Available via license: CC BY 4.0
Content may be subject to copyright.
Atmosphere 2018, 9, 73; doi:10.3390/atmos9020073 www.mdpi.com/journal/atmosphere
Article
Comparative Evaluation of the Third-Generation
Reanalysis Data for Wind Resource Assessment of
the Southwestern Offshore in South Korea
Hyun-Goo Kim *, Jin-Young Kim and Yong-Heack Kang
New and Renewable Energy Resources & Policy Center, Korea Institute of Energy Research, 34129 Daejeon,
Korea; jinyoung.kim@kier.re.kr (J.-Y.K.); yhkang@kier.re.kr (Y.-H.K.)
* Correspondence: hyungoo@kier.re.kr; Tel.: +82-42-860-3376; Fax: +82-42-860-3462
Received: 31 December 2017; Accepted: 14 February 2018; Published: 16 February 2018
Abstract: This study evaluated the applicability of long-term datasets among third-generation
reanalysis data CFSR, ERA-Interim, MERRA, and MERRA-2 to determine which dataset is more
suitable when performing wind resource assessment for the ‘Southwest 2.5 GW Offshore Wind
Power Project’, which is currently underway strategically in South Korea. The evaluation was
performed by comparing the reanalyses with offshore, onshore, and island meteorological tower
measurements obtained in and around the southwest offshore. In the pre-processing of the
measurement data, the shading sectors due to a meteorological tower were excluded from all
observation data, and the measurement heights at the offshore meteorological towers were
corrected considering the sea level change caused by tidal difference. To reflect the orographic
forcing by terrain features, the reanalysis data were transformed by using WindSim, a flow model
based on computational fluid dynamics and statistical-dynamic downscaling. The comparison of
the reanalyses with the measurement data showed the fitness in the following order in terms of
coefficient of determination: MERRA-2 > CFSR = MERRA > ERA-Interim. Since the measurement
data at the onshore meteorological towers strongly revealed a local wind system such as sea-land
breeze, it is judged to be inappropriate for use as supplementary data for offshore wind resource
assessment.
Keywords: reanalysis data; wind resource assessment; CFSR; ERA-Interim; MERRA; MERRA-2;
South Korea
1. Introduction
The opening of a wind power project is determined based on a feasibility study that estimates
the profit of wind energy production, cost of wind farm construction, and operations and
maintenance (O&M). The most significant risk in wind power projects, which are huge investment
projects, is cost estimation, i.e., wind resource assessment, because a wind farm is to be operated for
the next 20 to 30 years. Therefore, how to reduce uncertainty in relation to a long-term wind resource
using short-term observation data of a year or more is the most important key. In this regard, the
Korean Board of Audit and Inspection warned of the low economic feasibilities of the ‘Daejeong
Offshore Wind Power Project’ on Jeju Island even though it is the wind resource-richest province in
South Korea [1] and the USD 8-billion, three-stage ‘Southwest 2.5 GW Offshore Wind Power Project’
led by the Korean government. A number of studies on wind resource assessment have been
conducted to promote the ‘Southwest Offshore Wind Project’ in the west to the Korean Peninsula and
in the east to China (region marked with dashed line in Yellow Sea in Figure 1). Note, however, that
it is necessary to have in-depth re-investigation since considerable discrepancies in wind resource
Atmosphere 2018, 8, 73 2 of 13
prediction were reported. Kim et al. [2] pointed out that, if the ‘Southwest Offshore Wind Project’
depended on the short-term measurements of offshore meteorological towers only, uncertainties
became larger due to the variability of wind resource, resulting in increases in project risk.
Realistically, it is difficult to perform measurements at the offshore meteorological tower for five
years or longer.
Figure 1. East Asia map around the Korean Peninsula (the dashed circle indicates the Southwest 2.5
GW Wind Power Project site).
Nowadays, reanalysis data tend to be employed more as long-term reference data than
meteorological observation data on nearby islands or onshore. Since orographic forcing by terrain
features does not exist in offshore, reanalysis data shows more higher correlation in offshore than in
the land [3]. Several high-resolution reanalysis data sets are now freely available for use in wind
resource assessment, such as CFSR, ERA-Interim, MERRA, and MERRA-2. Compared to
independent meteorological tower measurements, all four perform significantly better than the
1990s-era first generation and 2000s-era second generation reanalysis at all-time scales. Eichelberger
et al. [4] evaluated the third-generation reanalysis data using high-rise meteorological tower
measurements at 35 sites; the comparison result showed that R2 of CFSR and ERA-Interim was about
17% higher than that of MERRA. (MERRA R2 = 0.46, CFSR & ERA-Interim R2 = 0.54) Brower et al. [5]
reported the same result in their study, which was performed using measurements of high-rise
meteorological towers at 37 sites around the world. R2 of CFSR and ERA-Interim was about 9% higher
than that of MERRA. (MERRA R2 = 0.66, CFSR R2 = 0.74, ERA-Interim R2 = 0.73) Carvalho et al. [6]
compared buoy measurements at five sites in the sea of the Iberian Peninsula with reanalysis data
and reported MERRA R2 = 0.87, CFSR R2 = 0.87, and ERA Interim R2 = 0.78. Chawla et al. [7] also
reported buoy measurements at 12 sites in the sea of the Gulf of Mexico in the Atlantic and at 10 sites
in the sea of Hawaii in the Pacific, showing CFSR R2 of 0.84 and 0.87.
