research article wrf model methodology for offshore wind...

Research ArticleWRF Model Methodology for Offshore WindEnergy Applications

Evangelia-Maria Giannakopoulou and Regis Nhili

EDF Energy RampD UK Centre 52 Grosvenor Gardens London SW1W 0AU UK

Correspondence should be addressed to Evangelia-Maria Giannakopoulou mariliagiannakopoulouedfenergycom

Received 6 February 2014 Accepted 24 March 2014 Published 23 April 2014

Academic Editor Huei-Ping Huang

Copyright copy 2014 E-M Giannakopoulou and R Nhili This is an open access article distributed under the Creative CommonsAttribution License which permits unrestricted use distribution and reproduction in any medium provided the original work isproperly cited

Among the parameters that must be considered for an offshore wind farm development the stability conditions of the marineatmospheric boundary layer (MABL) are of significant importance Atmospheric stability is a vital parameter in wind resourceassessment (WRA) due to its direct relation to wind and turbulence profiles A better understanding of the stability conditionsoccurring offshore and of the interaction between MABL and wind turbines is needed Accurate simulations of the offshorewind and stability conditions using mesoscale modelling techniques can lead to a more precise WRA However the use of anymesoscale model for wind energy applications requires a proper validation process to understand the accuracy and limitations ofthe model For this validation process the weather research and forecasting (WRF) model has been applied over the North Seaduring March 2005 The sensitivity of the WRF model performance to the use of different horizontal resolutions input datasetsPBL parameterisations and nesting options was examined Comparison of the model results with other modelling studies and withhigh quality observations recorded at the offshore measurement platform FINO1 showed that the ERA-Interim reanalysis data incombination with the 25-level MYNN PBL scheme satisfactorily simulate the MABL over the North Sea

1 Introduction

The offshore wind energy has recently become a rapidlygrowing renewable energy resource worldwide with severaloffshore wind projects in development in different planningstages [1 2] Despite this a better understanding of the inter-action between the MABL and the offshore wind turbinesis needed in order to contribute to a better energy captureand cost-effectiveness [2] The MABL is less studied than theplanetary boundary layer (PBL) over land and some obviousreasons are the difficulties scarcity and costs of obtainingoffshore observations However during the last years a greateffort has been made in finding solutions [3] For examplemesoscalemeteorologicalmodels are increasingly consideredlately for wind energy assessment and yield calculationsThe advantage of the mesoscale models in WRA relies ontheir ability to simulate with reasonable accuracy the lowerparts of the boundary layer including important atmosphericproperties such as the atmospheric energy balance andstability [4] However the usage of any mesoscale model

for wind energy applications requires a proper validationprocess to understand the accuracy and limitations of themodel

The impact of the atmospheric stability on the wind andturbulence profiles and on the wind turbine wakes raises theneed to consider in the WRA of an offshore wind farm thestability conditions observed in the MABL [5 6] Differentstability conditions result in different shear conditions whichin turn lead to different wind speed distributions acrossthe rotor swept area The different velocity distributions cansignificantly affect the power production and fatigue loadson the wind turbine [7] For example it is well known thatwind power losses as a result of wake losses are largest instable atmospheric conditions and least in unstable andorneutral conditions [6] Therefore a precise estimation of thewind and stability conditions of an offshore site is requiredCurrently stability is not measured in traditional WRAcampaigns and more accurate WRA could be obtained if thewind climate is classified not only by wind direction but alsoby stability classes

Hindawi Publishing CorporationAdvances in MeteorologyVolume 2014 Article ID 319819 14 pageshttpdxdoiorg1011552014319819

2 Advances in Meteorology

In the mesoscale models the vertical stratification isinherently included and the different PBL schemes influ-ence the accuracy of the simulated winds and atmosphericstratification in the MABL [8] For instance some PBLparameterisations are best in unstableneutral stratificationssome others are best in stable atmospheric conditions andsome in both [3] It is alsowidely known that the performanceof the PBL schemes in the mesoscale models depends on thearea of interest the meteorological variables examined andseason and time of day considered and one cannot identifya best model configuration in a general sense [9] The mostcommon approach to deal with this problem is to carry outstatistical studies over a sensible set of model configurationsand finally use the PBL scheme that reproduces better theobservations in an average sense

The attention of this study has been focused on thedevelopment of a mesoscale methodology by using theweather research and forecasting (WRF) model [10] so asto accurately simulate the wind conditions and to reproducethe stability effects on the wind profile over the MABL inthe North Sea The present work describes the verificationof the WRF model by comparing the model results withthat of other modelling studies (eg [8 11]) and with highquality observations recorded at the FINO1 offshore plat-form in the North Sea For this verification process severalWRF modelling simulations have been performed duringMarch 2005 In particular different horizontal resolutionsPBL parameterizations initial and boundary conditions andnesting options were tested The high probability (20) ofoccurrence of stable to very-stable atmospheric stratificationsituations in the spring and early summer at the FINO1platform and the impact of the high stability on the windprofile increases the need to focus on the stable atmosphericconditions and include them in our WRF model configura-tion andmesoscale methodology [3 5] A time period duringwhich all the atmospheric stability conditions are observedare also investigated in this study

A description of the verification process including infor-mation about the data sources and the statistics used foranalysis is given in Section 21 Section 22 focuses on theWRF model setup and the process followed for testing itsperformance Description of the synoptic conditions duringthe stable period in March 2005 and presentation of theWRF results for this case study with a number of keyderived statistics are provided in Section 3 In the samesection presentation of theWRF results for the whole March2005 with statistical metrics and error analysis is also givenConclusions follow in Section 4

2 Materials and Methods

21 Verification Process Two main steps of the verificationprocess are the definition of the data sources and the def-inition of the verification statistics to be produced by theanalysis

211 Observational Data FINO1 Measurements The FINO1offshore platform in Southern North Sea is located 45 kmnorth of the Borkum island (latitude 540∘N and longitude

635∘E) and performs multilevel measurements of windspeed wind direction air temperature relative humidity andair pressure since 2004 The height of the measurement mastis about 100m above mean sea level (MSL) Three ultrasonicinstruments (of 10Hz temporal resolution) are located at415m 615m and 815m height on northwesterly orientedbooms In addition eight cup anemometers with the lowerresolution of 1Hz are installed at different heights startingfrom 34m upto about 100m (every 10m) on boomsmountedin southeast direction of themeteorologicalmast (eg [8 11])

A list of parameters used for the analysis of the meteo-rological conditions and for the comparison with the WRFmodel results is shown in Table 1 In Table 1 we also providethe heights of the recorded parameters and the sensor typesincluding their accuracy according to Canadillas [6] Forthe WRF model validation we used data from the cupanemometers in order to have the best spatial coverage todescribe the vertical profiles of wind speed and wind direc-tion According to Deutsches Windenergie-Institut (DEWIGerman Wind Energy Institute) scientists at the RAVE 2012conference [12] the sonic anemometer data were sparse andwere associated with errors especially during the period2004ndash2007 In general from 2004 to 2011 the availabilityof the cup anemometers at the FINO1 platform was about98 while the availability of the sonic anemometers wasapproximately 83 [12]

212 Define Verification Statistics The following statisticalmetrics have been used in this study to verify the performanceof the WRF model when compared with the FINO1 obser-vations More details on the verification statistics could befound in [13]

Bias or mean error (ME) is defined as the mean ofthe differences between the WRF simulated meteorologicalparameters and the FINO1 observations In particular themean error is calculated for each hour of data and the timeaverage (over the period we have considered to study) biasis then provided for each measuring height of the FINO1platform

Mean absolute error (MAE) is defined as the quantity usedtomeasure how close the observed values are to themodelledones The MAE is given by

MAE = 1119899

119899

sum

119894=1

1003816100381610038161003816

119909119894minus 119910119894

1003816100381610038161003816

(1)

where |119909119894minus 119910119894| is the absolute error with 119909

119894and 119910

119894to repre-

sent the modelled and the observed values at the FINO1 metmast respectively 119899 is the sample size

Standard Deviation (STD) of the ME is defined as thedispersion of the biased values around the mean value A lowstandard deviation indicates that the data points tend to bevery close to themean high standard deviation indicates thatthe data points are spread out over a large range of valuesTheSTD uses the following formula

120590 =radic

sum

119899

119894=1

(119909119894minus 119909)

2

(119899 minus 1)

(2)

where 120590 is the standard deviation and 119909 is the sample mean

Advances in Meteorology 3

Table 1 Meteorological parameters with their associated heights and the sensor types including their accuracy [6] at the FINO1 offshoreplatform

Variable Heights (m) LAT Sensor type (accuracy)Wind speed (ms) (cup anemometer) 34 415 515 615 715 815 915 1045 Vector A100LK-WR-PC3 (plusmn001ms)Wind direction (degree) 415 515 615 715 815 915 Thies wind vane classic (plusmn1∘)Air temperature (∘C) 30 415 52 72 101 Pt-100 (plusmn01 K at 0∘C)Relative humidity () 345 90 Hydrometer Thies (plusmn3 RH)Air pressure (hPa) 225 Barometer Vaisala (plusmn003 hPa)

Root Mean Square Error (RMSE) is a frequently usedmeasure of the difference between values predicted by amodel and the values actually observed It measures theaverage magnitude of the error and it is defined as themeasure of the combined systematic error (bias) and randomerror (standard deviation) Therefore the RMSE will only besmall when both the variance and the bias of an estimator aresmall The RMSE uses the following formula

