Technology effects in repowering wind turbines

This research investigates, analyses, and quantifies the technological effects of wind turbine repowering (ie, where old turbines are removed and new turbines are installed at the same or a very close location, including the enhanced performance in energy production). In these cases, it is assumed that both old and new turbines are subject to the same wind regime, other than because of technological elements, such as hub height, and thus it is possible to isolate the effects of new technology from the effect of changing local wind conditions. This research is based on the analysis of empirical data on repowering turbines in Denmark and Germany, and on historical production data available for the Danish component of the data set. Technological innovations are expected to enable new wind turbines to capture more energy at the repowering site, mostly through larger rotors and higher hub heights, and this is what this study has analysed. The results show that new turbines in repowering projects are twice as high, have three times the rotor diameter, nine times the swept area, six times the nominal power, and nine times as much electricity as the old turbines. However, the most significant improvement is probably the increase of capacity factor of 7.1% on a per ‐ turbine basis, or 9.7% on a per ‐ production basis.

Repowering a wind farm implies dismantling the existing wind turbines and installing new turbines of a larger size with new technology 7 at or near the positions of the old turbines. Although repowering can apply (but has not been done so to date) to offshore wind farms, in this research, we focus on the experience of repowering onshore wind facilities.
The structure of this paper is as follows: Section 2 introduces the background to repowering wind turbines. Section 1 presents the methodology, model, and main data issues; Section 2 represents the results of our research, the technology effects of repowering in Germany and Denmark on the most significant technical characteristics of the turbines, and the impact of repowering on energy production for Danish wind turbines. Finally, some conclusions are drawn in Section 3.
Throughout the paper, the following definitions will apply: • Repowering: the process of replacing existing wind turbines with new turbines, which either have a larger nameplate capacity or more efficiency, resulting in a net increase of power generation (according to del Rio et al 8 ).
• Dismantled: refers to any turbine that has been decommissioned; removed refers to turbines that were decommissioned and dismantled in the context of a repowering project; and new refers to the replacing of turbines.
• Repowering year: the year the new turbine was commissioned.

