Offshore wind energy is one of the renewable energy sources to be exploited in Europe, Asia and North America to reduce anthropogenic emission of greenhouse gases from the energy mix. At present, the offshore wind climatology and derived wind power statistical information such as wind resources, wind speed decadal trends, changes in seasonal winds and inter-annual wind speed variability have insufficient detail for siting and planning of offshore wind farms. The main reason is that observations at offshore meteorological masts are very limited in space and time. Data from global reanalysis during recent 25 years are unfortunately not consistent in time due to changes in available input and difficulty in homogenization, thus artificial trends may occur [1
Wind climate statistics must be known locally for any potential wind farm project. Wind farms are planned with around 30-year economical lifetimes. Often only one or few years of meteorological observations are available and resource estimation has high uncertainty as surface winds change considerably between years [2
]. An implicit assumption is that the mean wind resource assessed from stationary climate during the analysis period will be similar into the future.
The collected ship data in the International Comprehensive Ocean-Atmosphere Data Set (ICOADS) [3
] show wind climate trends at decadal scale over the global oceans. Wind speed from four first-generation re-analyses were compared to recently updated ship wind data at 10 m [5
], as well as to wind speeds from satellite passive microwave [6
] and altimeter [7
]. These studies report positive trends in global ocean mean wind speed. This is in contrast to the results of wind speed trend analysis over land based on atmospheric modeling for the USA [10
] and Europe [11
] and from network of meteorological stations, e.g., in Scotland [12
] and China [13
]. There is much variability in the decadal wind speed trends for different periods and observational data sets at regional scale for land and sea and include both negative and positive trends. Offshore wind power estimates at global scale based on satellite scatterometer winds show large spatial variations [14
The present study focuses on quarter-century offshore wind speed trends and addresses, with similar methodology and observations, two regional seas both highly relevant for offshore wind farming but with contrasting wind climates. In the North Sea many wind farms are in operation in the UK, Denmark, Germany, the Netherlands and Belgium, while in the South China Sea wind farms are planned with 1.2 GW for 2020, and beyond this time, up to 3.95 GW according to the China National Energy Administration including the Hainan Floating wind farm demo project (20 MW), Guangdong Dongfang wind farm (350 MW), Huaneng Hainan Wenchang #1 offshore wind farm (200 MW) and Guangdong—Xuwen offshore wind power demonstration project (48 MW) (under construction) [16
Regional studies using scatterometer winds demonstrate good wind energy potential in the North Sea [17
] and South China Sea [19
]. Wind resource analysis at higher spatial resolution from satellite synthetic aperture radar (SAR) confirms the good wind energy potential for the North Sea [21
] and South China Sea [23
] and reveals spatial variability at kilometer scale often with sharp gradients near coastlines, where most offshore wind farm projects are located.
The most common problem for assessment of offshore long-term wind climate is the lack of meteorological mast observations. Therefore, satellite-based sea surface wind data are used in this study. Data from SAR satellites cover near-coastal regions, but their temporal coverage is too low for trend analysis. In contrast, passive microwave satellite wind products cover only open oceans but are frequent in time and have a quarter-decadal data record. For long-term wind trend analysis, it is assumed that regional effects rather than local effects govern potential long-term changes. Therefore, we examine what useful information we may extract from the Special Sensor Microwave Imager (SSM/I) ver. 7 that has accuracy sufficient for trend analysis for decadal changes at 1% [24
]. We compare the SSM/I wind speed data to available in-situ data and mesoscale model results. We investigate the strength and weaknesses of the different data sets. Meteorological observations are only available in the North Sea while SSM/I and output from mesoscale simulations using the Weather, Research and Forecasting (WRF) model are available for both seas.
The first objective of this paper is to examine wind power statistics at hub-height of offshore wind turbines at 100 m from SSM/I and WRF for both seas.
The second objective is to assess the amplitude of temporal wind speed variations and how these vary across the North Sea and the South China Sea by performing long-term trend analysis at monthly and yearly time scale using Student’s t-test for significance based on SSM/I.
The analysis is based on 26 years of SSM/I ocean wind speeds extrapolated from 10 m to 100 m using a logarithmic wind profile and an assumption of neutral stability and wind-induced roughness of the sea [25
]. The information on atmospheric stability and boundary layer height is not used within the long-term trend analysis, even though both parameters are known to influence marine boundary-layer wind profiles [26
]. We choose this because we otherwise would have to use information on these parameters from a model. We do not know if stability and boundary layer heights are constant during the years and if they are modeled correctly. Thus, using this information could contaminate the analysis based on raw data. We investigate the quality of the SSM/I observations compared to in-situ observations on a weekly basis at 100 m level. The in-situ observations are available at around 100 m in the North Sea [28
]; thus, this comparison study is complementary to classical satellite wind comparison based on 10 m buoy observations [24
]. In addition, we add an analysis on stability based on WRF to estimate the bias due to stability effects on the mean wind speed at the two sites. The two novelties in this study are: (1) a consistent method applied in both a temperate area and in the tropics based on SSM/I for wind energy applications; and (2) the estimation of stability correction.
