Abstract
This study makes use of data from six meteorological masts to examine five different data sources of roughness lengths. The data sources can be applied in most regions of the world. The consequences of applying wrong roughness lengths can impact the business case of a wind project. The experiment confirmed the preliminary expectation, as the optimized roughness approach provided better results than the remaining four approaches and, furthermore, was able to treat different tree heights. The initial test was conducted using a spatial resolution of 20 m for optimized roughness approach, while the other data sources used a greater resolution. As a response, optimized roughness approach was reused for the other spatial resolutions showing better results than the remaining approaches. One other remarkable finding associated with this study was the relationship between spatial resolution and errors in the estimation, as a resolution above 100 m provided random results with no relationship whatsoever.
Introduction
Wind turbines are an efficient technology for harvesting one of the Earth’s renewable energy sources: wind. The methods used to convert wind into energy have been employed by humankind for millenniums (Sahin, 2004), yet a more significant expansion began in the 1970s and 1980s with the Danish and Californian wind adventures (Van Est, 1999). In the recent decades, wind power has expanded in capacity installed, countries covered, and configurations from turbines on mountaintops to large offshore wind farms, which can be seen as a response to increased political support as well as improved technologies and innovations (Enevoldsen et al., 2018). One of the relatively new wind farm configurations is the development of wind farms in forested areas, which in particular seems to be an option for countries with a large portion of installed onshore wind farms (Enevoldsen and Valentine, 2016).
In 2015, 30.8% of the Earth was covered by forests (The World Bank, 2016). With the continuous development of onshore wind farm installations, it was and is inevitable to have wind turbines located in forested areas. Besides the obvious fact that wind turbines eventually would be placed in that one-third part of the Earth’s land cover, more direct reasons can explain the deployment of wind turbines in forests. Such as, the increasing costs for land acquirement, lack of space in areas with great wind conditions, and various reasons for social opposition, though majorly based on the increasing size of the wind turbines (Enevoldsen, 2016a). Simultaneously, the same development of wind turbine size has made it possible to operate wind farms in forests without the blades touching the forest canopies and, less dramatically, with a decrease in the needed deforestation. Despite the increased size of wind turbines, and thereby the distance between the lower tip height of the wind turbine blade and the forest canopy, the trees still affect the wind turbines’ performance by changing the wind conditions in the surface boundary layer (Arnqvist, 2013). This phenomenon of changes in wind conditions in and above forests has been studied extensively during the past decades (Baldocchi, 1988; Lee et al., 2011; Raupach and Thom, 1981), and recently with a specific focus from the wind industry (Bergström et al., 2013). Nevertheless, estimating wind conditions above forest canopies remains a risk for the wind industry (Enevoldsen, 2016a). The consequences of not being able to estimate the wind conditions in the swept area of the wind turbine may cause increased loads, which at one point could break vital components in the wind turbine. Besides this worst-case scenario, the annual energy production may be less than expected, which potentially would ruin a business case for the wind farm investor. Therefore, the industry and researchers have tried to determine methods that can explain and, more importantly, estimate the trees’ impact on wind conditions.
Roughness as a method to describe the impact of the trees
One way of determining the impact from forests on wind conditions is by defining the roughness length of trees, where several studies have proposed methods for using tree heights as the key input. The most dominating approach includes a combination of the roughness length, z0, and the displacement height, d. The importance of these factors can be seen from examining the logarithmic wind profile, where the two parameters are vital for estimating the wind speed, Umean, at a specific height, z as
As illustrated in equation (1), the roughness length is a vital parameter, which is why it is critical to find. Nevertheless, a great diversity in the various approaches for determining the roughness length of coniferous trees was discovered, ranging from (a) 0.3 times the tree height divided by the displacement to (b) the three height times 0.1, and to (c) the tree height divided by 30 (Freris, 1990; Garratt, 1992; Hicks et al., 1975; Jarvis et al., 1976). As a response to the lack of consensus, Enevoldsen (2016b) examined more than 12 meteorological (met) masts in a few case studies in forested areas in order to test 28 conversion methods, and found that a combination of the proposal from Hicks et al. (1975), estimating the roughness length applying (2) and the proposal from Garratt (1992) for displacement height of (3) are
Floors et al. (2018) described how roughness lengths can be applied for wind resource assessments, and the study furthermore stressed that online dataset of roughness lengths cannot replace converted in situ data. Therefore, this research aims at testing and comparing optimized roughness approach (ORA) presented by Enevoldsen (2016b) and ranking it against the different data sources presented by Floors et al. (2018), in order to provide information to academia and the wind industry about which models to trust when examining wind conditions in forested areas dominated by coniferous tree types. Thus, this research focuses only on finding the best approach using roughness as an indicator of the trees’ impact on the wind conditions. Other approaches exist, where the leaf area density (LAD) is determined by advanced measurements such as aerial LiDAR scans (Dellwik et al., 2016) to find the porosity of the trees (leaf surface (m2)/space volume (m3)). Yet, such data collection methods and subsequent analyses are expensive and, furthermore, still under development (Dellwik et al., 2016; Floors et al., 2018), for which reasons the aim of this research is to propose a method that can be used as of today.