RMSE = radic1205902 + 1199092 (3)

where 119909 is the sample mean and 120590 is the standard deviationPearson correlation coefficient (R) is defined as the mea-

sure of the linear dependence between the WRF resultsand the FINO1 data giving a value between +1 and minus1inclusive It thus indicates the strength and direction of alinear relationship between these two variables A value of1 implies that a linear equation describes the relationshipbetween WRF and the observations perfectly with all datapoints lying on a line for which the WRF values increaseas the data values increase The correlation is minus1 in case ofa decreasing linear relationship and the values in betweenindicates the degree of linear relationship between the WRFmodel and the observations The formula for the Pearsonproduct moment correlation coefficient is

119877 =

sum

119899

119894=1

(119909119894minus 119909) (119910

119894minus 119910)

radicsum

119899

119894=1

(119909119894minus 119909)

2

(119910119894minus 119910)

2

(4)

where the 119909 and 119910 are respectively the sample means of theWRF results and the measurements

22 Model and Setup In this study the numerical weatherprediction (NWP) model of the National Centre for Atmo-spheric Research (NCAR) advancedWRFmodel version 34[10] was used The model was run for this project on a XeonX54 system at EDF RampD in France with 96 CPUs TheWRFmodel is based on the fully compressible nonhydrostaticEuler equations and for the purposes of this research theLambert conformal projection was chosen A third orderRunge-Kutta (RK3) integration scheme and Arakawa C-gridstaggering were used for temporal and spatial discretizationrespectively The modelling setup including the selecteddomains and the initial and boundary conditions as well asthe physics schemes is described belowThe strategy followedfor the WRF modelling setup followed up to a certain pointthe strategy of previous WRF studies (eg [14 15])

WPS domain configuration

d03

d04

d02

60∘N

58∘N

56∘N

54∘N

52∘N

50∘N

48∘N

46∘N

0∘

5∘E 10

∘E 15∘E

Figure 1 WRF modelling domain 27 km (d01) 9 km (d02) 3 km(d03) and 1 km (d04) grid spacing

Domain Setup The WRF model was run in a series of two-way nested grids (centred in the FINO1 offshore platform atlatitude 540∘N and longitude 635∘E) The horizontal gridspacing was refined by a factor of 3 through three nesteddomains until 1 km resolution In particular the WRF modelwas built over a parent domain (d01) with 67 times 65 horizontalgrid points at 27 km an intermediate nested domain (d02)of 9 km spatial resolution (151 times 121 grid points) and twoinnermost domain (d03) with 3 km spacing (217 times 205 grids)and 1 km spacing of 202 times 190 grids (see Figure 1) On thevertical coordinate 88 vertical levels were used The verticalresolution was 10m upto 200m height to accurately resolvethe lower part of the MABL Above 200m grid spacing isprogressively stretched

Initial and Boundary Conditions The WRF model was usedto refine the state of the atmosphere especially the PBLby downscaling both the global NCEP final analysis (FNL)data and the Era-Interim reanalysis data produced by theECMWF The NCEP FNL data have horizontal resolution of1 times 1 degree (sim100 km) and 52 model levels The Era-Interimreanalysis project covers the period from 1979 to present andhas a spectral T255 horizontal resolution (sim79 km spacing ona N80 reduced Gaussian grid) and 60 vertical model levels


[16] The time step was set equal to 120 seconds to fulfilthe Courant-Friedrichs-Lewy (CFL) condition for horizontaland vertical stability and the first 24 hours were discarded asspin-up time of the model

Topographic Inputs For the WRF the topographic informa-tion was developed using the 20-category moderate reso-lution imaging spectroradiometer (MODIS) WRF terraindatabase The 27 km domain was based on the 5 minutes(sim925 km) global data the 9 km domain was based on the2 minutes (sim370 km) data and the 3 km and 1 km domainswere based on the 30 seconds (sim900m) data

Four-Dimensional Data Assimilation (FDDA) Nudging isone method of FDDA that is implemented in the WRFmodel The WRF model supports three types of nudgingthe three-dimensional upper-air andor surface analysis theobservation and the spectral nudging Because the purposeof the stable case study was mainly to perform the sensitivityexperiments rather than to keep the simulations close tothe reanalysis and measurement data the WRF model wasrun without analysis and observation nudging for this stableperiod When the model was run for the whole March 2005analysis nudging was configured to nudge temperature watervapourmixing ratio and horizontal wind components on theoutermost domain (d01) with time intervals of six hours Thestrategy followed for analysis nudging was based on previousstudies [17]

Physics Schemes Microphysics was modelled using the newThompson scheme with ice snow and graupel processessuitable for high-resolution simulations The rapid radiativetransfermodel (RRTM) (an accurate and widely used schemeusing look-up tables for efficiency) and the Dudhia (a simpledownward integration allowing for efficient cloud and clear-sky absorption and scattering) schemes were used for thelongwave and shortwave radiation options respectively TheNoah land surface model (LSM) was chosen to simulate soilmoisture and temperature and canopy moisture Finally thecumulus physics was modelled with the Grell 3D schemewhich is an improved version of the Grell-Devenyi (GD)ensemble scheme that can be used on high resolution(in addition to coarser resolutions) However the cumulusparameterization was turned off in the fine-grid spacing of3 km and 1 km selected for these simulations Theoreticallyit is only valid for parent grid sizes greater than 9 km [10]Finally four PBL schemes were selected for this study Theseinclude two turbulent kinetic energy (TKE) closure schemesthe Mellor-Yamada-Janjic (MYJ) PBL [18] and the Mellor-YamadaNakanishi andNiino Level 25 (MYNN)PBL [19] andtwo first-order closure schemes the Yonsei University (YSU)PBL [20] and the asymmetric convective model version 2(ACM2) [21] More details of the model of the physicsschemes and references could be found in [10]

3 Results and Discussion

31 Stable Period (16ndash18 March 2005) In this section WRFmodel results are presented and compared with other mod-elling studies and with high quality observations recorded at

the FINO1 offshore platform in the North Sea The modelvalidation is for stable atmospheric conditions during March2005 According to Krogsaeligter [3] and Saint-Drenan et al [5]at the FINO1 offshore platform stable atmospheric conditionsare observed 20of the timeduring spring and early summerIt is well known that the mesoscale models handle both highatmospheric stability and high instability with difficulty dueto their PBL schemes The stable MABL is very complexand its structure is more complicated and variable thanthe structure of the unstableneutral MABL [22] Stableatmospheric conditions aremainly observed during the nightand it is well known that the nocturnal boundary layer isdriven by two distinct processes low turbulence and radiativecooling which both are very difficult to describe and tomodel Therefore making precise offshore wind resourcemaps under stable conditions is a great challenge For therepresentative case of the stable MABL the period from 16to 18 March 2005 at 0000 UTC was selected for analysisSimilar studies have been performed in the past for thisparticular time period and we could therefore compare ourmethodologies and results with that of other researches (eg[8 11])

During the stable period the synoptic situation overthe North Sea is determined by a low-pressure system overthe Atlantic and a high-pressure system over the southernEurope which later in the examined period shifts to thenorth-west [11] In Figure 2 we present the temporal variationof thewind and temperature conditions at the FINO1 offshoreplatform during the stable period As can be seen in Figure 2at the beginning of the period very strong southwesterly(sim210ndash240∘) winds dominate the area which later becomewesterlies (sim270∘) and slow down It can be assumed thatthe MABL structure over the North Sea is influenced bylarge scale circulations since no diurnal variation is observedin the air temperature and wind speed According to Suseljand Sood [11] the MABL structure over the North Sea isprimarily established by the properties of advected air massesand has small or almost no diurnal cycle For example atthe FINO1 platform mainly during the winter the south-westerlies to westerlies advect warm air over the cold searesulting in a stable MABL while north-westerlies to north-easterlies advect cold air resulting in an unstable MABLIn general we could characterise the period 16ndash18 March2005 as a two-day stable period with the air temperaturehigher than the water temperature andwith highwind speeds(see Figure 2) Actually the FINO1 recordings gave even a5∘C difference between the air and the water temperature atthe night of 16 March 2005 corresponding closely to stableconditions Considering the air-water temperature differenceis a very simple and pragmatic method of describing thestability conditions In the current study this method wasshown to be sufficiently precise In addition the Richardsonnumber has been also calculated to confirm the stabilityconditions

To determine the atmospheric stability the Richardsonnumber has been calculated using the WRF model outputsThe Richardson number quantifies the respective contribu-tion of the wind shear and the buoyancy to the production


200210220230240250260

3579

11131517

0000 0600 1200 1800 0000 0600 1200 1800 0000

(deg

)

Time (UTC)

(ms

) (∘

C)

50m temp (∘C)SST (∘C)

50m wspd (ms)

50m wdir (deg)

Figure 2 Temporal variation of the sea-surface temperature (SST)air temperature (∘C) wind speed (ms) and wind direction (degree)at 50m height as these were recorded at the FINO1 platform forthe stable period 16ndash18 March 2005 at 0000 UTC Temperature andwind speed are represented in the left 119910-axis and the wind directionin the right 119910-axis of the plot

0

005

01

015

02

025

160

320

05 0

000

160

320

05 0

600

160

320

05 1

200

160

320

05 1

800

170

320

05 0

000

170

320

05 0

600

170

320

05 1

200

170

320

05 1

800

180

320

05 0

000

Rich

ards

on n

umbe

r

Time (UTC)