The quarter-century wind speed statistics for the North Sea and South China Sea relevant for wind energy planning are examined. Wind derived from SSM/I vs. Fino1 data have R2
0.72 and std. dev. 1.75 m/s and mean difference −0.12 m/s at 100 m. Better agreement is found for SSM/I vs. WRF at 10 m with R2
0.88, std. dev. 0.98 m·s−1
and mean difference −0.12 m/s. For the 10 m winds, the comparison is very good. This is despite the fact that SSM/I are equivalent neutral winds and WRF stability-dependent winds. Our analysis based on WRF show near-neutral conditions. Other studies [26
] show the atmospheric stratification at Fino1 is near neutral towards slightly stable. It means that WRF-simulated real winds are expected to be slightly higher than SSM/I equivalent neutral winds.
At 100 m height, the average wind speed from SSM/I during 26 years based on 45,146 observations is 9.75 m/s and based on 227,760 WRF (hourly) results is 9.68 m/s. It may be by chance that a negative mean difference is seen, as the difference is less than 1%. A possible cause is the position of the SSM/I grid cell, which is shifted slightly further offshore where winds are expected to be (slightly) higher than at the meteorological mast. Another reason for SSM/I winds to be slightly higher could be land contamination of SSM/I data observed relatively near the coast (around 60 km for Fino1 and 200 km for Hainan). The safe zone is around three times 3 dB beam size from which 98% on-Earth radiance originates. This corresponds to an ellipse of size 207 km by 129 km at 19 GHz for SSM/I. The land has much higher brightness temperature than ocean and therefore even small fractions of land will result in too high winds [42
]. The coastal mask seems to be suitable at Fino1. The above results provide confidence in using SSM/I winds for long-term wind resource statistics for this region.
In the South China Sea, SSM/I- and WRF-derived wind speeds are compared. At the Hainan study site at 10 m R2
is 0.71, std. dev. 1.10 m/s and mean difference 0.83 m/s while the statistics at 100 m are R2
0.67, std. dev. 1.41 m/s and mean difference 0.83 m/s. It may be noted that the mean difference is high for these collocated samples. The average wind speed at 100 m from 26 years of SSM/I observations is 7.58 m/s whereas 25 years of data from WRF is 8.53 m/s, i.e., 11% difference. One year less data from WRF than SSM/I is unlikely to explain the difference. The SSM/I-derived winds are equivalent neutral winds and our extrapolation to 100 m did not include a stability correction. This means that SSM/I winds at 100 m are expected to be positively biased (overestimated) as the climatology of stability at Hainan is dominated by unstable to neutral stratification. From our stability analysis the bias on mean wind speed is estimated as 0.5 m/s. Different reanalysis (ERA Interim vs. CFSR), horizontal grid spacing, and PBL schemes (YSU vs. MYJ) are used in the WRF simulations over the North and South China Sea. While little sensitive to these two factors in the North Sea [29
], the impact is unknown over the South China Sea. It is not possible to conclude whether SSM/I or WRF are more reliable. We note that for Hainan the WRF monthly mean wind speeds and distribution are systematically higher than SSM/I.
Comparison of point observations vs. spatial data from satellites has well-known scaling issue. The surface-layer theory offers weighting functions, traditionally called footprints, for scaling between the upwind area and the measurement point at a certain height [43
]. Assuming homogenous and stationary conditions the observed winds will be mainly related to the upwind near-field area of the footprint. In other words, the time-averaged winds from observations, e.g., as we use hourly values, compare to an upwind area in the spatial domain. For 10 m winds, the area will be smaller than for winds observed at 100 m. The two sites investigated in this study are relatively far from sharp horizontal wind speed gradients, which otherwise would limit the analysis [23
The seasonal wind speed trend at Fino1 in February (−0.92 m/s per decade) is statistically significant at the 95% significance level and in June (0.56 m/s per decade). The 95% percentile wind speeds trend (1.0 m/s per decade) in June is statistically significant at the 95% level. The trends in average wind speeds and 95% percentile winds are similar at Fino1 for all months. Large-scale climate dynamics such as NAO most likely dominate the wind speed trends.
At Hainan the 95% percentile wind speed trend (0.57 m/s per decade) show statistical significance above 95% in June. At the 90% confidence level the average wind speed trend is positive (0.27 m/s per decade) in June and negative in February, March and October (around −0.55 m/s per decade). The average wind speed trend and 95% percentile wind speed trend follow each other from June to January but not from February to May at Hainan. As stated earlier decadal trend analysis based on SSM/I should be possible for trends in wind speed above 0.1 m/s per decade while WRF may show artificial trends due to inconsistency in time due to changes in available input used in the driving reanalysis [1
] and are therefore not used for trend analysis in the current study.