Materials and methods
The research design applied in this study consists of a comparative analysis of data inputs for forestry information based on the usage of WAsP (2017) and WindPRO (Electro-Motive Diesel (EMD), 2017), where WAsP previously has been used for studies examining wind conditions above forest canopies (Crockford and Hui, 2007; Venäläinen et al., 2004). The software programs have been running 32 simulations for the purpose of this study, with the only difference being the source for the roughness length. The linearized numerical simulation of the wind conditions is considered sufficient for this location, as no sudden changes in elevation cause complexity in the topography and thereby potential recirculation. Nevertheless, it has to be stressed that the aim of this study is to determine the best roughness approach for linearized simulations, and not to discuss and compare linearized and numerical models.
The wind data from seven well-instrumented met masts located in the central part of Sweden have been applied to conduct the study, and as much information as possible has been presented in Table 1. Sweden is an excellent case study in that the country has experienced a rapid development of installed wind power capacity in recent years with the majority in forested areas (Enevoldsen and Permien, 2018).
Meteorological mast information.
The exact locations of the met masts have been anonymized as agreed by the data provider. Mast 5 is the reference mast that has been used to generate the wind input for estimating the conditions at the other met masts. Figures 1 and 2 will further introduce the specifications of each met mast position.
Tree heights.
Elevation map.
As mentioned in the introduction of this study, a literature review revealed diversity in the approaches to estimate wind conditions in forested areas and an even greater difference in approaches for roughness lengths of coniferous trees. The literature review was conducted using the following search words: “wind” in combination with “forest” + “roughness” using Google Scholar and ScienceDirect.com. The result of published studies was 72 papers, in which the search words were all a part of the title, abstract, or keywords. A snowball technique was effectively applied to discover additional literature, as this method has previously been found useful for literature reviews (Greenhalgh and Peacock, 2005).
The data inputs examined in this study are presented in Table 2 revealing the source, resolution, usage, and potentially preparation of data.
Input data sources.
ORA: optimized roughness approach.
Each dataset was prepared in WindPRO, where the first four data sources can be downloaded directly. The final data input was downloaded from the Swedish University of Agricultural Sciences (2010), which provided a map of tree heights that was converted to roughness length following the findings from Enevoldsen (2016b). The tree height map is pictured in Figure 1 with the white areas being water.
Figure 1 furthermore reveals that the area is heavily forested, yet with several clearings without forest (the complete red and white areas). This makes the site ideal for testing the different data input methods. The impact of the surface obstacles is tested in this study, yet the topography will affect the results and the forest’s impact on the wind conditions. The elevation is based on data from Viewfinder Panoramas (2016), which are presented in Figure 2. The significant changes in elevation are experienced over long distances, which is why the site is considered suitable for this experiment.
Figure 2 also reveals that the reference met mast is located at an upper point, for which reason wind speeds are likely to decrease when estimating the other points. The wind conditions at the reference mast are presented in Figure 3, revealing the Weibull distribution and the wind rose from the measurement at 101 m above ground level. The mean wind speed at that height is 7.7 m/s after filtering periods with icing. The overall wind conditions on-site can be classified as International Electrotechnical Commission (IEC) class IIA, following the mean wind speed and the measured turbulence intensity at the six met masts.
Wind conditions at the reference mast (Mast 5).
When analyzing and comparing the wind rose of met. mast five to the elevation map in Figure 1, it becomes clear that the reference mast will be affected by the small elevation from north-northwest (NNW), which ultimately puts WAsP and WindPRO to a test beyond the targeted obstacle of forestry. The same applies when estimating the wind conditions at the elevated location east of the reference mast. Naturally, one could have applied computational fluid dynamic (CFD) solvers to determine the impact of forests, yet such models are often based on Reynolds-averaged Navier–Stokes equations (RANS), and require information about the forests to determine drag coefficients and momentum sinks (Enevoldsen et al., 2017). Several commercial CFD software programs come with a standardized forest solver, yet Enevoldsen (2017) found that they produce less reliable results than ORA.
Analyzing the different approaches
The five different data sources all have individual specifications, making it possible to rank and set expectations for the impact of each dataset. Such ranking was applied in Table 2. The pattern in the expected ranking depends solely on the spatial resolution of each dataset. This is due to the fact that the approach with a 20-m resolution is most likely to capture changes in tree heights and, more importantly, clearings and forest edges which will change the wind conditions (Dellwik et al., 2016; Enevoldsen, 2016a) when compared to a dataset with a resolution of 1000 m, which in the worst case could miss an entire cluster of forest. Another important feature of each approach is the roughness length applied for the trees for which reason Table 3 presents the range of roughness lengths from each approach.
Roughness length definitions.
Based on Table 3, the roughness approach “E” seems to be the best choice, as this is the only approach that takes the diversity of the tree heights into consideration. Also, and based on the information regarding tree heights from Figure 2, approach “E” will have a more aggressive definition of the roughness lengths.