Figure 3 Temporal variation of the Richardson number as this wascalculated for the stable period 16ndash18 March 2005 at 0000 UTC

of turbulent kinetic energy [5] and the equation used in thecomputation of the Richardson number was

Ri =119892

1205791

(1205791minus 1205792) (1199111minus 1199112)

(1199061minus 1199062)

2

(5)

where 119892 is the gravitational acceleration 1199061 1199062 1205791 and 120579

2

are the wind speed and the virtual potential temperature atthe height 119911

1and 1199112 respectively [23]

It is worth noting that negative Richardson numbers indi-cate unstable conditions positive values indicate staticallystable flows and values close to zero or zero are indicative ofneutral conditions As an example the Richardson numberis plotted in Figure 3 for the period 16ndash18 March 2005 Thepositive values of the Richardson number (varying from01 to02) confirm the statically stable flows that dominated duringthis particular time period

311 Sensitivity to Horizontal Resolution Increasing the hor-izontal resolution in the mesoscale models increases the abil-ity of the models to resolve major features of the topography

Table 2 Statistical metrics (MAE in ms ME in ms STD of theME in ms and 119877) for the 100m wind speed so as to evaluate theperformance of different horizontal resolutions (3 km versus 1 km)

At 100m MAE (ms) ME (ms) STD (ms) 119877

1 kmmdashERA-I 245 minus231 158 0623 kmmdashERA-I 238 minus227 147 066

and surface characteristics such as the coastal boundariesand therefore produces more accurate wind climatologies[4] Generally speaking the higher the resolution of thesimulation the better the representation of the atmosphericprocesses we obtain [24] especially if the terrain is complex[4] However it is difficult to define a priori the grid spacingneeded to achieve a desired level of accuracy The sensitivityto horizontal resolutions is thus tested in this section todefine optimum grid spacing For the runs the ERA-Interimreanalysis was selected to initialise the model and the YSUscheme to model the MABL

To evaluate the performance of the different horizontalresolutions four statistical parameters were used meanabsolute error (MAE) mean error (ME) standard deviation(STD) of the ME and correlation coefficient (R) As anexample statistical metrics for the 100m wind speed areshown in Table 2 for the 3 km and 1 km spacing It seems thatthe 3 km horizontal resolution results in a slight reductionof MAE bias and STD and improves correlations with theobservations compared to what has been simulated at the1 km spacing We concluded that the 3 km is the optimalresolution and that the finest 1 km spacing does not bringimprovement in theWRF results worth considering It seemsthat the parameterizations used in the WRF model impose alimit to the downscaling beyond which there is a minor orany improvement of the model performance According toTalbot et al [25] it is beneficial to use very fine horizontalresolution (le1 km) during stable atmospheric conditionsand over heterogeneous surfaces when local properties arerequired or when resolving small-scale surface features isdesirable In this modelling study the North Sea can becharacterised as a homogeneous surface and therefore thereis no advantage of using the computationally expensive 1 kmspacing (as the model resolution increases more computerprocessing required) It is worth noting that similar resultsto the one presented in this section were obtained for all thevertical levels below 100m

312 Sensitivity to Input Data The impact of using differentreanalysis datasets on the initialisation of the WRF modelis investigated in this section Comparing the model resultswhen the ERA-Interim and the NCEP data were used at the3 km horizontal resolution and with the YSU PBL scheme itis found that the simulated winds are better correlated withthe FINO1 observations and have lower STD andMAE whenthe ERA-Interim dataset is used (see Table 3) On averageover all the vertical levels the ERA-Interim dataset yields091ms lower wind speeds than recorded 117ms STD andcorrelation between the model and the measurements equal077


Table 3 Statistical metrics (MAE in ms ME in ms STD of the ME in ms and 119877) for wind speed at 8 vertical levels (30ndash100m) so as toevaluate the performance of different input datasets (NCEP versus ERA-Interim) at the 3 kmWRF modelling domain

30m 40m 50m 60m 70m 80m 90m 100m AverageNCEP FNL and YSU PBL

MAE (ms) 109 109 106 123 144 194 220 240 156ME (ms) 051 044 minus025 minus075 minus11 minus18 minus207 minus233 minus092STD (ms) 135 137 139 141 142 142 142 157 142119877 063 062 062 062 062 062 063 059 062

ERA-I and YSU PBLMAE (ms) 089 090 088 107 129 184 210 238 142ME (ms) 046 039 minus023 minus074 minus109 minus175 minus205 minus227 minus091STD (ms) 101 105 110 114 118 121 123 147 117119877 081 080 079 078 077 076 075 066 077

Table 4 A synopsis of the WRF model configuration used by EDFRampD for the stable period 16ndash18 March 2005

Simulation period 15ndash20032005 0000 UTCModel version V34Domains 4Horizontal resolution 27 9 3 and 1 km

Grid sizes 67 times 65 151 times 121 217 times 205and 202 times 190

Vertical resolution 88 (eta levels)Input data ERA-InterimTime step 120 sOutputs frequency 180 180 60 and 60 minutesGridded analysis nudging NONesting 2-way nestingPhysics schemes

Microphysics NewThompsonCumulus Grell 3DShortwave DudhiaLongwave RRTMLSM NoahPBL YSU MYNN MYJ ACM2

Note that in both experiments with different input dataan increased MAE with the elevation is observed and theWRF model is found to overestimate the winds below 40mand underestimate them above that height It is also observedthat the correlations slightly decrease and the bias continuesto increase as the altitude increases this trend in the meanerror and correlation coefficient with the increasing heightis an indication that the current model configuration needsfurther improvement It seems that this decreasing trend ismore pronounced when the ERA-Interim data were usedbut the average results (considering the average statisticalmetrics correlation bias STD and MAE) were closer to theobservations with this dataset As will be shown later thesereduced correlations with increasing height are a result ofseveral factors including the input data and mainly the PBLscheme

313 Sensitivity to PBL Schemes it is well accepted that theaccuracy of the simulated by the mesoscale models offshorewinds is strongly affected by the PBL schemes [26]Thereforea way to improve the WRF performance at the lower levelsof the atmosphere is to try different PBL parameterisationsThe behaviour of four different PBL schemes (YSU MYJMYNN and ACM2) on the stable MABL is examined andthe importance of defining the appropriate model setup forstudying pure offshore wind conditions is noticed in thissectionThe PBL experiment was run four times one run foreach PBL scheme A synopsis of the model configuration isgiven in Table 4

Statistical metrics (MAE ME STD and R) for the windspeed at heights from 30m upto 100m are presented inTable 5 Note that the MYNN PBL scheme predicts lowervalues of MAE and ME in all vertical levels than the otherPBL schemes It seems that the MYNN scheme is the onethat yields to the highest degree of correctness mainly above50m and thus showing in average the best agreementwith theFINO1 observations It is also noticed that the bias betweenthe WRF model and the measurements is lower and closeto the surface for the YSU MYJ and ACM2 PBL runswhereas the bias increases with altitude The YSU run seemsto correlate better with the FINO1 measurements than theother PBL schemes below 70m but above that height theME significantly increases and lower values of correlationwere found On the other hand theMYNN scheme performsalmostwith the same accuracy at all heightswith low values ofME and STD In particular the MYNN scheme resulted (inaverage over all the heights) in 001ms higher wind speedsthan measured STD of 121ms and correlation coefficientequal to 066 (see Table 5)

It is well known that the atmospheric stability has asignificant effect on the vertical wind profile and on estimatesof the wind resource at a given height [4] A detailed analysisof the offshore vertical wind profile upto hub heights andabove is important for a correct estimation of the MABLwind and stability conditions wind resource and powerforecasting

The time-averaged vertical profiles of wind speed tem-perature andwind direction for all the PBL runs are providedin Figure 4 for comparison with the FINO1 measurements


Table 5 Statistical metrics (MAE in ms ME in ms STD of the ME in ms and 119877) for the wind speed at 8 vertical levels (30ndash100m) so asto evaluate the performance of four PBL schemes (YSU MYJ ACM2 and MYNN) at the 3 kmWRF modelling domain

30m 40m 50m 60m 70m 80m 90m 100m AverageYSU PBL


MYJ PBLMAE (ms) 138 125 143 155 156 179 176 164 155ME (ms) minus103 minus084 minus112 minus125 minus123 minus152 minus146 minus133 minus122STD (ms) 124 123 124 127 132 136 139 143 131119877 068 067 067 066 064 063 063 065 065

ACM2 PBLMAE (ms) 108 097 120 138 146 177 179 186 144ME (ms) minus074 minus056 minus091 minus115 minus123 minus161 minus164 minus159 minus118STD (ms) 113 113 115 117 118 120 121 141 120119877 067 066 065 063 061 061 060 056 062

MYNN PBLMAE (ms) 091 092 088 091 093 100 099 103 095ME (ms) 012 038 016 001 003 minus027 minus021 minus012 001STD (ms) 122 121 120 121 123 123 126 128 123119877 067 067 066 066 064 064 063 067 066

30405060708090

100

12 13 14 15 16 17 18 19 20

Hei

ght (

m)

OBSMYNNACM2

MYJYSU

(a) Time-averaged wind speed (ms)

30405060708090

100

230 235 240 245 250

Hei

ght (

m)

OBSMYNNACM2

MYJYSU

(b) Time-averaged wind direction (degree)

30405060708090

100

5 6 7 8 9 10

Hei

ght (

m)

OBSMYNNMYJ

ACM2YSU

(c) Time-averaged temperature (C)