| BACKGROUND-REPOWERING WIND TURBINES
Repowering wind turbines (or wind farms) brings a number of benefits. First, and most important, repowering will increase performance and electricity production of wind projects, as demonstrated under different circumstances by previous studies. 1,4,[8][9][10][11] This increase in performance is partly because the sites with the best wind conditions were often used in the 1980s or 1990s. 12 Compared with a greenfield project, financing conditions tend to be better for repowering projects because the wind resource is known already and planning costs will be lower. 6,13,14 In some cases, parts of existing infrastructure might be usable, 15 11,16,17 A theoretical case study for a wind farm in India has shown that energy yield could increase by a factor of four. 18 Other studies have reported an increase in capacity factor by 10 percentage points. 14 Interestingly, research based on actual wind farm repowering cases found that while maintaining the same rated power, electricity production was doubled. 19 FIGURE 1 Age distribution of wind fleet in selected countries. Sources: Global Wind Statistics, 29 Wind in Power 2017, 30  Repowering of wind turbines also offers advantages for the whole electricity system 8 and the society at large. In general, repowering will lead to a reduction of reactive power consumption and voltage variations. 18 Turbines have evolved to better support the grid by adding increasingly complex features, such as low-voltage ride-through. Further, new turbines have lower rotational speeds with reduced noise emissions. 20 Bigger rotors and slower rotational speeds provide a less visually intrusive and more pleasant view than fast-rotating turbines. 21 In terms of environmental effects, wind farm repowering has a number of benefits. Fatalities for raptors and other birds are reduced, as shown by a research in California, which found that repowering resulted in a reduction of fatalities by 83% for raptors and 87% for all birds for the same amount of energy generated. 22 In Mediterranean mountain ecosystems, repowering was found to reduce the relative mortality of skylark males as compared with new turbine installation. 23 Repowering impact on global warming has also been investigated, and researchers found that the impact of removing the old and installing the new turbines (and other works) is "clearly offset by the benefits of increasing the generation of electrical power from renewable sources." 24 The visual effect was investigated based on a real case, and it was found that the repowering wind farm project achieved a 37% power increase with no additional visual effects. 25 Last but not least, local acceptance is also usually higher for repowering projects compared with greenfield developments. 26,27 A significant portion of the installed European Union (EU) wind fleet will come to the end of its lifetime between 2020 and 2030. 28 Approximately, 3.3 GW of the wind turbines installed in the EU by the end of 2017 were 20 years and older. This group, along with the approximately 18 GW of turbines between 15 and 19 years old are the obvious candidates for repowering ( Figure 1). Notwithstanding this, there are cases where younger turbines can be suited for repowering, and this would include some of the 33 GW of turbines between 10 and 14 years old. The largest markets for repowering in the EU are Germany, Denmark, Spain, and Italy. The repowering market is also large in the United States and India, with about 1.1 and 0.3 GW of wind turbines being 20 years and older.
Two countries that have been frontrunners in wind energy have accumulated significant experience with repowering so far: Denmark and Germany. 32 Since 2001, Denmark has supported repowering through various incentive programmes, which led to the repowering of a significant amount of the oldest wind turbines. Fifty-six percent of turbines installed before 2000 and 84% of turbines installed before 1994 had already been removed by the end of 2017. More than 3200 turbines were dismantled in Denmark before 2018.
In Germany, about 5470 MW (approximately 2040 turbines) of wind power capacity has been installed before 2018 in repowering projects.
Those turbines replaced about 2900 old turbines (2280 MW). 33 Table 1 shows the annual evolution of repowering in Germany.
So far, the technological effects, efficiency gains, and performance improvements of actual repowering projects have not been researched in a detailed manner. Some evidence is available from theoretical studies. 1,10,34 Some case studies have been performed on an individual wind farm basis (see, eg, Castro-Santos et al 34 and Villena-Ruiz et al 20 for actual repowering wind farms in Spain), then the focus has been economic or techno-economic, rather than technological. Also, at a national level, the focus of the assessment or modelling of repowering has been from an economic (eg, de Simón-Martín et al, 27 ) or techno-economic (eg, Serri et al., 17 ) perspective.
The objective of this research is therefore to fill this gap by analysing the technological and performance (in terms of energy production) changes due to wind turbine repowering independent from locational changes in wind resource. In addition, unlike previous research, this research focuses on a large number of cases in which different analytical tools have been applied.

| Methodology
There are two main methodological elements in this research. First, technology trends were uncovered through graphical representation. Second, regression analysis was performed to show how variations in performance are linked to the key technology trends. The data sample consists of sets of two wind turbines, one removed and a second one newly installed in the vicinity (see the next paragraphs) around the same period. Data include the technical characteristics of both turbines and the corresponding energy produced during a reasonably long period of time.
Microsoft Excel and Visual Basic for Applications were used as the main modelling tools before applying regression analysis to the results. A Visual Basic for Applications macro selected pairs of removed/newly installed turbines under the following conditions: • maximum distance between them was 1500 m; • new installation occurred between 30 days before and 500 days after dismantling the old turbine.
The maximum distance of 1500 m was decided after Monforti and González-Aparicio found that "uncertainty in the wind farm locations of the order of a few kilometres is not expected to visibly decrease the quality of the wind power assessment at national level." 35 Figure 7 in that article shows that a separation of 1500 m hardly affects the simulated wind conditions.
The period of 30 days before and 500 days after dismantling the old turbine was chosen to follow the reasonable project management process while not letting excessive time impact the available technology at the time of repowering. A much longer time period might involve completely disconnecting the old and new turbines (thus not a repowering project).

