Accurate pre-construction assessments are critical to securing project financing and ensuring investor confidence. However, North America's wind power industry has developed a reputation for producing energy below levels predicted by pre-construction wind resource and energy assessments. What are the reasons for over-prediction, and how can such assessments be improved?

Wind power projects involve diverse risks for investors. Along with credit, technology and interconnection risks, production risk is a key component in project evaluation. Accurate wind resource and energy assessments are therefore critical when evaluating project feasibility and obtaining financing. However, North American wind projects have been performing significantly below expectations on average, revealing a systemic bias in pre-construction energy assessments.

Concerns emerged in the early 2000s from lenders and other industry participants. Third-party energy assessments were developing a reputation for over-predicting annual energy generation, and banks and investors relying on the estimates were losing confidence. Many investors adapted by applying a 'haircut' to the estimates, limiting the ability of developers to leverage debt.

In response to the emerging underperformance trend, DNV KEMA began assembling a database to assess the magnitude of the bias and identify the key drivers. Its most recent update and assessment of project performance was completed in April.

This database contains production data from 89 North American wind power facilities, and 476 years of operation (369 independent project years) for projects that began operations in 2000 or later. The database also includes pre-construction production estimates for each project, and captures a diverse sampling of regions, project sizes and turbine technologies. This information was used to develop a distribution of operating performance relative to pre-construction estimates.

Analysis of the recently updated database shows that on average significant bias remains in pre-construction energy assessments. But the bias is substantially smaller for assessments with comprehensive, realistic and site-specific loss estimates, and do not rely on generic wind flow and wake modelling tools not tuned to site conditions. While the average industry underperformance bias is about 5%-8%, depending on the method and time period, when examined by the organisation performing the assessment, the data show average underperformance biases between about 1% and 10%, depending on the organisation that completed the analysis.

Pre-construction estimates may be improving. Our database in 2006 showed an average project performance at about 87% of estimates (i.e. a 13% over-prediction). As more projects and more years of operations were included, average over-prediction decreased to 11% (in 2008), 9% (in 2009 through 2011), and 8% for the current analysis. This trend may reflect the larger data set each year. But we expect the trend to improve as consultants' methods come more in line with operating experience. That said, there is still considerable room for improvement.

Where assessment can go wrong

The site-specific wind resource and energy assessment process typically starts with the project developer installing meteorological towers and sometimes deploying remote sensing equipment (sodar or lidar) at a site to obtain the data necessary to support an assessment. A comprehensive pre-construction wind resource and energy assessment typically includes:

  • Evaluating wind speed and wind direction data from on-site instrumentation. This requires filtering for erroneous data caused by icing, tower shadow, flow obstructions, sensor failures, and other problems;
  • Extrapolating met tower measurements from lower heights up to the wind turbine hub height after assessing how wind speeds change with distance above ground (shear), through the use of tower measurements, non-linear model approximations, and, when available and appropriate, remote sensing data;
  • Developing long-term annual hub-height wind speed and direction frequency distributions for each met tower location. In some cases, this may require using longer-term data sources that are off site (if they are available and appropriate) and/or using relationships between on-site measurements to fill in gaps or extend the data record;
  • Estimating wind speeds at the proposed turbine locations using wind flow models and other tools;
  • Estimating 'gross' energy production for the project using the power curve for the proposed turbine and the estimated wind speeds at the proposed turbine locations;
  • Estimating energy losses based on analysis of the project equipment downtime, array losses, turbine performance, electrical losses, environmental losses, and other sources of downtime or degraded performance;
  • Evaluating individual uncertainties for all of the steps listed above. This requires some understanding or estimation of how uncertainties related to wind speeds impact energy generation non-linearly as well as an overall assessment of the total uncertainty at both annual and longer term (10 or 20 year) timescales.

It should be noted that any organisation's ability to conduct sound wind resource and energy assessments is generally limited by the data collected, and often those data are insufficient to preclude introducing a bias if standard industry numerical models are relied on. As a result, experience must be relied on to a great degree. In general, the better the measured data represents the conditions the project turbines are expected to experience, the lower the uncertainty and potential for bias in the resulting energy production estimate.

Causes of underperformance

Year-to-year variations in wind resource (bad and good wind years) account for some of the observed performance bias. Also, no attempt was made to correct for build out of additional phases or curtailment losses not considered in the original estimate. These factors may contribute to the observed bias, but several other root causes for over-predictions that are likely more significant contributors include:

  • Inappropriate handling of topographic effects (often aggravated by inappropriate siting of measurement systems);
  • Higher than modelled array losses;
  • Lower equipment availability than assumed;
  • Turbine power performance effects (often associated with atmospheric conditions unlike those used to calculate the turbine power curve).

Ongoing research suggests these four points are indeed major contributors to overprediction. Some projects, however, are significantly impacted by other factors including curtailment and higher than expected losses due to environmental conditions, such as icing.