Figure 4 Time-averaged (16ndash18 March 2005) vertical profiles upto 100m ASL for (a) wind speed (ms) (b) wind direction (degree) and (c)temperature (∘C) The model results of different PBL schemes are compared with the FINO1 measurements


0100

0203

04

05

06

07

08

09

095

099

10

2

2

3

3

44 1

1

250

225

200

175

150

125

100

075

050

025

000025 050 075 REF 125 150 175 200 225 250

Stan

dard

ized

dev

iatio

ns (n

orm

aliz

ed)

Correlation

Figure 5 Taylor diagram displaying a statistical comparison withfourWRFmodel estimates (YSU 1MYJ 2MYNN 3 andACM2 4)of the wind speed during the stable period 16ndash18March 2005 Red isthe EDF statistics and blue is the Munoz-Esparza et al [8] statistics

This kind of representation allows us to have some under-standing about the behaviour of different PBL schemes inthe lower levels of the MABL The highly stable conditionsobserved during the examined period pose a particularchallenge As can be seen in Figure 4 the YSU scheme wasin good agreement with the observed data at the 30m heightbut higher up underestimated the wind resource This resulthighlights the importance of studying the whole MABL foroffshore wind energy applications Also an increase in windspeed with increasing altitude is noticed at all PBL runs butonly the MYNN scheme produces wind speeds which are inexceptionally good agreement with the observations It seemsthat the MYNN PBL scheme manages to model the correctamount of mixing in the boundary layerTheMYNN schemeis also shown to be the most consistent with the temperaturemeasurements (approximately 1∘C difference at 100m height)and gives less than 10∘ difference in the wind direction at allvertical levels (see Figure 4)

Comparison with Other Modelling StudiesMunoz-Esparza etal [8] in their WRF modelling study for the stable period16ndash18March 2005 compared six different PBL schemesTheirmodel configuration is presented in Table 6

Statistical metrics (ME RMSE and R) of the 100m windspeed for the common PBL runs as these were calculated byMunoz-Esparza et al [8] and by EDF are shown in Table 7Note that all PBL runs of this study result in a decreased biasand RMSE as well as in an increased correlation coefficientThe differences between EDFrsquos andMunoz-Esparza et alrsquos [8]model setup such as the input data the number of the verticallevels and in general the combination of physics schemes(eg microphysics and cumulus) may be a reason for thebetter accuracy in this study For example EDF initialised

Table 6 A synopsis of the WRF model configuration used byMunoz-Esparza et al [8] for the stable period 16ndash18 March 2005

Simulation period 14ndash18032005 0000 UTCModel version V32Domains 4Horizontal resolution 27 9 3 and 1 kmVertical resolution 46Input data NCEPGridded analysis nudging NONesting 2-way nestingPhysics schemes

Microphysics WSM3Cumulus Kain-FritschShortwave DudhiaLongwave RRTMLSM Noah

PBL YSU MYJ MYNN ACM2QNSE BouLac

Table 7Munoz-Esparza et al [8] versus EDF statisticalmetrics (MEin ms RMSE in ms and 119877) for the 100m wind speed so as toevaluate the WRF performance with different PBL schemes

PBLschemes

[8] EDFME(ms)

RMSE(ms) 119877 (mdash) ME

(ms)RMSE(ms) 119877 (mdash)

YSU minus283 321 046 minus227 270 066MYJ minus193 260 038 minus133 194 065QNSE minus123 190 053 mdash mdash mdashMYNN minus044 156 055 minus012 128 067ACM2 minus161 213 059 minus159 211 056BouLac minus308 342 046 mdash mdash mdash

the model with the ERA-Interim data instead of the NCEPreanalysis and used 88 vertical levels instead of the 46 usedby and Munoz-Esparza et al [8] It is well expected that partof the problem with the simulation of stable conditions in theMABL is having enough vertical levels A reason is the factthat a stable layer acts as an effective barrier to mixing and ifthe model layers at the lower levels of the MABL are widelyspaced the simulated barrier may be too weak [4]

Another important point seen in Table 7 is that bothMunoz-Esparza et al [8] and EDF concluded that the bestmodel performance was observed when the MYNN PBLschemewas selected and that theWRFmodel tends to under-estimate the wind speed at the hub height The similaritiesbetween the PBL experiments ofMunoz-Esparza et al [8] andEDF are quantified in terms of their correlation their RMSEand their STD by using the Taylor diagram [27] In particularFigure 5 is a Taylor diagram which shows the relative skillwith which the PBL runs of Munoz-Esparza et al [8] and ofEDF simulate the wind speeds during the stable period 16ndash18 March 2005 Statistics for four PBL runs (YSU 1 MYJ 2MYNN 3 andACM2 4) are used in the plot and the position


Table 8 Statistical metrics (MAE in ms ME in ms STD of the ME in ms and 119877) for the wind speed at 8 vertical levels (30ndash100m) so asto evaluate the performance of one- and two-way nesting options at the 3 kmWRF modelling domain

30m 40m 50m 60m 70m 80m 90m 100m AverageMYNN and 2-way

MAE (ms) 091 092 088 091 093 100 099 103 095ME (ms) 012 038 016 001 003 minus027 minus021 minus012 001STD (ms) 122 121 120 121 123 123 126 128 123119877 067 067 066 066 064 064 063 067 066

MYNN and 1-wayMAE (ms) 097 100 098 100 102 107 108 111 103ME (ms) 017 043 021 005 008 minus022 minus017 minus007 006STD (ms) 128 128 128 130 133 134 136 140 132119877 064 063 062 061 059 059 058 062 061

of each number appearing on the plot quantifies how closelythat run matches observations If for example run number3 (MYNN PBL run) is considered its pattern correlationwith observations is about 067 for the EDF MYNN PBLexperiment (red colour) and 055 for theMunoz-Esparza et al[8]MYNNPBL run (blue colour) It seems that the simulatedpatterns between EDF and [8] that agree well with each otherare for the ACM2 PBL run (number 4) and the pattern thathas relatively high correlation and low RMSE are for the EDFMYNN PBL run (red-number 3)

314 Sensitivity to Nesting Options In the WRF model thehorizontal nesting allows resolution to be focused over aparticular region by introducing a finer grid (or grids) intothe simulation [10] Therefore a nested run can be defined asa finer grid resolution model run in which multiple domains(of different horizontal resolutions) can be run either inde-pendently as separate model simulations or simultaneously

The WRF model supports one-way and two-way gridnesting techniques where one-way and two-way refer tohow a coarse and a fine domain interact In both the one-way and two-way nesting options the initial and lateralboundary conditions for the nest domain are provided by theparent domain together with input from higher resolutionterrestrial fields and masked surface fields [10] In the one-way nesting option information exchange between the coarsedomain and the nest is strictly downscale which means thatthe nest does not impact the parent domainrsquos solution Inthe two-way nest integration the exchange of informationbetween the coarse domain and the nest goes both ways Thenestrsquos solution also impacts the parentrsquos solution

To this point the best model performance was achievedwith the following combination Era-Interim as input andboundary data 3 km as the optimal horizontal resolutionand the MYNN PBL scheme to simulate the MABL In thissection both the one-way and two-way nesting options withthe aforementioned model combination were tested so as toinvestigate how the different nesting options can affect theWRF performance

It seems that the differences between the two-way andone-way nesting options are small However it can be con-cluded that the two-way nest integration produces on average

wind speeds closer to the FINO1measurements (see Table 8)When the two-nesting option was selected the average overall heights ME was 001ms and the STD of the ME was123ms (instead of 006ms ME and 132ms STD for theone-way nesting option) As can be also seen in Table 8 theWRF model resulted in smaller MAE (averaged over all theheights) and correlated better with the observations whenthe two-way nesting option was selected It is thus concludedthat when the exchange of information between the coarsedomain and the nest goes both ways it results in better modelperformance

315 Summary The validation of the WRF model at theFINO1 met mast during the stable case study (16ndash18 March2005) resulted in the following conclusions in terms of modelconfiguration

(i) Selecting the finest horizontal resolution of 1 kminstead of the 3 km does not bring improvement inthe WRF results worth considering it was very timeconsuming in terms of computational time

(ii) Using the ERA-Interim reanalysis data instead of theNCEP FNL data allows reducing the bias betweensimulated and measured winds

(iii) Changing PBL schemes has a strong impact on themodel resultsTheMYNNPBL schemewas in the bestagreement with the observations (confirmed by othermodelling studies as well)

(iv) A small improvement in the model performance wasobserved when the two-way nesting option was used

32 March 2005 Accuracy verification at the FINO1 plat-form indicated that theWRFmethodology developed for thestable period (16ndash18 March 2005) results in more accuratewind simulation than any other simulation in the studyof Munoz-Esparza et al [8] though there are similaritiesbetween the studies in terms of findings But is the concludedWRF model configuration appropriate for other stabilityconditions (eg neutral and unstable) and for longer timeperiods during which all atmospheric stability conditions areobserved


000020040

Win

d sp

eed

bias

(ms

)

MYNNACM2

MYJYSU

minus100

minus080

minus060

minus040

minus020

(a)

MYNNACM2

MYJYSU

000050100150200250

AverageWin

d sp

eed

STD

(ms

)

30m 40m 50m 60m 70m 80m 90m 100m

(b)

Figure 6 Monthly (March 2005) bias and standard deviation for wind speed at 8 vertical levels (30ndash100m) and their average for the fourPBL runs (YSU MYJ ACM2 and MYNN)