| Data sources and data availability
For Denmark, the publicly available Danish master data register for wind turbines (Stamdataregister for vindkraftanlaeg) provided by the Danish Energy Agency was used. 36 The register contains data on geographical coordinates, turbine model, rated power, hub size, rotor diameter, date of commissioning and decommissioning, and annual energy production for most wind turbines. However, some wind farms do report global production data; in such cases, the register presents the average production per turbine. On the negative side, this database lacks wind farm names; therefore, it is not possible to unequivocally associate old and new wind farms and even turbines.
In Germany, a publicly available register of all notifications (eg, commissioning and decommissioning) of renewable energy installations since August 2014 is available from the Federal Network Agency for Electricity, Gas, Telecommunications, Post and Railway. 37 For wind energy, the register contains the geographical coordinates, turbine model, rated power, hub size, rotor diameter, and date of commissioning and decommissioning per wind turbine. The register also specifies that if a new wind turbine was commissioned as part of a repowering project. Wind turbines that have been commissioned before August 2014 are not included in the sample. Also, operators are not obliged to report dismantling of old turbines, thus we cannot assume that the set of decommissioned turbines is complete. Energy production data were provided by the Federal Network Agency to the JRC under a confidentiality agreement and cover the years 2012 to 2016 only. Unfortunately, given the limited period, these data did not enable energy production analysis as in the Danish case.
For Denmark, data about repowering projects from as early as 2000 were available, whereas for Germany, the sample only includes repowering projects from the last 4 years. The data available varies according to the specific parameter analysed because of the different levels of completeness in the data fields. However, there are two broad categories of parameters: technological or structural elements relate to the turbine characteristics (eg, hub height, rotor diameter, and specific power) and energy production resulting from the interaction of structural elements with the wind; in other words, turbine data vs production data.
Any data point containing valid data in structural fields was used for the respective analysis. However, the analysis related to energy production was restricted to those cases where at least three full years of data were available in order to accommodate the variability of the wind resource.
For Denmark, the data sample that could be used for the subsequent analyses included 232 pairs of turbines for the technical parameters, which is just 6% of all dismantled turbines. Data on energy production was available for 200 pairs (5.5% of all dismantled turbines). The data sample for Germany with complete information about the technical parameters contained 442 pairs of turbines and data on energy production was not used. The appendices contain detailed information about data issues and the data improvements performed.

| RESULTS
In Denmark, three "waves" of repowering can be identified from the data: For Germany, data about repowering projects and production data were only available for 2014 onward. Thus, in the remaining of this section we will focus mainly on results for Denmark and will compare them with German data (whenever data were available).  In Germany, the average power rating of removed wind turbines was between 900 and 1300 kW, while the new turbines showed more similar ratings to Denmark (for 2014 onwards), which was a capacity of between 2500 and 3050 kW on average.

|
The average (or mean) of power rating increases in Denmark was 11.6 times throughout the period, but the median was 8.8 times. This suggests a high number of cases where the power rating increases (times) were low, as was the case during the second part of the period, from 2008 ( Figure 4). For the average power rating of new turbines (1833 kW) versus the average of old turbines (302 kW), the increase is sixfold.
New turbines in Denmark were significantly more powerful than removed turbines during the first repowering wave, when they averaged 11 to 12 times the rated power of removed turbines. Later waves saw reduced differences and in the latest wave new turbines were only 5 to 6 times as powerful as decommissioned turbines. This trend from larger to smaller differences between new and old turbines has been observed as well in

| Rotor diameter
The rotor is perhaps the element of the turbine that has a more direct relationship (through its swept area) to the energy produced. In Denmark, the rotor diameter of turbines, both for the decommissioned and new sets, was on average more homogeneous than in the case of turbine power The relative increase of rotor diameter has been, on average, very stable: new turbines had a rotor three times as large as old turbines (in metres), equivalent to a 9-time increase in swept area. Given the direct relationship of swept area to energy produced, it could be concluded that the increase in rotor diameter is the single most important structural element impacting an increase in energy production.