The Korean Peninsula located in the Far East Asia belongs to the monsoon climate zone
characterized by complex seasonal and diurnal meteorological pattern due to the effect of the oceanic
climate in summer and the continental climate in the winter. This study sought to determine data
suitable for the southwest offshore wind resource assessment in the Korean Peninsula among third-
generation reanalysis data CFSR, ERA-Interim, MERRA, and MERRA-2. To do this, a comparative
evaluation was performed on the measurement data around the target site of the ‘Southwest Offshore
Wind Project’—which are offshore, onshore, and island meteorological tower measurements – and
the third-generation reanalysis data. Furthermore, pre-processing was conducted to take optimal
measurements analysis. The variation of measurement heights caused by tidal change in the Yellow
Sea and the meteorological tower shading effect were taken into consideration. Since meteorological
Atmosphere 2018, 8, 73 3 of 13
towers installed on islands and coasts are affected by flow transform caused by terrain features,
numerical analysis using CFD was conducted to correct the effect.
2. Research Data
2.1. Reanalysis Data
Reanalysis refers to a systematic approach for reproducing systematic and consistent
meteorological data by accommodating 7–9 million observation data acquired through various
sources (e.g., radiosonde, buoy, airplanes, ships, etc.) every 6–12 h in the invariant data assimilation
structure and meteorological model in an integrated manner. The first-generation reanalysis data
refers to the global reanalysis data produced at the National Center for Atmospheric Research/
National Centers for Environmental Prediction (NCAR/NCEP) and Europe Center for Medium-
Range Weather Forecast (ECMWF) in the 1990s [8]. The second-generation reanalysis data were then
produced in the early 2000s by resolving several problems and adding the parametrization of
physical processes in the same grid network [9]. In 2010s, third-generation reanalysis data such as
CFSR, ERA-Interim, MERRA, and MERRA-2—which are significantly improved spatio-temporal
resolution—were developed.
Table 1 shows the comparison of the specifications in the third-generation reanalysis data. The
temporal resolution of MERRA, MERRA-2, and CFSR is one-hour interval, whereas ERA-Interim is
three-hour interval. The spatial resolution is 0.5° × 0.5° for CFSR, 0.67° × 0.5° for MERRA, and 0.75° ×
0.75° for ERA-Interim. This study employed the reanalysis data in the grid point closest to the location
of offshore meteorological tower HeMOSU-1, a representative location of the southwest offshore
where wind direction and wind speed data were extracted at 100 m above sea level to be compared
with the observation data.
2.1.1. CFSR (Climate Forecast System Reanalysis)
CFSR (http://rda.ucar.edu/pub/cfsr.html) was produced by the atmosphere-ocean-surface-
glacier coupled climate forecast system with global high resolution in the NCEP based on the same
meteorological data and analysis model as that of MERRA [10].
2.1.2. ERA-Interim (European Center for Medium-Range Weather Forecasts Interim)
ERA-Interim (https://www.ecmwf.int/en/forecasts/datasets/reanalysis-datasets/era-interim) was
produced as a prior step to producing next-generation reanalysis data following ERA-15 and ERA-40
among the reanalysis data series of the ECMWF. ERA-Interim applies a 12-h, 4-dimensional variation
analysis (4D-Var) with adaptive estimation of biases in satellite radiance data (VarBC) based on the
ECMWF Integrated Forecast Model [11].
2.1.3. MERRA (Modern Era Reanalysis for Research and Applications)
MERRA (http://gmao.gsfc.nasa.gov/merra/) refers to the reanalysis data of the National
Aeronautics and Space Administration (NASA) for the satellite era. The official data production was
launched in 2008 using the up-to-date GEOS-5 (Goddard Earth Observing System Data Assimilation
System Version 5) produced in NASA GMAO (GSFC Global Modeling and Assimilation Office) [12].