55N548N546N544N542N54N

538N536N536N534N532N53N

35E 4E 45E 5E 55E 6E 7E65E 75E 85E8E 9E

78 81 84 87 9 93 9996 102 105 108 111

Figure 7 Average wind speed (ms) and wind direction at 100mheight over the 3 km WRF modelling domain during March 2005when the MYNN PBL scheme was selected

To answer this question the WRF model is validated inthis section for the whole of March 2005 in which neutralunstable and stable atmospheric conditions are observedTo determine the atmospheric stability at the FINO1 metmast forMarch 2005 the Richardson number was calculatedIt was found out that 35 of the time during March 2005stable atmospheric conditions were dominant at the FINO1offshore platform while 65 of the time unstable and neutralconditions were observed

321 Sensitivity to PBL Schemes The WRF model is testedwith the same four PBL schemes (YSU MYJ ACM2 andMYNN) used for the stable scenario The monthly bias andstandard deviation for each height of the FINO1 offshoreplatform is given in Figure 6 for each PBL run In Figure 6 wecan clearly observe that the MYNN PBL run gives the lowestbias at all vertical heights and the second lowest standarddeviation The biases of the YSU and ACM2 PBL schemestend to increase as the height increases As before for thestable case study we concluded that theMYNN scheme is theone that yields to the highest degree of correctness with theFINO1 observational data In particular the MYNN schemeresulted in average over all the heights in 001ms lower windspeeds STD of 171ms and very high correlation coefficientequal to 093 (see Table 9)

Table 9 For March 2005 statistical metrics (ME in ms STD of theME in ms and 119877) for wind speed [averaged over all the heights 30to 100m)] so as to evaluate the performance of four PBL schemes(YSU MYJ ACM2 and MYNN) in the WRF model

PBL YSU MYJ ACM2 MYNNME (ms) minus043 minus026 minus048 minus001STD (ms) 175 175 169 171119877 092 092 092 093

In Figure 7 a map plot of the average 100m wind speedand wind direction as these were simulated during March2005 from theMYNNPBL run over the 3 kmWRFmodellingdomain is provided At the FINO1 mast (latitude 540∘Nand longitude 635∘E) and the areas nearby the platform overthe North Sea the WRF model simulated hub height-windspeeds of the order of 112ms and westerly to southwesterlydirection In Figure 8 the wind roses are shown for March2005 from both the FINO1 observations and all theWRF PBLsimulations It seems that the model in every PBL simulationoverestimates the occurrence of wind speed in the range 12ndash16ms and results in a small veering of the winds in theclockwise direction between 0∘ and 180∘

322 Error Analysis and Q-Q Plots In Figure 9 the Q-Q plots of the 100m wind speed for March 2005 displaya quantile-quantile plot of two samples modelled versusobserved Each blue point in the Q-Q plots corresponds toone of the quantiles of the distribution the modelled windfollows and is plotted against the same quantile of the FINO1observationsrsquo distribution If the samples do come from thesame distribution the plot will be linear and the blue pointswill lie on the 45∘ line 119910 = 119909

As can be seen in Figure 9 the two distributions (the onemodelled with the MYNN PBL scheme versus the observedone at FINO1) being compared are almost identical Theblue points in the Q-Q plot are closely following the 45∘line 119910 = 119909 with high correlation coefficient equal to093 (see Table 9) The hourly observed and modelled windspeeds for March 2005 follow the Weibull distribution andupon closer inspection of Figure 10 the Weibull distributionresulted from the MYNN PBL run is the one that fits very


10

20

30

FINO1

West East

South

(a)

(d) (e)

(b) (c)

North

10

20

30

WRF-ACM2

West East

South

North

10

20

30

WRF-YSU

West East

South

North

10

20

30

WRF-MYJ

West East

South

North

10

20

30

West East

South

NorthWRF-MYNN

0ndash44ndash8

8ndash1212ndash16

16ndash2020ndash24

Figure 8 Wind roses based on March 2005 data from the FINO1 measurements and the WRF PBL runs (ACM2 YSU MYJ and MYNN) at100m height

well with the FINO1 observed distribution at the 100mheight In particular the shape and scale parameters of theWeibull distribution as these were calculated with the FINO1measurements were 251 and 1274 respectively The shapeparameter modelled by the WRF model varied from 220 to285 and the scale parameter ranged from 1141 to 1251 for thedifferent PBL schemes used However only the MYNN PBLscheme resulted in shape and scale parameters closer to theone observed at the offshore platform (see Figure 10)

In order to further understand the performance of theWRF model and show the variation of the model results inthis section an error analysis is also performed In general theerrors are assumed to follow normal distributions about theirmean value and so the standard deviation is themeasurementof the uncertainty If it turns out that the random errors inthe process are not normally distributed then any inferencesmade about the process may be incorrect In Figure 9 we

provide a normal probability plot of theMEof the 100mwindspeed The plot includes a reference line indicating that theMEof thewind speed (blue points) closely follows the normaldistribution Figure 9 also shows a histogram of the windspeed mean error In the 119910-axis of this histogram we havethe number of records with a particular velocity error Theblue bars of the histogram show the relative frequency withwhich each wind speed error (which is shown along the 119909-axis) occurs Clearly the ME follows the normal distributionand the mean is very close to zero which is an indication thatthe mean error is unbiased

4 Conclusions

The main goals of the present study were to achieve a betterunderstanding of the offshore wind conditions to accurately


0 2 4 6 8 10 12 14 16 18 20 2202

4

68

10

12

14

16

18

20

22

Observation

WRF

H = 100m R = 0933 ratio data = 07

(a)

0 2 4 6

00010003

001002005010

025

050

075

090095098099

09970999

Prob

abili

ty

Normal probability plot

minus8 minus6 minus4 minus2

Error 100m

(b)

0 5 10 150

10

20

30

40

50

60

70

Error 100mminus15 minus10 minus5

(c)

Figure 9 (a) Q-Q plot theWRFMYNNPBL run (y-axis) against the FINO1 100mwind speed distribution (x-axis) (b) Normal distributionagainst the 100m wind speed ME and (c) histogram of the wind speed ME

simulate the wind flow and atmospheric stability occurringoffshore over the North Sea and to develop a mesoscalemodelling approach that could be used for more accurateWRA In order to achieve all these goals this paper examinedthe sensitivity of the performance of the WRF model to theuse of different initial and boundary conditions horizontalresolutions and PBL schemes at the FINO1 offshore platform

The WRF model methodology developed in this studyappeared to be a very valuable tool for the determinationof the offshore wind and stability conditions in the NorthSea It resulted in more accurate wind speed fields than theone simulated in previous studies though there were similarfindings It was concluded that the mesoscale models mustbe adaptive in order to increase their accuracy For example

the atmospheric stability on the MABL (especially the stableatmospheric conditions) the topographic features in orderto adjust the horizontal resolution accordingly and the inputdatasets are needed to be taken into account

It was concluded that the 3 km spacing is an optimal hor-izontal resolution for making precise offshore wind resourcemaps Also using the ERA-Interim reanalysis data instead ofthe NCEP FNL data allows us to reduce the bias betweensimulated and measured winds Finally it was found thatchanging PBL schemes has a strong impact on the modelresults In particular for March 2005 the MYNN PBL runyields in the best agreement with the observations with001ms bias standard deviation of 171ms and correlationof 093 (averaged over all the vertical levels)


0 5 10 15 20 25 300

001002003004005006007008009

01

Prob

abili

ty

Wind speed (msminus1)

FINO1 (k = 251 120582 = 1274)

WRF-ACM2 (k = 220 120582 = 1141)WRF-MYJ (k = 285 120582 = 1246)WRF-YSU (k = 278 120582 = 1203)

WRF-MYNN (k = 268 120582 = 1251)

Figure 10 Weibull distribution based onMarch 2005 data from theFINO1 measurements and the WRF PBL runs (ACM2 YSU MYJand MYNN) at 100m height

In future studies in order to increase the reliability of thewind simulations and forecasting it is necessary to expandthe simulation period as long as possible

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

References

[1] O Krogsaeligter J Reuder and G Hauge ldquoWRF and the marineplanetary boundary layerrdquo in EWEA pp 1ndash6 Brussels Belgium2011

[2] A C Fitch J B Olson J K Lundquist et al ldquoLocal andMesoscale Impacts of Wind Farms as Parameterized in aMesoscale NWPModelrdquoMonthly Weather Review vol 140 no9 pp 3017ndash3038 2012

[3] O Krogsaeligter ldquoOne year modelling and observations of theMarineAtmospheric Boundary Layer (MABL)rdquo inNORCOWEpp 10ndash13 2011

[4] M Brower J W Zack B Bailey M N Schwartz and D LElliott ldquoMesoscale modelling as a tool for wind resource assess-ment and mappingrdquo in Proceedings of the 14th Conference onApplied Climatology pp 1ndash7 American Meteorological Society2004

[5] Y-M Saint-Drenan S Hagemann L Bernhard and J TambkeldquoUncertainty in the vertical extrapolation of the wind speed dueto the measurement accuracy of the temperature differencerdquo inEuropean Wind Energy Conference amp Exhibition (EWEC rsquo09)pp 1ndash5 Marseille France March 2009

[6] B Canadillas D Munoz-esparza and T Neumann ldquoFluxesestimation and the derivation of the atmospheric stability at theoffshore mast FINO1rdquo in EWEAOffshore pp 1ndash10 AmsterdamThe Netherlands 2011

