Imagine that a year ago, two wind farms you invested in became operational – one in Texas (Project 1) and one in New York (Project 2). To set your expectations for how much energy the projects will produce, you contracted an independent engineer to conduct a pre-commercial operation date (COD) energy assessment for you; in response, you were provided with a lengthy report, a P50 energy estimate and various production levels that corresponded to different probabilities of exceedance (one-year P99 and 10-year P90, for example) for each project.
The independent energy assessments for the two projects had the same P50 and the same one-year P99, and you find that after a year of production, both Project 1 and Project 2 produced 5% less than the previously predicted P50. Concerned, you return to the same independent engineer and ask for an explanation. After a detailed investigation, you receive the following response:
Esteemed wind farm owner,
We have reviewed the operational data from your two projects and re-evaluated the expected future energy output. Given the windiness of the operational period in the two regions where the projects are located, you should expect the new long-term P50 of Project 1 will be 2% lower than the pre-COD P50, and the new long-term P50 of Project 2 will be 3% higher than the pre-COD P50.
Still concerned, and now a bit confused, you demand an explanation for how this could possibly be the case. You then receive the following response from the independent engineer:
Both of the regions where Project 1 and Project 2 are located experienced lower-than-average wind speeds last year. The region where Project 1 is located is known to have relatively low variability in wind resource year to year; last year, the winds were 3% less energetic than the long-term average. The region where Project 2 is located is known to have high variability in wind resource year to year; in addition, Project 2’s production is particularly sensitive to changes in wind speed – last year, the winds were 8% less energetic than the long-term average. Both of these events represent events approximately one standard deviation off from expected production. Therefore, the data should be considered within the normal range of expected production.
You are a bit surprised that you should be expecting such vastly different fluctuations in production year to year from projects that looked so similar on paper. Had you recognized these differences prior to investing in the projects, you may have valued the projects differently, or at least taken different approaches to mitigating the production risk.
For more than 30 years, the wind industry has used the same language to understand the potential production of a wind project. The 10-year P90 and one-year P99 are terms that have decided the fate of most financed wind projects. As the wind industry matures, financial margins are tightening and off-take agreements are becoming more complicated, making it time for the language around preconstruction production estimates to evolve to allow for a deeper understanding of a project’s value and risk.
The current process of assessing the uncertainty for a preconstruction wind energy assessment has remained relatively unchanged for the last 30 years. The uncertainty around discreet steps in the energy assessment process is quantified as standard deviations of normal distributions, wind speed uncertainties are converted to energy uncertainties using a “sensitivity ratio,” and then the energy uncertainties are combined to tally up the overall project uncertainty.
Broadly speaking, the categories of uncertainty that are considered in a preconstruction wind energy assessment include measurement uncertainty, historical wind resource, vertical extrapolation, horizontal extrapolation, plant performance and project evaluation period variability. These are, in fact, the current proposed uncertainty categories coming from the International Electrotechnical Commission’s (IEC) 61400-15 Working Group, 2017. The first four of these categories are wind speed uncertainties that address the uncertainty of predicting the wind regime at the turbine locations from the measurements that are recorded on the project. Plant performance addresses the uncertainties associated with predicting the net energy from the wind regime at the turbine locations (including uncertainty in wake predictions, system availability, turbine performance, electrical losses, etc.).
The last category, project evaluation period variability, addresses fluctuations in the plant output due to varying conditions at the project over a period of concern (e.g., over a one-year period or the life of the project). The dominant, and most recognized, variability is the wind resource, which has a natural variation from year to year. System availability is another variability that might be considered in this category. The expectation with the variabilities captured in this category is that the impact they have on the overall project uncertainty diminishes over time. That is, if you are concerned with the energy output over any one year, then the variability of the wind will be a critical consideration; if you are concerned with the energy output over a 20-year period, then the variability of the wind will be less of a concern, as you expect the high-wind years to be largely balanced out by the low-wind years.
In the current process for assessing the uncertainty of a preconstruction energy assessment, the standard deviations from each of these six categories are combined to determine the shape of the expected project production profile over the evaluation period of concern. From this process, the P50, “10-year P90” and “one-year P99” are all derived. This process has served the industry well, but as our analysis methods have advanced, allowing for higher geographic and temporal resolution of project variabilities, for example, the traditional language for describing project production risk is beginning to fail us.
The project production profile for a preconstruction energy assessment, from which the 10-year P90 and one-year P99 are derived, is determined by a combination of uncertainties and variabilities. The proposed refinement to the language we use to describe the evaluation of a preconstruction project is captured in the table.
Using the framework proposed by the IEC 61400-15 working group, the uncertainties can be quantified in five categories: measurement uncertainty, historical wind resource, vertical extrapolation, horizontal extrapolation and plant performance. These uncertainties represent the distribution of possible results around the various inputs to the energy production estimate and capture the potential range of long-term P50 energy production levels of the project. It is these categories that should be evaluated to understand the risk in the long-term P50 output of a project.
The variabilities are quantified in the project evaluation period variability category and attempt to capture the year-to-year fluctuation in production, independent of the uncertainties already considered. Even if the long-term P50 value were known perfectly, the production each year will vary, which is what the variabilities model demonstrates. Generally, the largest contributor to project variability is the wind resource, followed by project availability. It is the variabilities that should be evaluated to understand production volatility on a project, whether on an annual, seasonal or hourly basis.
Change is hard. Why go through the hassle of adopting a new vocabulary to evaluate the production risk in a project? Understanding the contribution of both uncertainties and variabilities to a project’s production profile allows investors and developers to fine-tune the financing or development of a project.
To illustrate this, let’s travel back in time to when we just received the pre-COD energy assessment for the two example projects, but this time, the difference between the uncertainty and variabilities has been made clear.
Project 1 has an uncertainty of 8% of the P50, and the variabilities are 4% of the P50. This could be a project with a single measurement mast being used to predict across a large region with a relatively high capacity factor and low expected inter-annual variation of winds. In other words, Project 1 has a relatively high risk on the long-term P50 but will experience small year-to-year variation. For Project 1, additional high-quality measurements or advanced modeling would provide the most risk reduction. If feasible, a true-up assessment done after the date of commercial operation would also significantly reduce production risk.
Project 2 has an uncertainty of 4% of the P50, and the variabilities are 8% of the P50. This could be a project with multiple hub-height measurements and advanced wind flow modeling but with a lower capacity factor and high expected inter-annual variation of wind speeds. Thus, Project 2 has a relatively low risk on the long-term P50 but will experience large year-to-year production variation. For Project 2, where the uncertainties are low but high variabilities mean the variation in production year to year, or even season to season, could introduce temporary budget shortfalls, a wind insurance product or a production hedge might add substantial value.
The one-year P99 for these projects would be the same in this example, but how the production risk is treated and the potential tools to reduce the risk are very different.
As evidenced by more sophisticated methods of demonstrating value, such as the Monte Carlo statistical approach, analysis techniques have kept pace. However, the language to describe these advances has not. Any refinement worth making should provide additional clarity to the process it is describing.
Taylor Geer is service line leader for project development at DNV GL. He can be reached at (503) 222- 5590 or email@example.com.