2.1.4. MERRA-2 (Modern Era Reanalysis for Research and Applications Version 2)
MERRA-2 (https://gmao.gsfc.nasa.gov/reanalysis/MERRA-2/) is MERRA version 2 that
improved the data assimilation of modern hyperspectral radiance and microwave observations,
along with GPS-Radio Occultation datasets. In particular, it was designed such that space-based
observations of aerosols including NASA ozone observations after 2005 were assimilated to represent
their interactions with other physical processes in the climate system [13].
Atmosphere 2018, 8, 73 4 of 13
2.2. Meteorological Tower Measurements
The measurement data secured in the target area were offshore meteorological towers
(HeMOSU-1 and HeMOSU-2), onshore meteorological tower (Gochang), island meteorological
towers (Hoenggyeong-do, Wangdeung-do, Wangdeungyeo), and wind resource campaign database
(Dongho, Sinsi-do, Julpo, and Saemangeum). Figure 2 shows the location of the in-situ measurement
sites around the southwest offshore target area. The yellow lozenge in the map center is a proposed
construction location of the pilot wind farm site as the first phase of the ‘Southwest Offshore Wind
Project’. Table 2 summarizes the meteorological tower measurements used in this study.
Figure 2. Location of in-situ measurement sites around the pilot wind farm of the ‘Southwest Offshore
Wind Project’ (yellow lozenge in the map center). Filled squares, triangles, circles and star symbols
indicates offshore, island, onshore meteorological towers and a tidal station.
2.2.1. Measurement Data of Offshore Meteorological Towers
The Korean Ministry of Trade, Industry & Energy announced the ‘Southwest 2.5 GW Offshore
Wind Power Roadmap’ in October 2010. Next, the Korea Electric Power Research Institute installed
Korea’s first offshore meteorological tower, HeMOSU (Herald of Meteorological and Oceanographic
Special Unit) No. 1 unit (since October 2010) and No. 2 unit (since September 2013).
2.2.2. Measurement Data of Onshore Meteorological Tower
As the first phase of the ‘Southwest Offshore 2.5 GW Wind Project,’ the 80 MW-capacity pilot
wind farm is under construction until 2018 and will be grid-connected to the Gochang Power Test
Center of Korean Electric Power Corporation (KEPCO) located near the coast. The wind speed data
for 34 months obtained from a 120 m-high transmission tower (altitude above sea level: 21 m) inside
the Gochang Power Test Center were employed.
Atmosphere 2018, 8, 73 5 of 13
Table 1. Comparison of the third-generation reanalysis data.
Name Source Time Range Assimilation Model Resolution Publicly Available
Dataset Resolution
Dataset Output Times and Time
Averaging
CFSR NCEP 1979–present 3D-VAR T382 L64 0.5 × 0.5 and 2.5 × 2.5 Hourly, 4 times daily
ERA Interim ECMWF 1979–present 4D-VAR TL255L60 and N128 reduced
Gaussian
User-defined, down to
0.75 × 0.75
3 h for most surface fields; 6 h for
upper-air fields
MERRA NASA 1979–present
3D-VAR, with
incremental update
2/3 lon. × 1/2 lat. deg.; 72
sigma levels
2/3 lon. × 1/2 lat. deg. 3d
analysis and 2D variables;
1.25 deg. 3D diagnostics;
72 model levels and 42
pressure levels
2D diagnostics—1-h avg., centered
on half hour; 3D diagnostics—3-h
avg., centered on 01:30, 04:30 ...
22:30; 3D analysis -instantaneous 6-
h; 2D diagnostics
MERRA-2 NASA
GMAO 1980–present
3D-VAR with
incremental update;
Includes aerosol
data assimilation
Native cube sphere grid
output is interpolated to 5/8
lon. × 1/2 lat. deg.; 72 sigma
levels
5/8 lon. × 1/2 lat. deg. 3D
analysis and 2D variables;
3D diagnostics; 72 model
levels and 42 pressure
levels
2D Diagnostics—1-h avg., centered
on half-hour; 3D diagnostics—3-h
avg., centered on 0130, 0430 ... 2230;
3D analysis - Instantaneous 6-h; 2D
diagnostics
Source: https://reanalyses.org/atmosphere/comparison-table (accessed on 16 February 2018).
Table 2. Summary of the meteorological tower measurements.