Neither in the North Sea, nor in the South China Sea does the current analysis show significant wind speed trends based on 26 years of observations from SSM/I from 1988 to 2013 at the 95% confidence level. The meteorological wind speed observations across China show significant decreasing trends [13
] among several other studies. Decreasing winds over land are likely a consequence of land cover changes, e.g., urbanization near meteorological masts. However, it is remarkable that our results on decadal wind speed trends in the South China Sea do not show similar pattern, as the interpretation for the changes in wind speed are large-scale dynamics [5
]. Our results based on SSM/I shows weak positive trends at Fino1 (0.10 m/s per decade) and Hainan (0.04 m/s per decade) but these trend values are not significant at the 95% confidence level. Several other studies have found positive trend in wind speeds over the open ocean with statistically significant values [6
Wind resource statistics at 100 m based on SSM/I and WRF at Fino1 show good agreement for mean wind speed (less than 1% difference) while the energy density differs around 14%. At Fino1 estimation of energy density at 100 m from mast data shows 1002 W/m2
from September 2003 to August 2007 (4 years) with the average energy density for each full year 870 W/m2
, 1077 W/m2
, 836 W/m2
and 1225 W/m2
]. The temporal yearly variability is high. The energy density based on SSM/I extrapolated to 100 m is 1131 W/m2
whereas WRF is 975 W/m2
for 26 years. It seems likely that the SSM/I estimate for Fino1 is too high, which in part also may be explained from the SSM/I grid position being shifted slightly further offshore than the actual mast or SSM/I is influenced by land contamination. The observations at Fino1 for a much shorter period (10 years vs. 26 years) show Weibull distribution closer to WRF than to SSM/I. No firm conclusion can be drawn as to which energy density is correct.
In the South China Sea, in-situ observations are not available. Available estimates of wind resources in the South China Sea include results from one year (2011) showing strong gradients in wind energy density near Hainan ~500 W/m2
at 100 m [23
]. Maps on the offshore wind energy density are presented in [48
] based on [49
] and the International Energy Agency (IEA) and Energy Research Institute [50
]. The spatial patterns in these maps are dissimilar. Near Hainan the values are around 400 W/m2
and 450 W/m2
, respectively. The data source for production is not described nor is its years, height or method. Our results on wind energy density from SSM/I and WRF vary considerably. At Hainan, the WRF-derived means show higher values 648 W/m2
while SSM/I show 470 W/m2
, i.e., around 30% difference. The WRF values are likely overestimated. One indication is the diurnal wind speed variation (Figure 9
) where SSM/I and WRF show clear diurnal variation but with WRF winds consistently higher than SSM/I. Lack of wind speed observations from SSM/I during several hours each day may be more problematic near Hainan than in the North Sea as the diurnal winds are more variable in the South China Sea. SSM/I are equivalent neutral winds and extrapolated to 100 m using neutral wind profile. We expect the energy density from SSM/I to be overestimated near Hainan because the atmosphere is often unstable and our analysis on stability shows around 0.5 m/s lower winds at 100 m using stability correction than if assuming neutral conditions.
In summary, SSM/I and WRF results are compared in both study areas. The analysis of SSM/I observations is straightforward and the Charnock relationship and the logarithmic profile method have been used to extrapolate the winds to 100 m. The WRF model is a strong tool for wind resource assessment. Many more parameters than wind speed are available from model results, e.g., stability, temperature, etc. However, the accuracy of the results has to be checked versus reliable observations to ensure optimal choice of initial and boundary conditions and physical parameterizations. The number of possible choices in model set up is overwhelmingly large and even though computational power has increased rapidly, there is continued need for research and comparison analysis, in particular, over ocean. The winds in the North Sea (predominantly westerly’s coming from the North Atlantic Sea) are relatively simpler to simulate than those in the South China Sea, which are influenced by many different large-scale, seasonal (e.g., monsoon) and mesoscale phenomena. SSM/I can be used for consistency check of model results, e.g., when in-situ observations are too few for extensive model evaluation and the choice of planetary boundary layer scheme [29
SMM/I surface ocean wind observations provide new insight to offshore wind statistics relevant for wind energy planning. Novel results presented here are the seasonal wind pattern with springtime winds arriving one month earlier at Fino1 in the North Sea from 1988 to 2013 and higher winds in June. The results are based on SSM/I wind speeds extrapolated from 10 m to 100 m using Charnock and a logarithmic wind profile, thus equivalent neutral winds. The weekly average SSM/I values are successfully compared to WRF model results (better at 10 m than 100 m) and Fino1 mast data of stability-dependent winds. The SSM/I observations do not show significant decadal wind speed changes at quarter-century timescale. The inter-annual variability is 4.6% based on SSM/I and 4.0% based on WRF at Fino1.
At Hainan, in the South China Sea, SSM/I and WRF wind speeds also correlate well (better at 10 m than 100 m) but there is a bias of overestimated wind speeds in WRF. This is despite the fact that the climatology of atmospheric stratification is mainly unstable, and it is therefore expected that SSM/I equivalent neutral winds would be bias high using Charnock and a logarithmic profile approach used to extrapolate the winds from 10 m to 100 m. The stability analysis based on WRF output and adaptation of the Monin–Obukhov similarity theory show unstable conditions with around 0.5 m/s lower mean wind speed. The energy density varies across the South China Sea but the absolute level is not certain from this analysis. There is no significant trend found in wind speeds during the 26 years; however, seasonal changes are observed with negative trend in February, March and October and positive trend in June.