[7] D Munoz-Esparza and B Canadillas ldquoForecasting the DiabaticOffshore Wind Profile at FINO1 with the WRF MesoscaleModelrdquo DEWI Magazine vol 40 pp 73ndash79 2012

[8] D Munoz-Esparza J Beeck Van and B Canadillas ldquoImpact ofturbulence modeling on the performance of WRF model foroffshore short-termwind energy applicationsrdquo in Proceedings ofthe 13th International Conference on Wind Engineering pp 1ndash82011

[9] M Garcia-Diez J Fernandez L Fita and C Yague ldquoSeasonaldependence of WRF model biases and sensitivity to PBLschemes over Europerdquo Quartely Journal of Royal MeteorologicalSociety vol 139 pp 501ndash514 2013

[10] W Skamarock J Klemp J Dudhia et al ldquoA description ofthe advanced research WRF version 3rdquo NCAR Technical Note475+STR 2008

[11] K Suselj and A Sood ldquoImproving the Mellor-Yamada-JanjicParameterization for wind conditions in the marine planetaryboundary layerrdquo Boundary-Layer Meteorology vol 136 no 2pp 301ndash324 2010

[12] F Kinder ldquoMeteorological conditions at FINO1 in the vicinityof Alpha Ventusrdquo in RAVE Conference Bremerhaven Germany2012

[13] J M Potts ldquoBasic conceptsrdquo in Forecast Verification A Prac-titionerrsquos Guide in Atmospheric Science I T Jolliffe and D BStephenson Eds pp 11ndash29 JohnWiley amp Sons New York NYUSA 2nd edition 2011

[14] E M Giannakopoulou and R Toumi ldquoThe Persian Gulf sum-mertime low-level jet over sloping terrainrdquoQuarterly Journal ofthe Royal Meteorological Society vol 138 no 662 pp 145ndash1572012

[15] EMGiannakopoulou andR Toumi ldquoImpacts of theNileDeltaland-use on the local climaterdquo Atmospheric Science Letters vol13 no 3 pp 208ndash215 2012

[16] D P Dee S M Uppala A J Simmons et al ldquoThe ERA-Interimreanalysis configuration and performance of the data assim-ilation systemrdquo Quarterly Journal of the Royal MeteorologicalSociety vol 137 no 656 pp 553ndash597 2011

[17] P Liu A P Tsimpidi Y Hu B Stone A G Russell and ANenes ldquoDifferences between downscaling with spectral andgrid nudging using WRFrdquo Atmospheric Chemistry and Physicsvol 12 no 8 pp 3601ndash3610 2012

[18] G L Mellor and T Yamada ldquoDevelopment of a turbulenceclosure model for geophysical fluid problemsrdquo Reviews ofGeophysics amp Space Physics vol 20 no 4 pp 851ndash875 1982

[19] M Nakanishi and H Niino ldquoAn improved Mellor-YamadaLevel-3 model with condensation physics its design and veri-ficationrdquo Boundary-Layer Meteorology vol 112 no 1 pp 1ndash312004

[20] S-Y Hong Y Noh and J Dudhia ldquoA new vertical diffusionpackage with an explicit treatment of entrainment processesrdquoMonthly Weather Review vol 134 no 9 pp 2318ndash2341 2006

[21] J E Pleim ldquoA combined local and nonlocal closure modelfor the atmospheric boundary layer Part II application andevaluation in a mesoscale meteorological modelrdquo Journal ofApplied Meteorology and Climatology vol 46 no 9 pp 1396ndash1409 2007

[22] B Storm J Dudhia S Basu A Swift and I GiammancoldquoEvaluation of the weather research and forecasting model onforecasting low-level jets implications for wind energyrdquo WindEnergy vol 12 no 1 pp 81ndash90 2009

[23] J L Woodward ldquoAppendix A atmospheric stability classifica-tion schemesrdquo in Estimating the Flammable Mass of a VaporCloud pp 209ndash212 American I Wiley Online Library 1998


[24] L Fita J Fernandez M Garcıa-Dıez and M J GutierrezldquoSeaWind project analysing the sensitivity on horizontal andvertical resolution on WRF simulationsrdquo in Proceedings of the2nd Meeting on Meteorology and Climatology of the WesternMediterranean (JMCMO rsquo10) 2010

[25] C Talbot E Bou-Zeid and J Smith ldquoNested mesoscale large-Eddy simulations with WRF performance in real test casesrdquoJournal of Hydrometeorology vol 13 no 5 pp 1421ndash1441 2012

[26] S Shimada T Ohsawa T Kobayashi G Steinfeld J Tambkeand D Heinemann ldquoComparison of offshore wind speedprofiles simulated by six PBL schemes in the WRF modelrdquoin Proceedings of the 1st International Conference Energy ampMeteorology Weather amp Climate for the Energy Industry (IECMrsquo11) pp 1ndash11 November 2011

[27] K E Taylor ldquoSummarizing multiple aspects of model perfor-mance in a single diagramrdquo Journal of Geophysical Research Dvol 106 no 7 pp 7183ndash7192 2001

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ClimatologyJournal of

EcologyInternational Journal of


EarthquakesJournal of


Hindawi Publishing Corporationhttpwwwhindawicom

Applied ampEnvironmentalSoil Science

Volume 2014

Mining


Journal of

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014

International Journal of

Geophysics

OceanographyInternational Journal of


Journal of Computational Environmental SciencesHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal ofPetroleum Engineering


GeochemistryHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

Atmospheric SciencesInternational Journal of


OceanographyHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Advances in


MineralogyInternational Journal of


MeteorologyAdvances in

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Paleontology JournalHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

ScientificaHindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Geological ResearchJournal of


Geology Advances in


In the mesoscale models the vertical stratification isinherently included and the different PBL schemes influ-ence the accuracy of the simulated winds and atmosphericstratification in the MABL [8] For instance some PBLparameterisations are best in unstableneutral stratificationssome others are best in stable atmospheric conditions andsome in both [3] It is alsowidely known that the performanceof the PBL schemes in the mesoscale models depends on thearea of interest the meteorological variables examined andseason and time of day considered and one cannot identifya best model configuration in a general sense [9] The mostcommon approach to deal with this problem is to carry outstatistical studies over a sensible set of model configurationsand finally use the PBL scheme that reproduces better theobservations in an average sense

The attention of this study has been focused on thedevelopment of a mesoscale methodology by using theweather research and forecasting (WRF) model [10] so asto accurately simulate the wind conditions and to reproducethe stability effects on the wind profile over the MABL inthe North Sea The present work describes the verificationof the WRF model by comparing the model results withthat of other modelling studies (eg [8 11]) and with highquality observations recorded at the FINO1 offshore plat-form in the North Sea For this verification process severalWRF modelling simulations have been performed duringMarch 2005 In particular different horizontal resolutionsPBL parameterizations initial and boundary conditions andnesting options were tested The high probability (20) ofoccurrence of stable to very-stable atmospheric stratificationsituations in the spring and early summer at the FINO1platform and the impact of the high stability on the windprofile increases the need to focus on the stable atmosphericconditions and include them in our WRF model configura-tion andmesoscale methodology [3 5] A time period duringwhich all the atmospheric stability conditions are observedare also investigated in this study

A description of the verification process including infor-mation about the data sources and the statistics used foranalysis is given in Section 21 Section 22 focuses on theWRF model setup and the process followed for testing itsperformance Description of the synoptic conditions duringthe stable period in March 2005 and presentation of theWRF results for this case study with a number of keyderived statistics are provided in Section 3 In the samesection presentation of theWRF results for the whole March2005 with statistical metrics and error analysis is also givenConclusions follow in Section 4

2 Materials and Methods

21 Verification Process Two main steps of the verificationprocess are the definition of the data sources and the def-inition of the verification statistics to be produced by theanalysis

211 Observational Data FINO1 Measurements The FINO1offshore platform in Southern North Sea is located 45 kmnorth of the Borkum island (latitude 540∘N and longitude

635∘E) and performs multilevel measurements of windspeed wind direction air temperature relative humidity andair pressure since 2004 The height of the measurement mastis about 100m above mean sea level (MSL) Three ultrasonicinstruments (of 10Hz temporal resolution) are located at415m 615m and 815m height on northwesterly orientedbooms In addition eight cup anemometers with the lowerresolution of 1Hz are installed at different heights startingfrom 34m upto about 100m (every 10m) on boomsmountedin southeast direction of themeteorologicalmast (eg [8 11])

A list of parameters used for the analysis of the meteo-rological conditions and for the comparison with the WRFmodel results is shown in Table 1 In Table 1 we also providethe heights of the recorded parameters and the sensor typesincluding their accuracy according to Canadillas [6] Forthe WRF model validation we used data from the cupanemometers in order to have the best spatial coverage todescribe the vertical profiles of wind speed and wind direc-tion According to Deutsches Windenergie-Institut (DEWIGerman Wind Energy Institute) scientists at the RAVE 2012conference [12] the sonic anemometer data were sparse andwere associated with errors especially during the period2004ndash2007 In general from 2004 to 2011 the availabilityof the cup anemometers at the FINO1 platform was about98 while the availability of the sonic anemometers wasapproximately 83 [12]

212 Define Verification Statistics The following statisticalmetrics have been used in this study to verify the performanceof the WRF model when compared with the FINO1 obser-vations More details on the verification statistics could befound in [13]

Bias or mean error (ME) is defined as the mean ofthe differences between the WRF simulated meteorologicalparameters and the FINO1 observations In particular themean error is calculated for each hour of data and the timeaverage (over the period we have considered to study) biasis then provided for each measuring height of the FINO1platform