Site Location Coordinates Meausrement Period Data Average Measurement Heights (m)
HeMOSU-1 Offshore 35°27′55.17′′ N, 126°07’43.30′′ E October 2010~March 2016 10 min 26, 46, 56, 66, 76, 86, 96, 99
HeMOSU-2 Offshore 35°49′25.50′′ N, 126°12′23.06′′ E January 2014~July 2015 10 min 40, 60, 80, 100, 107, 117
Hoenggyeong-do Island 35°51′26′′ N, 126°25′03′′ E December 2008~November 2009 Monthly 20, 30
Wangdeung-do Island 35°39′43′′ N, 126°06′24′′ E May 2011~April 2012 Monthly 12
Wangdeunyeo Island 35°22′15′′ N, 126°07′50′′ E January 2010~December 2011 10 min 20, 40, 60, 70, 80
Gochang Onshore 35°27′43′′ N, 126°26′57′′ E November 2008~May 2009 10 min 80, 100, 120
Dongho Onshore 35°30′37′′ N, 126°28′48′′ E April 1997~April 1998 Hourly 15, 30
Sinsi-do Island 35°49′10′′ N, 126°28′40′′ E December 1999~April 2000 Hourly 15, 30
Julpo Onshore 35°34′58′′ N, 126°39′50′′ E February 1998~August 2000 Hourly 15, 30
Saemangeum Onshore 35°43′39′′ N, 126°31′41′′ E August 1999~October 2001 Hourly 15, 30
Atmosphere 2017, 8, 73 6 of 13
2.2.3. Measurement Data of Island Meteorological Towers
Meteorological towers were installed on nearby islands to alleviate the uncertainty in the wind
resource assessment at the pilot wind farm. The measurement data for 12 months—which were
measured by attaching sensors to a broadcasting tower (altitude above sea level: 184 m) in
Wangdeung-do, which was 22 km away from the north of HeMOSU-1—were employed. Jeong et al.
[14] installed an 80 m-high meteorological tower in Wangdeungyeo (altitude above sea level: 30 m),
8 km away from the south of HeMOSU-1; the monthly wind speed data for 24 months were adopted
from their paper. In addition, Shim et al. [15] installed a 30 m-high meteorological tower in
Hoenggyeon-do (altitude above sea level: 40 m) located north of Saemangeum Gogunsan-do; the
monthly wind speed data for 12 months were adopted from their paper.
2.2.4. Wind Resource Database
The Korea Institute of Energy Research (KIER) disclosed the results of the wind resource
assessment performed throughout South Korea through the web service of the renewable energy
resource map (www.kier-atlas.org). Wind data measured at 30 m-high meteorological towers in
Dongho, Sinsi-do, Julpo, and Saemangeum were used for comparison.
3. Analysis Methods
3.1. Comparative Evaluation between Reanalysis Data
Four kinds of reanalysis data closest at the installation location of HeMOSU-1, which was
regarded as a representative location of southwest offshore in Korea, were extracted to examine the
differences between the reanalysis data systematically and were reconfigured into 1 h-averaged time
series data.
The reference height was set to 100 m above sea level, which is the highest measurement height
of HeMOSU-1. The mean average error (MAE) was calculated to evaluate the difference between the
reanalysis data quantitatively as follows:
MAE=1
,−,
(1)
In Equation (1), V refers to the wind speed in the reanalysis data; the data period was selected
as 24 years (from 1992 to 2015, n = 24 × 8760), which included all measurement periods of observation
data used in this study. Number 1 and 2 indicate two different reanalyses. The MAE of wind speed
difference was calculated and an inter-annual pattern was compared.
3.2. Measurement Data Preprocessing
The measurement data around the pilot wind farm were pre-processed to remove uncertainty
factors, and orographic forcing caused by terrain features at each of the meteorological tower
locations was reflected on the reanalysis data using a flow model, WindSim (available online:
www.winsim.com, accessed on 16 February 2018).
3.2.1. Exclusion of Tower Shading Sectors
Sensors were mounted at the end of a long horizontal boom attached to the meteorological tower
to minimize flow interference caused by the tower structure following the IEC 61400-12-1 Annex G.
[16] Note, however, that wind reaches an anemometer through the meteorological tower in a specific
wind direction. Here, a measurement error due to the shading of the meteorological tower could
occur. Thus, the wind direction sectors where meteorological tower shading occurred were identified
and excluded.
Atmosphere 2017, 8, 73 7 of 13
3.2.2. Correction of Sea-Level Variation
Because the West Sea of Korea (Yellow Sea in Figure 1) has tidal difference of more than 7 m, the
measurement height of the sensors installed at the offshore meteorological tower should be corrected
to the reference of sea level. The measurement height was corrected using hourly tidal data (Figure
3) in the Wi-do observation station (star mark in Figure 2) of the Korea Hydrographic &
Oceanographic Administration, which is closest to the HeMOUS-1 offshore meteorological tower. In
other words, the wind speed measured at the anemometer installed between the top of the offshore
meteorological tower (99 m based on the mean sea level) and below (86 m) was corrected considering
the tidal difference through linear interpolation.
Figure 3. Histogram of hourly tide range at the Wi-do observation station (from October 2010 to
December 2016).
