How to Check Wind Farm Wake Effects: How Large They Are (Wake Is One of the Largest Energy Losses; a 15–40% Figure the Review's Discussion Calls Wake AEP Loss While Its Fig. 9 Is Labelled an Uncertainty Plot; Estimated Wake Loss Exceeds All Other Predicted Losses Combined; a Predicted Energy-Uncertainty Range From Wake of About 6%), Why They Are Under-Estimated (Ørsted 2019: Blockage and Wake Under-Forecast Across Its Portfolio, Likely Industry-Wide; External Wake 'a Bigger Known Unknown'), How Neighbours Change the Answer (Danish Waters: 9–11% Wake Loss as Built in 2021, 19–21% in the 2030 Scenario), What a Chinese Micrositing Case Shows (28-Turbine Offshore Layout: 10.58% Average and 15.58% Maximum Single-Turbine Wake Loss Against the NB/T 10103-2018 Guidance of Below 8% and Within 15%, as Quoted in the Project Report) and Which Models and Controls Exist (the Gauss-Curl Hybrid Model's Documented Deep-Array and Long-Distance Discrepancies and the Cumulative-Curl Model Offered to Fix Them; Wake Steering's Measured 14% Downstream and 4% Pair Gain Over a 10° Sector)

Short answer. Wake loss is one of the largest items between a wind farm's gross and net energy, and the published evidence says it has been both large and under-estimated. Lee and Fields' 2021 review finds that turbine wake causes a substantial portion of energy loss with large variation among estimates, that estimated wake loss is larger than the predicted non-wake loss (all other categories combined), and that total-loss predictions in eight studies ranged from 9.5% to 22.5%; in its discussion the review states wake AEP loss estimated at 15–40% citing its Fig. 9 — a figure the paper's own Section 4 (closing paragraph) says shows the uncertainty of loss percentages, not AEP loss, an inconsistency the reader should carry — gives the range of predicted energy uncertainty from wake as about 6%, and calls external wake from neighbouring farms "a bigger known unknown" than internal wake. Ørsted told investors in October 2019 that its forecasts had under-estimated both blockage and wake across its portfolio, that its new model showed slower wind-speed recovery and higher wake effects, that third-party estimates had trended more positive than its own, that under-estimation was likely industry-wide, and that the whole industry will see higher wake effects from neighbouring farms as offshore build-out accelerates. The neighbour effect now has a government-commissioned number: DTU's 2025 study for the Danish Energy Agency simulates 9–11% wake loss for Danish farms as built in November 2021, doubling to 19–21% under the 2030 scenario, with higher percentage losses in low-wind years. A Chinese case shows the design trade-off in practice: a January 2026 sea-use report for a 500 MW, 28-turbine offshore project compares four layouts at 10.58–11.62% average and 15.58–18.52% maximum single-turbine wake loss, notes that even the best exceeds the NB/T 10103-2018 guidance of below 8% average and within 15% per turbine, and argues that compressing spacing further to meet a sea-area utilisation ratio would sharply raise wake and affect turbine safety. On tools, FLORIS — released under a BSD 3-Clause (open-source) licence — describes itself as incorporating steady-state engineering wake models; for one of them, the Gauss-curl hybrid model, its developers document poor predictions for turbines in many wakes and for long-distance interactions and offer the cumulative-curl model to address them; wake steering has been measured in the field at about 14% more energy on a downstream turbine and 4% for the pair over a 10° sector. This guide reads open-access papers, a company announcement, a government-commissioned report, a Chinese project document and standard records; it quotes no operating-farm wake measurement from China.

1. How large wake losses are, and how uncertain

The review categorises losses by the IEC 61400-15 working group's framework, which splits wake into internal, external and future wake effects (the last based on confirmed or predicted new project development or decommissioning), and where a source gave only a range (for example 5–15%) it used the arithmetic mean. IEC 61400-15-1:2025 itself defines a framework for assessing and reporting site suitability conditions — long-term meteorological and wind-flow characteristics, their integration with turbine and balance-of-plant characteristics, and traceable documentation — while excluding the analytical procedures of IEC 61400-1 and related parts.

CNWattHub reading: the 15–40% figure comes from the review's discussion, which cites a figure the paper elsewhere labels as an uncertainty plot, so it should be carried as an indication of spread in the literature rather than as a confirmed loss range; the review's own point is the spread. The reviewer's first question to an energy-yield report is therefore not "what is the wake loss" but "is it reported separately as internal, external and future wake, with the model and its validation named", since the review found external wake the least-supported category.

2. Why wakes have been under-estimated

Ørsted's 29 October 2019 announcement describes two effects its forecasts had under-stated: blockage — wind slowing as it approaches the turbines, with a global farm effect larger than the sum of individual effects — and wake within farms and between neighbouring farms. Its new model, still being refined, suggested slower wind-speed recovery and higher wake effects; it had factored in a more extensive offshore build-out that raises neighbour wake; and because most third-party estimates had trended more positive than its own, it concluded that under-estimation of blockage and wake was likely an industry-wide issue.

