration with Claude (Anthropic).