3.2.3. CFD to reflect Orographic Forcing by Terrain Features
Since the spatial resolution of the reanalysis data (Table 1) is in order of dozens of km, orographic
forcing of local scale by terrain features was not reflected. Moreover, since meteorological towers
installed on islands and coasts are strongly affected by orographic forcing by terrain features,
numerical analysis using the CFD software WindSim was conducted to reflect the effect. In other
words, reanalysis data were inputted to ‘Climatology’ in the WindSim analysis and then ‘Transferred
Climatology’, on which orographic forcing by terrain features was reflected at each of the
measurement locations, was simulated.
3.3. Evaluation between Reanalysis Data and Measurement Data
Regression analysis on the four kinds of third-generation reanalysis data, which were
transferred to meteorological tower locations by reflecting orographic forcing by terrain features,
with the pre-processed measurement data was conducted. Here, superiority and inferiority among
the reanalysis data were evaluated in terms of mean bias (BIAS), root mean square error (RMSE), and
coefficient of determination (R
2
).
BIAS=1
−
(2)
-300 0300 600 900
0
1
2
3
4
5
6
Frequency ( %)
Tide (cm)
Atmosphere 2017, 8, 73 8 of 13
RMSE=
1
−
(3)
In Equations (2) and (3), M and O refer to the predicted and observed wind speed respectively.
The measurement data of meteorological towers were expected to have different characteristics
depending on the installation location of the tower, such as offshore, onshore, and island. To examine
the meteorological characteristics, the energy pattern factor [17], one-hour autocorrelation coefficient,
and Weibull shape factor were compared. The energy pattern factor (EPF) is defined as presented in
Equation (4).
=1
(4)
Assuming that air density (ρ) is a constant, EPF will be a ratio of wind power density (/)
calculated with mean wind speed () to actual mean wind power density (). In other words,
=12
(5)
Thus, the larger EPF is, the larger the dispersion of wind speed distribution from the mean wind
speed, which is then followed by a decreasing Weibull shape factor (k). This correlation is also
expected to occur in the 1 h-autocorrelation coefficient. Since larger dispersion of wind speed means
large wind speed variance, the 1 h-autocorrelation coefficient will tend to decrease.
4. Results and Discussion
4.1. Comparative Analysis on Reanalysis Data
Tables 3 and 4 present the statistical and correlation analysis results of hourly reanalysis data at
HeMOSU-1 location, respectively. For the mean/maximum wind speed and Weibull factors, the
difference between reanalysis data is within the range of 8%. The correlation among reanalysis data
showed that air temperature has R2 of more than 0.98, wind speed has R2 of more than 0.76, and wind
direction has R2 of more than 0.91. From the meteor-statistics viewpoint, the four kinds of reanalysis
data are very similar time-series data, but a difference in mean wind power density should be noted.
Table 3. Wind statistics of hourly reanalysis data at HeMOSU-1 location (100 m above sea level).
Wind Statistics CFSR ERA-Interim MERRA MERRA-2
Mean wind speed (m/s) 6.9 6.9 6.6 6.4
Max wind speed 30.8 29.9 29.0 29.0
Weibull shape parameter (k) 1.96 1.99 1.98 1.95
Weibull scale parameter (c, m/s) 7.73 7.73 7.48 7.22
Mean wind power density (W/m2) 385 377 343 322
Table 4. Coefficient of determination (R2) among hourly reanalysis data at HeMOSU-1 location (left:
wind speed, right: wind direction).
Reanalysis Data ERA-Interim MERRA MERRA-2
CFSR 0.76 0.91 0.82 0.92 0.83 0.93
ERA-Interim 1 0.85 0.94 0.83 0.93
MERRA 1 0.89 0.95
In other words, the maximum value (CFSR) and the minimum value (MERRA-2) of wind power
density are almost 20% different, which cannot be ignored at all in wind resource assessment. This
result implies that the economic feasibility results for the wind power project can be altered,
depending on which reanalysis data are selected.
Atmosphere 2017, 8, 73 9 of 13
Noticeable differences between third-generation reanalysis data are seen mostly since satellite
data were assimilated into the reanalyses after 2000s (see Figure 4). However, MERRA and MERRA-
2 that use the same production scheme shows the smallest difference.
Figure 4. Inter-annual variation of MAEs of wind speed differences between reanalyses.