Examining the different approaches
This section will test the expectations from section “Analyzing the different approaches.” The analyses were carried out simulating the wind conditions for six different scenarios with the only difference being the roughness input. The results revealed an overprediction of wind speeds in all heights for the majority of the approaches. Nevertheless, a number of descriptive statistical analyses have been carried out to determine the differences between the approaches presented in Table 4, in which the results from the analyses have been summarized into differences between the measured wind speeds at six met masts and the five approaches.
Summary of differences between measured and estimated wind speeds.
ORA: optimized roughness approach.
Based on Table 4, “ORA” is the most reliable approach with an absolute mean difference of 0.13 m/s for all heights, followed by the “Global Land Cover Facility Land Cover Use” with an absolute mean difference of 0.28 m/s. The worst approach was the “Corine Land Cover 2006,” with an absolute mean difference of 0.51 m/s. Table 5 furthermore reveals the minimum and maximum difference at three heights for all the five approaches.
Spatial resolution in the range 20–80 m.
Besides elaborating the findings presented in Table 4, Figure 4 also reveals a pattern in the relationship between the differences in the estimated versus predicted and the height above ground level, as the difference decreases proportionally with the height of the estimation. The explanation for this event could be related to changes in wind conditions in the surface layer as a reaction to the forest causing the wind profile to shift from exponential inside the forest to logarithmic above the forest canopy (Enevoldsen, 2016b); however, statistical analyses based on more datasets have to be carried out to draw conclusions on any trends. The measurements at 58 and 80 m are naturally more impacted by the forest and the drag from the trees, which is why the lack of a properly defined logarithmic wind velocity makes it difficult to estimate the roughness impact (Wenzel et al., 1997).
Comparison of results on different datasets.
Discussing the importance of spatial resolution
Based on the findings presented in Tables 4 and Figure 4, the spatial resolution is not necessarily an indicator of the correctness for a roughness length approach. Nevertheless, the approach applied in ORA or approach “E” could be biased by the spatial resolution, for which reason the spatial resolution of the input data has been changed to 100, 300, 500, and 1000 m and compared in Figure 5.
Comparison of wind profiles based on different spatial resolutions.
The graphs in Figure 5 indicate that a spatial resolution of 20 m provides the best results. When examining Figure 5, it becomes clear that there is no relationship between spatial resolution and estimations at a resolution of 100 m or more. The explanation for this outcome is based on the randomness of the different resolutions, as some might capture important events at the border of each cell, while others may oversee them. This was furthermore confirmed by Floors et al. (2018), who discovered that applying the suggested displacement height of ORA for sptaial resolutions above 100 m would result in worse results. Nevertheless, additional tests were conducted in order to define a potential boundary between 20 and 100 m. The results are illustrated in the graph in Figure 6.
Resolution versus difference for ORA.
The plots in the graph in Figure 6 reveals the relationship between the differences of estimated versus measured wind speed and the size of the spatial resolution in the range of 20–80 m. The difference in percentage is illustrated using the first y-axis, and the difference in m/s is shown using the second y-axis. A descriptive statistical analysis has been carried out, revealing an R2 value of 0.75 for the relationship between the results, which has been further summarized in Table 5.
When examining Table 5, it is clear that there is an increase in difference, as the spatial resolution increases. This finding has been further analyzed using a regression analysis, where all results for the different spatial resolutions have been plotted. The R2 value for all spatial resolutions from 20 to 1000 m is 0.00784, which indicates no relationship at all.
Conclusion
This research has examined five different approaches for roughness lengths through the usage of WAsP and WindPRO. Preliminary expectations were set, which were partly confirmed, as the ORA defined by Enevoldsen (2016b) had an average difference of 2.18% between the estimated and the measured results for 101, 81, and 58 m combined. In comparison, the second best approach, which was the “Global Land Cover Facility Land Cover Use,” had a difference of 4.54% between the estimated and the measured results, despite having the greatest spatial resolution of 1000 m. The worst solution was the approach of the “Global Land Cover Map 2009” with a difference of 8.46%, using a spatial resolution of 300 m. However, a clear bias existed in the test, since ORA can apply any resolution, that is, it can make use of different roughness lengths for different tree heights. It was, therefore, decided to apply the same resolutions as the other datasets to reveal (a) if ORA still would be the best approach and (b) the impact of resolution. Figure 5 and Table 4 revealed promising numbers using ORA for the entire range of spatial resolutions, and thus, this approach is better than the other four approaches. In order to verify this statement, Table 6 presents a comparison between ORA and the matching spatial resolution for other data sources.
Comparing ORA with the other approaches using a range of spatial resolutions.
As presented in Table 6, the research statement “Estimating Wind Conditions in Forests using Roughness Lengths: A Matter of Data Input” can be considered true, as errors can indeed be minimized by using a proper approach. In conclusion, ORA is the best approach and should therefore be used for estimating wind conditions in forested areas.
One of the most significant findings was the evidence from the dataset that, while the accurateness of the estimations was superior at a spatial resolution of 20 m, no pattern seemed to exist between the resolutions exceeding 100 m, which was explained by the impact of clearings that may or may not be included at different spatial resolutions, which ultimately impacts the displacement height. It is therefore proposed to make use of the highest spatial resolution as possible for roughness maps to ensure the positive trend of installing wind power in forested areas.
Footnotes
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.