Mean absolute error (MAE) is defined as the quantity usedtomeasure how close the observed values are to themodelledones The MAE is given by

MAE = 1119899

119899

sum

119894=1

1003816100381610038161003816

119909119894minus 119910119894

1003816100381610038161003816

(1)

where |119909119894minus 119910119894| is the absolute error with 119909

119894and 119910

119894to repre-

sent the modelled and the observed values at the FINO1 metmast respectively 119899 is the sample size

Standard Deviation (STD) of the ME is defined as thedispersion of the biased values around the mean value A lowstandard deviation indicates that the data points tend to bevery close to themean high standard deviation indicates thatthe data points are spread out over a large range of valuesTheSTD uses the following formula

120590 =radic

sum

119899

119894=1

(119909119894minus 119909)

2

(119899 minus 1)

(2)

where 120590 is the standard deviation and 119909 is the sample mean





RMSE = radic1205902 + 1199092 (3)



119877 =

sum

119899

119894=1

(119909119894minus 119909) (119910

119894minus 119910)

radicsum

119899

119894=1

(119909119894minus 119909)

2

(119910119894minus 119910)

2

(4)




d03

d04

d02

60∘N

58∘N

56∘N

54∘N

52∘N

50∘N

48∘N

46∘N

0∘

5∘E 10

∘E 15∘E















200210220230240250260

3579

11131517

0000 0600 1200 1800 0000 0600 1200 1800 0000

(deg

)

Time (UTC)

(ms

) (∘

C)


50m wspd (ms)

50m wdir (deg)


0

005

01

015

02

025

160

320

05 0

000

160

320

05 0

600

160

320

05 1

200

160

320

05 1

800

170

320

05 0

000

170

320

05 0

600

170

320

05 1

200

170

320

05 1

800

180

320

05 0

000

Rich

ards

on n

umbe

r

Time (UTC)



Ri =119892

1205791

(1205791minus 1205792) (1199111minus 1199112)

(1199061minus 1199062)

2

(5)


2

































30405060708090

100

12 13 14 15 16 17 18 19 20

Hei

ght (

m)

OBSMYNNACM2

MYJYSU


30405060708090

100

230 235 240 245 250

Hei

ght (

m)

OBSMYNNACM2

MYJYSU


30405060708090

100

5 6 7 8 9 10

Hei

ght (

m)

OBSMYNNMYJ

ACM2YSU




0100

0203

04

05

06

07

08

09

095

099

10

2

2

3

3

44 1

1

250

225

200

175

150

125

100

075

050

025

000025 050 075 REF 125 150 175 200 225 250

Stan

dard

ized

dev

iatio

ns (n

orm

aliz

ed)

Correlation










PBLschemes

[8] EDFME(ms)
























000020040

Win

d sp

eed

bias

(ms

)

MYNNACM2

MYJYSU

minus100

minus080

minus060

minus040

minus020

(a)

MYNNACM2

MYJYSU

000050100150200250

AverageWin

d sp

eed

STD

(ms

)

30m 40m 50m 60m 70m 80m 90m 100m

(b)


55N548N546N544N542N54N

538N536N536N534N532N53N

35E 4E 45E 5E 55E 6E 7E65E 75E 85E8E 9E

78 81 84 87 9 93 9996 102 105 108 111










10

20

30

FINO1

West East

South

(a)

(d) (e)

(b) (c)

North

10

20

30

WRF-ACM2

West East

South

North

10

20

30

WRF-YSU

West East

South

North

10

20

30

WRF-MYJ

West East

South

North

10

20

30

West East

South

NorthWRF-MYNN

0ndash44ndash8

8ndash1212ndash16

16ndash2020ndash24





4 Conclusions



0 2 4 6 8 10 12 14 16 18 20 2202

4

68

10

12

14

16

18

20

22

Observation

WRF

H = 100m R = 0933 ratio data = 07

(a)

0 2 4 6

00010003

001002005010

025

050

075

090095098099

09970999

Prob

abili

ty



Error 100m

(b)

0 5 10 150

10

20

30

40

50

60

70


(c)







0 5 10 15 20 25 300

001002003004005006007008009

01

Prob

abili

ty


FINO1 (k = 251 120582 = 1274)


WRF-MYNN (k = 268 120582 = 1251)





References






































Volume 2014

Mining


Journal of



Geophysics







Journal of




Advances in











Geology Advances in





RMSE = radic1205902 + 1199092 (3)



119877 =

sum

119899

119894=1

(119909119894minus 119909) (119910

119894minus 119910)

radicsum

119899

119894=1

(119909119894minus 119909)

2

(119910119894minus 119910)

2

(4)




d03

d04

d02

60∘N

58∘N

56∘N

54∘N

52∘N

50∘N

48∘N

46∘N

0∘

5∘E 10

∘E 15∘E















200210220230240250260

3579

11131517

0000 0600 1200 1800 0000 0600 1200 1800 0000

(deg

)

Time (UTC)

(ms

) (∘

C)


50m wspd (ms)

50m wdir (deg)


0

005

01

015

02

025

160

320

05 0

000

160

320

05 0

600

160

320

05 1

200

160

320

05 1

800

170

320

05 0

000

170

320

05 0

600

170

320

05 1

200

170

320

05 1

800

180

320

05 0

000

Rich

ards

on n

umbe

r

Time (UTC)



Ri =119892

1205791

(1205791minus 1205792) (1199111minus 1199112)

(1199061minus 1199062)

2

(5)


2

































30405060708090

100

12 13 14 15 16 17 18 19 20

Hei

ght (

m)

OBSMYNNACM2

MYJYSU


30405060708090

100

230 235 240 245 250

Hei

ght (

m)

OBSMYNNACM2

MYJYSU


30405060708090

100

5 6 7 8 9 10

Hei

ght (

m)

OBSMYNNMYJ

ACM2YSU




0100

0203

04

05

06

07

08

09

095

099

10

2

2

3

3

44 1

1

250

225

200

175

150

125

100

075

050

025

000025 050 075 REF 125 150 175 200 225 250

Stan

dard

ized

dev

iatio

ns (n

orm

aliz

ed)

Correlation










PBLschemes

[8] EDFME(ms)
























000020040

Win

d sp

eed

bias

(ms

)

MYNNACM2

MYJYSU

minus100

minus080

minus060

minus040

minus020

(a)

MYNNACM2

MYJYSU

000050100150200250

AverageWin

d sp

eed

STD

(ms

)

30m 40m 50m 60m 70m 80m 90m 100m

(b)


55N548N546N544N542N54N

538N536N536N534N532N53N

35E 4E 45E 5E 55E 6E 7E65E 75E 85E8E 9E

78 81 84 87 9 93 9996 102 105 108 111










10

20

30

FINO1

West East

South

(a)

(d) (e)

(b) (c)

North

10

20

30

WRF-ACM2

West East

South

North

10

20

30

WRF-YSU

West East

South

North

10

20

30

WRF-MYJ

West East

South

North

10

20

30

West East

South

NorthWRF-MYNN

0ndash44ndash8

8ndash1212ndash16

16ndash2020ndash24





4 Conclusions



0 2 4 6 8 10 12 14 16 18 20 2202

4

68

10

12

14

16

18

20

22

Observation

WRF

H = 100m R = 0933 ratio data = 07

(a)

0 2 4 6

00010003

001002005010

025

050

075

090095098099

09970999

Prob

abili

ty



Error 100m

(b)

0 5 10 150

10

20

30

40

50

60

70


(c)







0 5 10 15 20 25 300

001002003004005006007008009

01

Prob

abili

ty


FINO1 (k = 251 120582 = 1274)


WRF-MYNN (k = 268 120582 = 1251)





References






































Volume 2014

Mining


Journal of



Geophysics







Journal of




Advances in











Geology Advances in












200210220230240250260

3579

11131517

0000 0600 1200 1800 0000 0600 1200 1800 0000

(deg

)

Time (UTC)

(ms

) (∘

C)


50m wspd (ms)

50m wdir (deg)


0

005

01

015

02

025

160

320

05 0

000

160

320

05 0

600

160

320

05 1

200

160

320

05 1

800

170

320

05 0

000

170

320

05 0

600

170

320

05 1

200

170

320

05 1

800

180

320

05 0

000

Rich

ards

on n

umbe

r

Time (UTC)



Ri =119892

1205791

(1205791minus 1205792) (1199111minus 1199112)

(1199061minus 1199062)

2

(5)


2

































30405060708090

100

12 13 14 15 16 17 18 19 20

Hei

ght (

m)

OBSMYNNACM2

MYJYSU


30405060708090

100

230 235 240 245 250

Hei

ght (

m)

OBSMYNNACM2

MYJYSU


30405060708090

100

5 6 7 8 9 10

Hei

ght (

m)

OBSMYNNMYJ

ACM2YSU




0100

0203

04

05

06

07

08

09

095

099

10

2

2

3

3

44 1

1

250

225

200

175

150

125

100

075

050

025

000025 050 075 REF 125 150 175 200 225 250

Stan

dard

ized

dev

iatio

ns (n

orm

aliz

ed)

Correlation










PBLschemes

[8] EDFME(ms)
























000020040

Win

d sp

eed

bias

(ms

)

MYNNACM2

MYJYSU

minus100

minus080

minus060

minus040

minus020

(a)

MYNNACM2

MYJYSU

000050100150200250

AverageWin

d sp

eed

STD

(ms

)