4.2. Comparative Analysis between Reanalysis Data and Measurement Data
4.2.1. Offshore Meteorological Tower Data
HeMOSU-1 and HeMOSU-2 are approximately 40 km away from each other; the R
2
between the
two measurement data is 0.83 for wind speed and 0.94 for wind direction, respectively. The correction
of tidal difference in all cases showed about 3% improvement of R
2
. Wind power density was adjusted
to 451 W/m
2
at 437 W/m
2
, or an increase of approximately 3.2% (see Table 5). The highest reliable
fitness with the highest representation of the southwest offshore area between offshore
meteorological tower data and four kinds of reanalysis data was as follows: MERRA-2 = CFSR >
MERRA > ERA-Interim.
Table 5. Comparison between the HeMOSU-1 and HeMOSU-2 measurements and reanalysis data.
Site Measure CFSR ERA-Interim MERRA MERRA-2
HeMOSU-1
100 m ASL
Tower shading correction
BIAS (m/s) −0.07 0.18 −0.09 −0.06
RMSE (m/s) 2.10 2.22 2.20 2.08
R
2
0.76 0.70 0.73 0.76
Tower shading correction + tide correction
BIAS (m/s) −0.07 0.10 −0.07 −0.05
RMSE (m/s) 2.10 2.20 2.20 2.06
R
2
0.77 0.72 0.74 0.78
HeMOSU-2
100 m ASL
Tower shading correction + tide correction
BIAS (m/s) −0.13 −0.19 −0.19 −0.05
RMSE (m/s) 2.03 2.17 2.06 2.01
R
2
0.72 0.69 0.71 0.73
Atmosphere 2017, 8, 73 10 of 13
4.2.2. Onshore Meteorological Tower Data
As presented in Table 6, which summarizes the comparative evaluations with the observation
data of the onshore meteorological towers, MERRA-2 had the highest fitness in all cases. HeMOSU-
1 and Gochang are approximately 30 km away from each other; the R2 between the two measurement
data is 0.68 for wind speed that is fairly lower than that between HeMOSU-1 and 2.
Table 6. Comparison between onshore measurements and reanalysis data.
Site Measure CFSR ERA-Interim MERRA MERRA-2
Gochang
100 m AGL
BIAS (m/s) 1.53 1.71 1.42 1.30
RMSE (m/s) 2.77 2.94 2.69 2.44
R2 0.60 0.57 0.58 0.61
Dongho
30 m AGL
BIAS (m/s) 0.86 0.85 0.60 0.60
RMSE (m/s) 2.32 2.42 2.26 2.24
R2 0.52 0.48 0.50 0.52
Sinsi-do
30 m AGL
BIAS (m/s) 0.40 0.44 0.01 0.01
RMSE (m/s) 2.50 2.43 2.50 2.48
R2 0.65 0.58 0.66 0.67
Julpo
30 m AGL
BIAS (m/s) 1.47 1.50 1.32 1.30
RMSE (m/s) 2.88 2.88 2.78 2.71
R2 0.25 0.25 0.25 0.26
Saemangeum
30 m AGL
BIAS (m/s) 0.50 0.50 0.30 0.30
RMSE (m/s) 2.19 2.22 2.16 2.10
R2 0.49 0.49 0.52 0.53
The rank of fitness in the reanalysis data was similar to that of the offshore meteorological
towers: MERRA-2 > MERRA = CFSR > ERA-Interim.
4.2.3. Island Meteorological Tower Data
As presented in Table 7, which compares the correlation factor between the observation data of
island meteorological tower and reanalysis data, the correlation factor with the reanalysis data was
the same order as shown above: MERRA-2 > MERRA = CFSR > ERA-Interim. For Wangdeung-do
where sensors were mounted in the broadcasting tower, R2 noticeably improved 16% to the 27% level
with tower shading correction.
Table 7. Comparison between island measurements and reanalysis data (R2).
Site Correction CFSR ERA-Interim MERRA MERRA-2
Wangdeung-do
40 m AGL
No correction 0.64 0.55 0.59 0.60
Tower shading correction 0.74 0.68 0.73 0.76
Wangdeungyeo
80 m AGL – 0.56 0.65 0.58 0.74
Heonggueng-do
30 m AGL – 0.76 0.70 0.79 0.82
4.3. Offshore/Onshore Meteorology
The correlation between the onshore meteorological tower measurements and reanalysis data
was relatively low. In particular, Julpo was the lowest. To investigate the reason for this low
correlation, various meteor-statistical parameters were calculated as presented in Table 8. It was clearly
seen that EPF, 1 h autocorrelation coefficient, and Weibull factors in Julpo had highly different range
of values compared to that of other meteorological towers. One interesting fact is that a strong negative
correlation, −0.84 and −0.95, was revealed between EPF and 1 h autocorrelation coefficient and
between EPF and Weibull shape factor, respectively. This is similar to the expectation in Section 3.3.
Atmosphere 2017, 8, 73 11 of 13
Table 8. Comparison of wind statistics of the meteorological tower measurements.