| Hub height
The hub height is the last structural element analysed here that is strongly affected by repowering, and there are two reasons for this: first, larger rotors naturally require larger towers; second, at higher altitudes, winds are stronger and steadier, conditions that warrant more energy extracted and lower structural loads than turbulent winds.

| Specific power
Perhaps the most interesting result from repowering is the evolution of specific power, the ratio of power rating to swept area. This is because the profile value, which measures how valuable a wind turbine generation profile is to the electricity system, increases in line with a reduction in specific power. 39 All other elements being equal, a reduction in specific power results in higher capacity factor. A support mechanism based on the number of years of production promotes that turbines produce as much energy as possible during these years, whereas support based on a number of full-load hours promotes that turbines produce less energy per year in order for support to last longer.
During the second wave, both new and old turbines had similar specific power, albeit the former already below the latter: 373 vs 392 W/m 2 .
During the third wave, the situation reversed, with new turbines having significantly lower specific power: 354 vs 411 W/m 2 . This reversal can be linked both to technological evolution-the average specific power of a wind turbine has been decreasing with time, 42 and because of changes in 2014 to the Danish support scheme, which incentivised lower specific power turbines. 43 Interestingly, in Germany, average specific power was between 411 and 422 W/m 2 for removed turbines and 461 and 316 W/m 2 for new turbines in 2014 to 2017, revealing a clear downward trend that increased the profile value of new turbines.
It is important to note that the lowest specific power of new turbines in Denmark steadily happens in the 2014 to 2017 time period. This is consistent with the support scheme change in 2014, as mentioned earlier, to reduce the incentive for high specific power turbines. As Lena Kitzing stated, "The change (to the support scheme) in 2014 has also eliminated much of this incentive by using swept area instead (at least for 70% of the support duration calculation)." 44 Looking forward, we see elements that could differentiate repowering project turbines from greenfield ones that are related to the location of old turbines. In the past, turbines were placed considerably closer to human settlements than new wind farms. Therefore, repowering the old site is unlikely to get planning consent in a number of cases. Even in the cases when planning consent is obtained, it could come with restrictions in hub height or rotor diameter that can be similar to those placed on new projects in much of the United Kingdom (eg, Kelmarsh or Dunmaglass wind farms) or Ireland (eg, Meenadreen Extension), where 2+ MW machines have hub heights limited to 70 m.

| Electricity production
In this work, a number of wind turbine pairs did not contain reliable production data. Some others did not contain three full years of data. As a result, only 200 pairs make up the energy-production-related analysis. The average number of production years of old turbines was 16 and 12 for new turbines until the end 2018, the end of the data sample.
The German data were not used for studying the effect of repowering on production because they were not available for a time period long enough to obtain reliable results.

| Absolute annual energy production and increase
The annual energy production was calculated as the weighted average of the individual turbines for the years of production, which were considered complete. This includes from the year after commissioning to the year before decommissioning.
where the average annual turbine production (AATP) of a removed turbine is the sum of its annual energy production from year a (the year following commissioning) to year b (the year before decommissioning). In the case of new turbines, b is the last year of data (2018).
The increase in absolute annual energy production for each (t) repowering turbine is the result of subtracting AATP for the removed turbine (td) from AATP of the removed turbine (tr): As expected, the new turbines have increased annual energy production ( Figure 8)

| Specific annual electricity production
Specific annual energy production shows the amount of electricity that a turbine generates irrespective of rotor size, ie, per square metre of swept area. This indicator therefore collects technology improvements mostly due to turbine efficiency, power rating, and hub height.
Results consistently show that new turbines have increased specific energy production: 99% of the cases (197 pairs) show an increase after repowering ( Figure 9). The trend in both old and new turbines is towards higher specific electricity production in 2014, but 2015 shows a change. With no data after 2015, it is not possible to define whether the 2015 effect is short or long term. On average, specific electricity production increased by 320 kWh/m 2 /yr-a 45% increase-from 702 to 1021 kWh/m 2 /yr.