30m 40m 50m 60m 70m 80m 90m 100m

(b)


55N548N546N544N542N54N

538N536N536N534N532N53N

35E 4E 45E 5E 55E 6E 7E65E 75E 85E8E 9E

78 81 84 87 9 93 9996 102 105 108 111










10

20

30

FINO1

West East

South

(a)

(d) (e)

(b) (c)

North

10

20

30

WRF-ACM2

West East

South

North

10

20

30

WRF-YSU

West East

South

North

10

20

30

WRF-MYJ

West East

South

North

10

20

30

West East

South

NorthWRF-MYNN

0ndash44ndash8

8ndash1212ndash16

16ndash2020ndash24





4 Conclusions



0 2 4 6 8 10 12 14 16 18 20 2202

4

68

10

12

14

16

18

20

22

Observation

WRF

H = 100m R = 0933 ratio data = 07

(a)

0 2 4 6

00010003

001002005010

025

050

075

090095098099

09970999

Prob

abili

ty



Error 100m

(b)

0 5 10 150

10

20

30

40

50

60

70


(c)







0 5 10 15 20 25 300

001002003004005006007008009

01

Prob

abili

ty


FINO1 (k = 251 120582 = 1274)


WRF-MYNN (k = 268 120582 = 1251)





References






































Volume 2014

Mining


Journal of



Geophysics







Journal of




Advances in











Geology Advances in























30405060708090

100

12 13 14 15 16 17 18 19 20

Hei

ght (

m)

OBSMYNNACM2

MYJYSU


30405060708090

100

230 235 240 245 250

Hei

ght (

m)

OBSMYNNACM2

MYJYSU


30405060708090

100

5 6 7 8 9 10

Hei

ght (

m)

OBSMYNNMYJ

ACM2YSU




0100

0203

04

05

06

07

08

09

095

099

10

2

2

3

3

44 1

1

250

225

200

175

150

125

100

075

050

025

000025 050 075 REF 125 150 175 200 225 250

Stan

dard

ized

dev

iatio

ns (n

orm

aliz

ed)

Correlation










PBLschemes

[8] EDFME(ms)
























000020040

Win

d sp

eed

bias

(ms

)

MYNNACM2

MYJYSU

minus100

minus080

minus060

minus040

minus020

(a)

MYNNACM2

MYJYSU

000050100150200250

AverageWin

d sp

eed

STD

(ms

)

30m 40m 50m 60m 70m 80m 90m 100m

(b)


55N548N546N544N542N54N

538N536N536N534N532N53N

35E 4E 45E 5E 55E 6E 7E65E 75E 85E8E 9E

78 81 84 87 9 93 9996 102 105 108 111










10

20

30

FINO1

West East

South

(a)

(d) (e)

(b) (c)

North

10

20

30

WRF-ACM2

West East

South

North

10

20

30

WRF-YSU

West East

South

North

10

20

30

WRF-MYJ

West East

South

North

10

20

30

West East

South

NorthWRF-MYNN

0ndash44ndash8

8ndash1212ndash16

16ndash2020ndash24





4 Conclusions



0 2 4 6 8 10 12 14 16 18 20 2202

4

68

10

12

14

16

18

20

22

Observation

WRF

H = 100m R = 0933 ratio data = 07

(a)

0 2 4 6

00010003

001002005010

025

050

075

090095098099

09970999

Prob

abili

ty



Error 100m

(b)

0 5 10 150

10

20

30

40

50

60

70


(c)







0 5 10 15 20 25 300

001002003004005006007008009

01

Prob

abili

ty


FINO1 (k = 251 120582 = 1274)


WRF-MYNN (k = 268 120582 = 1251)





References






































Volume 2014

Mining


Journal of



Geophysics







Journal of




Advances in











Geology Advances in


0100

0203

04

05

06

07

08

09

095

099

10

2

2

3

3

44 1

1

250

225

200

175

150

125

100

075

050

025

000025 050 075 REF 125 150 175 200 225 250

Stan

dard

ized

dev

iatio

ns (n

orm

aliz

ed)

Correlation










PBLschemes

[8] EDFME(ms)
























000020040

Win

d sp

eed

bias

(ms

)

MYNNACM2

MYJYSU

minus100

minus080

minus060

minus040

minus020

(a)

MYNNACM2

MYJYSU

000050100150200250

AverageWin

d sp

eed

STD

(ms

)

30m 40m 50m 60m 70m 80m 90m 100m

(b)


55N548N546N544N542N54N

538N536N536N534N532N53N

35E 4E 45E 5E 55E 6E 7E65E 75E 85E8E 9E

78 81 84 87 9 93 9996 102 105 108 111










10

20

30

FINO1

West East

South

(a)

(d) (e)

(b) (c)

North

10

20

30

WRF-ACM2

West East

South

North

10

20

30

WRF-YSU

West East

South

North

10

20

30

WRF-MYJ

West East

South

North

10

20

30

West East

South

NorthWRF-MYNN

0ndash44ndash8

8ndash1212ndash16

16ndash2020ndash24





4 Conclusions



0 2 4 6 8 10 12 14 16 18 20 2202

4

68

10

12

14

16

18

20

22

Observation

WRF

H = 100m R = 0933 ratio data = 07

(a)

0 2 4 6

00010003

001002005010

025

050

075

090095098099

09970999

Prob

abili

ty



Error 100m

(b)

0 5 10 150

10

20

30

40

50

60

70


(c)







0 5 10 15 20 25 300

001002003004005006007008009

01

Prob

abili

ty


FINO1 (k = 251 120582 = 1274)


WRF-MYNN (k = 268 120582 = 1251)





References






































Volume 2014

Mining


Journal of



Geophysics







Journal of




Advances in











Geology Advances in



















000020040

Win

d sp

eed

bias

(ms

)

MYNNACM2

MYJYSU

minus100

minus080

minus060

minus040

minus020

(a)

MYNNACM2

MYJYSU

000050100150200250

AverageWin

d sp

eed

STD

(ms

)

30m 40m 50m 60m 70m 80m 90m 100m

(b)


55N548N546N544N542N54N

538N536N536N534N532N53N

35E 4E 45E 5E 55E 6E 7E65E 75E 85E8E 9E

78 81 84 87 9 93 9996 102 105 108 111










10

20

30

FINO1

West East

South

(a)

(d) (e)

(b) (c)

North

10

20

30

WRF-ACM2

West East

South

North

10

20

30

WRF-YSU

West East

South

North

10

20

30

WRF-MYJ

West East

South

North

10

20

30

West East

South

NorthWRF-MYNN

0ndash44ndash8

8ndash1212ndash16

16ndash2020ndash24





4 Conclusions



0 2 4 6 8 10 12 14 16 18 20 2202

4

68

10

12

14

16

18

20

22

Observation

WRF

H = 100m R = 0933 ratio data = 07

(a)

0 2 4 6

00010003

001002005010

025

050

075

090095098099

09970999

Prob

abili

ty



Error 100m

(b)

0 5 10 150

10

20

30

40

50

60

70


(c)







0 5 10 15 20 25 300

001002003004005006007008009

01

Prob

abili

ty


FINO1 (k = 251 120582 = 1274)


WRF-MYNN (k = 268 120582 = 1251)





References






































Volume 2014

Mining


Journal of



Geophysics







Journal of




Advances in











Geology Advances in


10

20

30

FINO1

West East

South

(a)

(d) (e)

(b) (c)

North

10

20

30

WRF-ACM2

West East

South

North

10

20

30

WRF-YSU

West East

South

North

10

20

30

WRF-MYJ

West East

South

North

10

20

30

West East

South

NorthWRF-MYNN

0ndash44ndash8

8ndash1212ndash16

16ndash2020ndash24





4 Conclusions



0 2 4 6 8 10 12 14 16 18 20 2202

4

68

10

12

14

16

18

20

22

Observation

WRF

H = 100m R = 0933 ratio data = 07

(a)

0 2 4 6

00010003

001002005010

025

050

075

090095098099

09970999

Prob

abili

ty



Error 100m

(b)

0 5 10 150

10

20

30

40

50

60

70


(c)







0 5 10 15 20 25 300

001002003004005006007008009

01

Prob

abili

ty


FINO1 (k = 251 120582 = 1274)


WRF-MYNN (k = 268 120582 = 1251)





References






































Volume 2014

Mining


Journal of



Geophysics







Journal of




Advances in











Geology Advances in


0 2 4 6 8 10 12 14 16 18 20 2202

4

68

10

12

14

16

18

20

22

Observation

WRF

H = 100m R = 0933 ratio data = 07

(a)

0 2 4 6

00010003

001002005010

025

050

075

090095098099

09970999

Prob

abili

ty



Error 100m

(b)

0 5 10 150

10

20

30

40

50

60

70


(c)







0 5 10 15 20 25 300

001002003004005006007008009

01

Prob

abili

ty


FINO1 (k = 251 120582 = 1274)


WRF-MYNN (k = 268 120582 = 1251)





References






































Volume 2014

Mining


Journal of



Geophysics







Journal of




Advances in











Geology Advances in


0 5 10 15 20 25 300

001002003004005006007008009

01

Prob

abili

ty


FINO1 (k = 251 120582 = 1274)


WRF-MYNN (k = 268 120582 = 1251)





References






































Volume 2014

Mining


Journal of



Geophysics







Journal of




Advances in











Geology Advances in















Volume 2014

Mining


Journal of



Geophysics







Journal of




Advances in











Geology Advances in

research article wrf model methodology for offshore wind...

Documents