Site Energy Pattern
Factor
1h
Autocorrelation
Weibull Scale
Factor, c (m/s)
Weibull Shape
Factor, k
HeMOSU-1 2.18 0.98 7.85 1.81
HeMOSU-2 2.36 0.95 7.28 1.69
Gochang 2.39 0.93 5.97 1.65
Dongho 2.41 0.92 5.13 1.72
Sinsi-do 2.43 0.94 6.15 1.55
Julpo 3.22 0.89 3.76 1.31
Saemangeum 2.35 0.93 5.48 1.70
Figure 5 depicts the autocorrelation of the wind speed measurement data. The measurements at
HeMOSU-1 influenced by oceanic weather with low variability, Gochang on the coast showed a
reduction tendency in autocorrelation coefficient monotonically according to a lag time. In contrast,
Julpo on the coast inside a bay located inwardly in the inland showed clear periodicity of 24 h due to
the effect of the sea-land breeze system. Dongho located in the entry of the bay also has 24-h
periodicity, although it is weaker than that in Julpo. Julpo is located 20 km inward to the inland from
the bay entrance, and it has topographic characteristics shielded by 509 m-high and 444 m-high
mountains in the north and south, respectively. Thus, ventilation would occur only through the open
east-west direction.
Figure 5. Auto-correlation plots of wind speed data.
From the analysis with Table 8 and Figure 5, it is conjectured that the intensity of local wind
system increases as penetrates into inland from offshore through onshore. This implies that offshore
wind resource assessment using onshore or inland measurement data would cause considerable
misinterpretation. According to a wind sector clustering map shown in Figure 6, classified by a
surface wind regionalization method proposed by Kim et al. [18], offshore meteorological towers
HeMOSU-1 and HeMOSU-2 belong to a different wind sector, whereas Dongho and Gochang belong
to the same wind sector along the coast; Julpo, which is located inland, is classified into a separate
wind sector.
Atmosphere 2017, 8, 73 12 of 13
Figure 6. Wind sector clustering of the southwest offshore region.
5. Conclusions
This study performed comparative analysis with the meteorological tower measurement data to
evaluate the applicability of four kinds of third-generation reanalysis data when wind resource
assessment on the ‘Southwest Offshore Wind Project,’ which is underway strategically in South
Korea at present, is conducted. The following conclusions were derived through this study:
(1) The difference in wind power density between four kinds of reanalysis data was more than 20%
in the southwest offshore, and this level of difference cannot be ignored in the wind resource
assessment. Accordingly, it is very important to select proper reanalysis data.
(2) According to the comparison of offshore, onshore, and island meteorological tower
measurements with the reanalysis data, MERRA-2 showed the best fitness among the four kinds
of data. In terms of the mean of R2 between the reanalysis data and all observation data, the
fitness is in the following order: MERRA-2 (0.67) > CFSR (0.63) = MERRA (0.63) > ERA-Interim
(0.61). For reference, the variance of R2 was 0.10 for all reanalysis data.
(3) Pre-processing is recommended when meteorological tower measurement data are used. For
example, R2 was improved by over 16% when tower shading correction was applied for
Wangdeung-do case; the correction of the tidal difference of HeMOSU-1 also improved R2 by
over 2%.
(4) According to the wind characteristics analysis such as energy pattern factor, 1 h autocorrelation,
and surface wind regionalization, 24-h periodicity due to sea-land breeze was revealed to be
stronger inland than on the coast. Therefore, onshore or inland measurements are inappropriate
for use as supplementary data for offshore wind resource assessment in the Southwestern
Offshore in South Korea.
Since many kinds of reanalysis data are available, the data selection criteria are not sufficiently
clear. Thus, this study limited the locations to the southwest offshore in the Korean Peninsula and
selected reanalysis data suitable for the wind resource assessment through comparative analysis with
various types of observation data. For future studies, investigating an ensemble method that can
reduce uncertainty by employing multiple reanalysis data is recommended.
Atmosphere 2017, 8, 73 13 of 13
Acknowledgments: This work was conducted under the framework of the research and development program
of the Korea Institute of Energy Research (B8-2424-02).
Author Contributions: Hyun-Goo Kim conceived and designed the investigations; Jin-Young Kim and Yong-
Heack Kang analyzed the data and performed the analysis; Hyun-Goo Kim wrote the paper.
Conflicts of Interest: The authors declare no conflict of interest.
References
1. Kim, H.G.; Kang, Y.H.; Hwang, H.J.; Yun, C.Y. Evaluation of inland wind resource potential of South Korea
according to Environmental Conservation Value Assessment. Energy Procedia 2014, 57, 773–781.