Model developers document the same weakness from the inside. Bay et al. (2023) report that a comparison of the Gauss-curl hybrid model with historical data from three offshore farms showed poor wake predictions for turbines experiencing many wakes and for interactions over long distances, and present the cumulative-curl model as improving accuracy in situations with large wake losses and long-distance recovery while matching the earlier model for single- and few-turbine cases. FLORIS, the BSD 3-Clause-licensed (open-source) framework in which these models live, describes itself as controls-focused software incorporating steady-state engineering wake models, in development since 2013.

CNWattHub reading: the two discrepancies the modellers name for the Gauss-curl hybrid model — deep arrays and long-distance recovery — arise in exactly the conditions of large offshore clusters, which is where Ørsted found the bias. For a project in a dense development zone, the yield report should state which wake model version was used and whether it was validated against an operating farm of similar depth.

3. Neighbours: the external wake number

DTU's 2025 report for the Danish Energy Agency simulated five meteorological years (2019–2023) with a mesoscale model under three scenarios (no farms, farms as of November 2021, 2030 deployment projected in Q1 2024). It concludes that Danish wind farms as built in 2021 cause wake losses of 9–11%, doubling to 19–21% under the 2030 scenario; that percentage wake losses are generally higher in years with lower mean wind speed; that mean wind speeds in Danish waters vary annually by up to ±1 m/s; and that planning must account for wake effects and promote cross-border cooperation to reduce cumulative wake impacts.

CNWattHub reading: a yield report dated before the neighbouring lease areas were awarded can only contain their wake if its "future wake" category was populated with confirmed or predicted development, which the IEC-framework definition allows. The check is whether that category has been populated with the planned build-out around the site, and whether the loss is stated as a range over wind years rather than a single long-term mean, since the Danish result moves with the wind year.

4. A Chinese micrositing case

The sea-use demonstration report (January 2026, draft for review) for the 华润汕尾红海湾五 offshore project, published for notice by the Shanwei municipal government, compares four 28-turbine, 500 MW layouts with neighbouring-farm wakes included:

The report quotes NB/T 10103-2018 clause 6.1.6 as advising that whole-farm average wake loss should preferably be below 8% and single-turbine wake loss controlled within 15%, acknowledges that its 10.58% and 15.58% exceed both, and argues that compressing row and column spacing further to meet a sea-area utilisation ratio of at least 6% would sharply increase wake and affect turbine safety. It also lists the layout principles it attributes to the standard — rows preferably perpendicular to the dominant wind-energy direction, unequal spacing allowed, analysis of mean and extreme wind, inflow angle and shear for turbine safety, and mutual influence with neighbouring farms kept within safety and wake requirements — and names a CEEIA group standard (T/CEEIA 561-2021) for energy-yield calculation at the offshore micrositing stage. NB/T 10103-2025, the revised micrositing specification, was published on 18 December 2025 and takes effect on 18 June 2026, applying to onshore and offshore micrositing, with the 2018 edition repealed from the same date; neither text was read for this guide.

CNWattHub reading: the case is useful because it shows the numbers a Chinese developer actually reports and the constraint that pushes them up — sea-area utilisation pulls turbines together while the wake guidance pushes them apart. By CNWattHub arithmetic, the spread between the best and worst layouts here is about 25 GWh a year (2,159.342 − 2,134.020), or roughly 1.2% of net generation, between layout options that differ in turbine positions and arrangement. Two cautions: the 8%/15% figures are read second-hand from the project report, not from the standard, and they are 宜 (preferably/should) guidance in the 2018 edition, whose 2025 successor was not read.

5. Controls: what wake steering has measured

NREL's 2019 field campaign at a commercial wind farm had two turbines implement a schedule of yaw offsets; for two closely spaced turbines it measured an approximate 14% increase in energy on the downstream turbine over a 10° wind-direction sector and a 4% increase for the combined pair.

CNWattHub reading: these are gains over a narrow direction sector on a pair, not farm-wide annual gains; a vendor claim of farm-level wake-steering uplift should be reconciled to a sector-weighted calculation and to the controller's effect on loads, which is outside this guide's sources.

6. A checking sequence built from the sources

1. Decomposition. Internal, external and future wake stated separately, with model and version named.
2. Validation. Whether the model was validated on a farm of similar array depth; which model version is used and whether the deep-array and long-distance discrepancies documented for the Gauss-curl hybrid model apply to it.
3. Neighbours. Planned build-out around the site included; loss given as a range over wind years.
4. Benchmarks. The project's average and maximum single-turbine wake against the 8%/15% guidance and the reasons for any exceedance.
5. Bias. Whether the forecaster has back-tested against operating data, given the documented industry under-estimation.
6. Uncertainty. Wake's contribution to the energy-uncertainty budget stated explicitly, with the review's about-6% range of predicted wake uncertainty as a reference point.
7. Standards. IEC 61400-15-1:2025 for the assessment framework; NB/T 10103-2025 from 18 June 2026 for Chinese micrositing.

Scope limits

This guide reads Lee and Fields (2021), Ørsted's 2019 announcement, DTU's 2025 report for the Danish Energy Agency, Fleming et al. (2019), Bay et al. (2023), the FLORIS README, a Chinese offshore project's sea-use demonstration report (draft for review) and the records of NB/T 10103-2025, NB/T 10103-2018 and IEC 61400-15-1:2025 (texts not read). It quotes no operating-farm wake measurement from China, no wake-model validation dataset, no IEC 61400-15-2 energy-yield text and no turbine-load analysis. Not engineering or investment advice.