Back to the programme printer.gif Print

Delegates are invited to meet and discuss with the poster presenters in this topic directly after the session 'How does the wind blow behind wind turbines and in wind farms?' taking place on Tuesday, 11 March 2014 at 16:30-18:00. The meet-the-authors will take place in the poster area.

Alfonso Pérez-Andújar DTU Wind Energy, Denmark
Alfonso Pérez-Andújar (1) F P
(1) DTU Wind Energy, Roskilde, Denmark

Printer friendly version: printer.gif Print

Presenter's biography

Biographies are supplied directly by presenters at EWEA 2014 and are published here unedited

As a physicist (5-year BSc) with 2 years of experience in a co-owned eco-sustainable dairy business, Alfonso Pérez-Andújar underwent a transition from academic life to a time-pressed environment such as a cheese factory, where you have to deal with complex machinery, real-time environmental control and of course a living product that will be out on the shelves. This early transition pushed him into the engineering field, which is why he left for Denmark to study a MSc in Wind Energy. His company is wind and solar-powered.




Long-term corrections (LTCs) are methodologies commonly used in wind resource assessment (both in research studies and in industry) to extend a short wind climatology (about 1—3 years) into a long-term one (about 10—50 years), as it is believed that such extension will include longer cycles of the wind and so we will be able to better estimate the future wind climate. The question is whether the LTC methods help to predict something really unknown such as the future wind climate at a site. If so, which method is best?


LTC methods work by correlating short-term measurements at a target site to the short-term period (i.e. the period concurrent to the site observations) of a longer “reference” time series (a long-term observation from a nearby site, a dataset from analysis or reanalysis data or results from numerical weather prediction models). From the concurrent period, some correction factors are established, by means of which we can transfer the long-term time series onto the target site (via a linear transformation of each data point of the long-term reference time series, or via a manipulation of its Weibull sector-wise parameters). Thus, a long-term corrected time series will span a long period of time and constitute the wind climatology that could have been measured at the target site, had measurements started earlier. This is why LTCs are not in essence, as often termed, predictions of the future wind climatology, but rather a could-have-been hypothesis regarding a past time. The energy yield of a long-term corrected climatology is often assumed to give a trustworthy idea of the future energy yield. This is the same as assuming that the long-term corrected climatology is representative of the future, which is a reasonable assumption only if the climatology of the area is known to vary mildly with time. Here we investigate the validity of this hypothesis using high-quality measurements at the Høvsøre site (Denmark) and focusing on a variety of LTC methods, which are found in the literature often classified as regression and non-regression methods (Riedel et al. [1], Nielsen et al. [2], Woods [3], Mortimer [4], and also summarised in Liléo et al. [5] and Rogers et al. [6]. Rogers et al. proposed their own method too).

Main body of abstract

Indeed, the assumption of the past being representative of the future has been object of study in recent years. Liléo et al. [5] investigated, for a given past period of reanalysis data, how well (always within this period) different “past” windows of wind speed represent a fixed “future” window of subsequent years. They did this for each grid point over a certain focus region, using wind speeds obtained from the Twentieth Century Global Reanalysis Version II (20CRv2). In order to get an idea of the past's representativeness of the future, the authors defined an error by taking the percentage difference in mean wind speed of the “past” and the “future” periods. They concluded that the mean wind speed of the near past is not necessarily the best predictor of the future mean wind speed, as well as that each grid point has an optimum length of “past” window, i.e. the number of “past” years needed to get the best prediction (i.e. the minimum percentage error) is specific to each grid point.
A similar approach is followed in this thesis over a total past period spanning 1999-2013. The short-term observations at Høvsøre span 2005-2013, while the long-term reference data span 1999-2013 and come from the Weather Reanalysis Forecast model (WRF). The long-term reference data are outputted at a single grid point located near to the measuring meteorological mast. “Past” and “future” windows will thus always lie inside the time frame 1999-2013. In this case, however, all possible “past” windows contain long-term corrected hourly wind speeds within 2005-2013 (which shall henceforth be referred to as long-term corrected “observations”) instead of pure reanalysis data as in the study carried out by Liléo et al. On the other hand, “future” windows contain hourly-averaged 10-minute observed wind speeds measured at Høvsøre.
This scheme is reproduced for several different LTC methods, and in order to determine which of them is the most effective at representing the future, the aim in this work is to calculate the bias ratios of “past” parameters obtained from the LTC methods and “future” parameters taken from observations. These parameters are defined as biases and are simply ratios of future and past Weibull scale factor A, Weibull shape factor k and mean wind power density P. The closest these ratios are to 1, the safer it is to assume that past long-term corrected “observations” are representative of future observations. Furthermore, these results are compared to bias ratios obtained using short-term (1 year) measurements at the target site instead of long-term corrected observations, i.e. taking “past” windows from observations and comparing them to “future” observations. The purpose was to see if long-term corrected “observations” generated via LTC methods are anyway better than pure short-term observations recorded at Høvsøre, in terms of predicting the aforementioned parameters.
The different regression LTC methods are also separately applied to both wind velocity components, in order to obtain long-term corrected u and v. After combining these again, long-term corrected time series for both wind speed and direction are obtained.