2. Kim, H.G.; Jang, M.S.; Ko, S.H. Long-term wind resource mapping of Korean west-south offshore for the
2.5 GW Offshore Wind Power Project. J. Environ. Sci. Int. 2013, 20, 1305–1316.
3. Sharp, E.; Dodds, P.; Barrett, M.; Spataru, C. Evaluating the accuracy of CFSR reanalysis hourly wind speed
forecasts for the UK, using in situ measurements and geographical information. Renew. Energy 2015, 77,
527–538.
4. Eichelberger, S.; Stoelinga, M.; McCaa, J. Performance of new reanalysis data sets for estimating the
temporal and spatial variability of wind resource. In Proceedings of the European Wind Energy Conference
2013, Vienna, Austria, 4–7 February 2013.
5. Brower, M.C.; Barton, M.S.; Lledó, L.; Dubois, J. A Study of Wind Speed Variability using Global Reanalysis
Data; Technical Report; AWS Truepower, Albany, NY, USA, 2013; p. 11.
6. Carvalho, D.; Rocha, A.; Gomez-Gesteira, M.; Santos, C.S. Comparison of reanalyzed, analyzed, satellite-
retrieved and NWP modelled winds with buoy data along the Iberian Peninsula coast. Remote Sens. Environ.
2014, 152, 480–492.
7. Chawla, A.; Spindler, D.M.; Tolman, H.L. Validation of a thirty-year wave hindcast using the Climate
Forecast System Reanalysis winds. Ocean Model. 2013, 70, 189–206.
8. Kalnay, E.; Kanamitsu, M.; Kistler, R.; Collins, W.; Deaven, D.; Gandin, L.; Iredell, M.; Saha, S.; White, G.;
Woollen, J.; et al. The NCEP/NCAR 40-Year Reanalysis Project. Bull. Am. Meteorol. Soc. 1996, 77, 437–471.
9. Kanamitsu, M.; Ebisuzaki, W.; Woollen, J.; Yang, S.K.; Hnilo, J.J.; Fiorino, M.; Potter, G.L. NCEP–DOE
AMIP-II Reanalysis(R-2). Bull. Am. Meteorol. Soc. 2002, 83, 1631–1643.
10. Saha, S.; Moorthi, S.; Pan, H.L.; Wu, X.; Wang, J.; Nadiga, S.; Tripp, P.; Kistler, R.; Woollen, J.; Behringer, D.;
et al. The NCEP Climate Forecast System Reanalysis, Bull. Am. Meteorol. Soc. 2010, 91, 1015–1057.
11. Dee, D.P.; Uppala, S.M.; Simmons, A.J.; Berrisford, P.; Poli, P.; Kobayashi, S.; Andrae, U.; Balmaseda, M.A.;
Balsamo, G.; Bauer, P.; et al. The ERA-Interim reanalysis: Configuration and performance of the data
assimilation system. Q. J. R. Meteorol. Soc. 2011, 137, 553–597.
12. Rienecker, M.M.; Suarez, M.J.; Gelaro, R.; Todling, R.; Bacmeister, J.; Liu, E.; Bosilovich, M.G.; Schubert,
S.D.; Takacs, L.; Kim, G.K.; et al. MERRA: NASA’s Modern-Era Retrospective Analysis for Research &
Applications. J. Clim. 2011, 24, 3624–3648.
13. Molod, A.; Takacs, L.; Suarez, M.; Bacmeister, J. Development of the GEOS-5 atmospheric general
circulation model: Evolution from MERRA to MERRA2. Geosci. Model Dev. 2015, 8, 1339–1356.
14. Jeong, M.S.; Moon, C.J.; Jeong, G.S.; Choi, M.S.; Jang, Y.H. The research on the Yeonggwang Offshore Wind
Farm generated energy prediction. J. Korean Sol. Energy Soc. 2012, 32, 33–41.
15. Shim,A.R.; Choi, Y.S.; Lee, J.H. Measurement and analysis of wind energy potential in Kokunsando of
Saemankeum. J. Korean Soc. New Renew. Energy 2011, 7, 51–58.
16. IEC (International Electrotechnical Commission). IEC Standard 61400-12-1, Wind Energy Generation
Systems—Part 12-1: Power Performance Measurements of Electricity Producing Wind Turbines; IEC Central
Office: Geneva, Switzerland, 2017.
17. Akdag, S.A.; Dinler, A. A new method to estimate Weibull parameters for wind energy applications. Energy
Convers. Manag. 2009, 50, 1761–1766.
18. Kim, J.S.; Kim, H.G.; Park, H.D. Surface wind regionalization based on similarity of time-series wind
vectors. Asian J. Atmos. Environ. 2016, 10, 80–89.
© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