| Capacity factor
The capacity factor (CF) is the percentage of actual production to theoretical production should the turbine have been producing continuously at the rated power. It is expressed either as a percentage or as the equivalent number of hours, on annual average.
The net effect of repowering on CF is clearly positive. Figure 10 shows the annual average capacity factor for all the turbines removed or In Figure 11, the CF of the old turbine is shown on the horizontal axis and the CF of the new turbine on the vertical axis. The black line divides pairs according to whether the new (top left) or old (bottom right) turbine has a higher CF. In the large majority of cases, the new turbine has a higher capacity factor. In 14% of the cases, the new turbine had a lower CF than the replaced turbine. Some cases could be explained by data issues (eg, errors in reporting of electricity production). Most often, those cases related to early repowering projects when turbine technology was more similar between the removed and new turbine (Section 2.1).
The results show that the old turbines had a capacity factor of 22.4% on average, whereas the new turbines reached 29.5%. This is a significant 7.1% increase in capacity factor on a per turbine basis.

| Regression analysis
In order to relate the variation in energy performance to the technological elements that caused them, and to identify the most important ones, we decided to use regression analysis.
Regression analysis has been applied in the energy field, eg, by Lee and Yang, 45 Fumo and Rafe Biswas, 46 and Ma et al. 47 In the wind energy field, Arias-Rosales and Osorio-Gómez 48 have applied regression analysis to wind turbines based on estimates of the cost of energy.
Among the statistical models commonly used, linear regression analysis has shown promising results because of the reasonable accuracy and relatively simple implementation when compared with other methods. 46 Under the multiple linear regression approach, the selection of the explanatory variables is a key issue because irrelevant variables have negative effects on the process. 49 To ensure that the multiple linear regression approach is the appropriate methodology, it has been tested so that the input variables selected are linear (ie, all of them follow a normal distribution) and independent from each other.
The correlation between the technological changes brought about by repowering was explored (ie, the increases with time in hub height, rotor diameter, and power rating) between the repowered and new turbines, and the increases in annual energy production (AEP) in each case. The regression analysis took AEP increase (ΔAEP) as the dependent variable and all other variables as independent variables. The reason for defining AEP as the dependent variable is that the final objective of repowering is increased production, which is also the natural result of the changes in technological variables.
The regression analysis used Minitab statistical software. Initially, the following predictor variables were considered: The first analysis trials quickly showed that two variables were not statistically significant (ΔHH, YR), as the P value was above.05 for these predictor variables.
The two remaining independent variables (ΔTR and ΔSP) were found to be of statistical significance for the regression model.
The coefficient of multiple determination, R 2 , takes an acceptable value of 93.37%, and adjusted R 2 is 93.30%. A small Mallows' Cp value of 3.0 was obtained, indicating that the model is sufficiently precise. It was concluded that the model fits the data well.
Other assumptions that are required for multiple regression analysis to give a valid result were checked as well. They are shown in Table 2 and summarised here: • the independent variables are significant, as P value is below.05 for both variables.
• The variance inflation factor (VIF) is 1.29 for the two independent variables of the regression model, indicating that the predictor variables are not correlated.
• The residuals show an approximate constant variance.
• The residuals are normally and randomly distributed. Note that the ratio between turbine rating and swept area is very important: it is used by turbine manufacturers to design products better suited for local wind conditions. For example, Siemens Gamesa currently offers two different rotor diameters (155 and 170 m) for their 5.8-MW wind turbine, and two different rated powers (3.4 and 4.5 MW) for their 132-m rotor diameter turbine. 50 Because of this reason, we carried out further analyses: the swept area (SA, m 2 ) was tried instead of specific power (W/m 2 ) in the regression analysis. Somehow, although R 2 reached 95%, the results showed a VIF value of 7.83 for both statistically significant variables ΔTR and SA, indicating a possible problem of multicollinearity.
Another analysis was based on the previous regression model result that change in hub height (ΔHH) was not significantly related to the increase in energy production. To explore this further, the data set was split into two subsets: from 2000 to 2005 and 2007 to 2015, to examine possible partial time correlation. However, the results of the analysis showed again that ΔHH remained a nonsignificant variable for the regression model.
The results of the regression analysis are shown in Table 2, whereas Figure 12 shows the adjusted probability plot and the residuals of the regression analysis.
Time-organised, plotted results in Figure 12 confirm what has been observed earlier: because of faster technological progress, repowering effects in the past few years had greater impact on turbine efficiency than early repowering projects. On the other hand, from the regression analysis, variations in increased electricity production can be mainly attributed to variations in two explanatory variables: turbine rating, ΔTR, and specific power, ΔSP.
The coefficients can be explained as follows: • for each increase of turbine rating by 1 kW, annual electricity production increases by 3.22 MWh.
• For each decrease in specific power by 1 W/m 2 , annual electricity production increases by 8.62 MWh.
The lack of a direct relation between the increase in energy production and the increase in hub height, or with time, came as a little surprise to the authors. This is because the technology has improved over time (YR), and because an increase in hub height is directly related to an increase in energy production. See, for example, a recent statement by Vestas, the market leader, "With hub heights of 152 m, the (…) customised tower solution increases the project's annual energy production by unlocking new wind resources at higher and more consistent wind speeds." 51 We think that the reason is that the impact of the increase in turbine rating and specific power is much more significant than the impact of having higher hubs.