From the results, two very different conclusions can be drawn. Firstly, long-term “observations” stemming from the different LTC methods show various degrees of success at recreating long-term observations that were actually taken at Høvsøre during the same time period. The could-have-been climatologies for certain “past” windows were thus compared to actual observations at Høvsøre that were recorded during the same time period, in order to find out which method yields the most trustworthy LTC in terms of A, k and P. This first test has yet nothing to do with future prediction, but rather simply shows which method is able to best reconstruct a past dataset. In this regard, the so-called Weibull method yields the best results of all non-regression methods, but in general regression methods give less variable results (with respect to A and P) as a function of different chosen “past” windows.
Secondly, it is seen that a method’s ability to recreate past long-term corrected “observations” that are similar to the observations that were actually recorded is generally unrelated to the effectiveness of these past long-term corrected “observations” at representing the future at the target site. As a matter of fact, LTC wind climatologies are found to be generally less effective (regardless their length and the length of the future period to be predicted) at predicting the future P than the short-term site observations.
As a side note, when applying LTC methods directly to u and v, LTC direction is found to match well the direction that was actually observed at Høvsøre at the same time period, but interestingly enough, LTC A and P bias ratios are farther apart from 1 than those obtained when applying the same LTC methods solely to wind speed.

Learning objectives
For various reasons, the measuring conditions at Høvsøre are quite unique. However, it is a coastal site and therefore the reanalysis long-term reference data used presents large deviations at certain direction sectors. Future work is therefore needed at non-coastal sites, or at sites where long-term reference data comes from another mast. Also, offshore sites and sites with complex terrain are worthy of investigation. Finally, although somewhat separate from this topic, are neural networks.

1. Riedel V., Strack M., Robust approximation of functional relationships between
meteorological data: Alternative measure-correlate-predict algorithms, Proc. EWEA,
2. Nielsen M., Landberg L., Mortensen N. G., Barthelmie, R. J., Joensen A., Application of
measure-correlate-predict approach for wind resource measurement, Proc. EWEA,
3. Woods J. C. and Watson S. J., A new matrix method of predicting long-term wind roses
with MCP, Journal of Wind Engineering and Industrial Aerodynamics, Vol 66, n. 2, Feb
1997, pp 85-94.
4. Mortimer A. A., A new correlation/prediction method for potential wind farm sites,
Mortimer, Proc. BWEA, 1994.
5. Liléo S., Berge E., Undheim O., Klinkert R., Bredesen R. E., Long-term correction of wind measurements, state-of-the-art, guidelines and future work, Elforsk report 13:18, Jan. 2013.
6. Rogers, A.L., Rogers, J.W. and Manwell, J.F., Comparison of the
performance of four measure-correlate-predict algorithms., J. Wind Eng. Ind.
Aerodyn, 93, 243–264, 2005.