| CONCLUSIONS
This study has, for the first time, assessed the technological effects caused by a large set of real repowering projects and their impacts on energy production on a turbine-by-turbine level. The average repowering occurred has brought nearly a three-time increase in rotor diameter, or a ninetime increase in swept area, and a doubling of hub height. New turbines were between 6 and 11.6 times as powerful as decommissioned turbines, depending on how the average was taken.
The results show that repowering has resulted in an increased capacity factor of 7.1% on per turbine basis, or 9.7% on a per production basis.
Interestingly, during the first years of repowering, new turbines had significantly higher specific power than the turbines they replaced, and this trend reversed in the 2014 to 2017 period. This was linked to changes to the financial support instrument being used at the time in Denmark, which from 2014 promoted turbines with low specific power.
New turbines have a higher annual energy production compared with the removed turbines. On a weighted average, production as a result of Also, the study shows that specific energy production (per m 2 swept area) has increased in 99% of the cases. On average, specific electricity production increased by 320 kWh/m 2 /yr.
A regression analysis was performed to assess the impact of the underlying changes in technology on energy output. It showed that the increase in energy production was directly related to the increase in turbine rating and the decrease in specific power of the new turbines. On average, every additional kilowatt of rated power added 3.22 MWh to the annual energy production, and each W/m 2 of lower specific power increased annual electricity production by 8.62 MWh.
Further, this study analysed the effects of repowering on a turbine-by-turbine level. This was done to mitigate the influence of local variations in the wind resource. Of course, it is highly unlikely that in a given wind farm would substitute each turbine with a newer, larger one when repowering in practice. Repowering projects most often concern whole wind farms where both turbine and power grid upgrades are performed and wind farm configuration is optimised for energy production and levelled cost of energy, often by reducing turbine counts but approximately maintaining power density. A follow-up to this study has been proposed to analyse with empirical data repowering of wind farms in order to characterise the actual change in turbine density and energy production resulting from deployment of new modern turbines in place of older facilities at the end of their life. A further research question would also be to analyse the financial aspects of repowering, for example, was the repowering performed at the optimal time from a cost perspective?
From a societal view point, and considering the growing market for repowering in the coming years, it is important to understand if marketdriven repowering projects will also deliver the socioeconomic benefits to society. In particular, will the wind resource be utilised optimally? What are indirect economic impacts (eg, on the value of land) of repowering? These questions need answers in order to determine if repowering could be more efficient or steered by policy instruments to bring the additional value for the economy and society.