R

Af

Sa

b

a

ARRAA

KWMBWWHEA

C

1

s

0d

Applied Ocean Research 34 (2012) 1– 9

Contents lists available at SciVerse ScienceDirect

Applied Ocean Research

journal homepage: www.elsevier.com/locate/apor

eview

ssessment of wind models around the Balearic Islands for operational waveorecast

. Ponce de Leóna,∗, A. Orfilaa, L. Gómez-Pujolb, L. Renaultb, G. Vizosoa, J. Tintoréa,b

IMEDEA (CSIC-UIB), Miquel Marqués 21, 07190, Esporles, Balearic Islands, SpainICTS SOCIB, Balearic Islands Coastal Observing and Forecasting System Parc Bit, Naorte, Bloc A P3, 07121, Palma de Mallorca, Balearic Islands, Spain

r t i c l e i n f o

rticle history:eceived 19 October 2010eceived in revised form 25 August 2011ccepted 6 September 2011vailable online 7 October 2011

eywords:ind waves

a b s t r a c t

A wave hindcast in the Western Mediterranean Sea is carried out in order to assess the performance ofthree atmospheric models in providing the forcing for a third generation wave model. The wind modelshave been used as forcing fields for the generation of waves and the resulting significant wave heighttime history compared with four buoys around the Balearic Islands. Two different wave-model gridresolutions are used to get the wave field in the entire Mediterranean and around the Balearic Islands.The present application was performed for three months: November 2008 and for July and August 2009.Results indicate that all data sources provide good forcing for operational wave forecast at large scales

editerranean Seaalearic IslandsAMRF

IRLAMCMWF

(wind forecast with grid resolution of 30 and 25 km). Near the coast or at the lee of islands, resolvingsmall scale topographical features result in a better forecast of wave fields. However, for the area studied,the atmospheric model that better represents summer and winter conditions is hourly WRF at 1.5 kmresolution.

© 2011 Elsevier Ltd. All rights reserved.

SCAT

ontents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12. Study area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23. Data and methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

3.1. Atmospheric models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23.2. Wave model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33.3. Buoy data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

4. Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44.1. Wind forcing experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44.2. Performance of WAMPRO1 in providing boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44.3. Performance of WAMPRO2 vs. WAMPRO1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64.4. Influence of wind spatial resolution on the nested grid (WAMPRO2) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64.5. Spatial variability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

5. Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

. Introduction

A proper assessment of wave climate is a requirement forcientific and engineering activities in the coastal zone. Beach

∗ Corresponding author. Tel.: +34 659027032; fax: +34 971611761.E-mail address: sponce@imedea.uib-csic.es (S. Ponce de León).

141-1187/$ – see front matter © 2011 Elsevier Ltd. All rights reserved.oi:10.1016/j.apor.2011.09.001

nourishment, port design and operability, dispersion and diffusionof pollutants are some examples that require a proper estima-tion of significant wave heights fields (diagnostic) as well as theirevolution (prognostic) forward in time. The diagnostic of waves

has traditionally been obtained using scalar and directional wavebuoys moored at specific locations. Moored instruments are themost reliable systems used to obtain wave conditions but they areexpensive providing only records at specific locations. In the last

2 ed Oce

dhds

ciagnpt

atotowtmabw

esttqm[tEAtd[usfi

otbfatogor

gfiadi2omnp

ga

S. Ponce de León et al. / Appli

ecade, satellites and more recently High Frequency Radar systemsave overcome to some degree the problem of the spatial lack ofata. However these platforms only provide information at limitedpatial and temporal coverage.

By contrast, numerical models provide on a regular basis, waveonditions at different spatial and temporal resolutions by integrat-ng physical principles forward in time. Wave generation modelsre able to reproduce complex physical processes involved in theeneration and transformation of waves [1,2]. To obtain reliableumerical simulations accurate wind fields with the adequate tem-oral and spatial resolution are mandatory since realistic forcingerms will provide accurate wave predictions.

Despite the diversity of wave generation models as well as intmospheric models, numerical predictions of wave fields in a par-icular region still fail mainly due to an inadequate representationf the physical processes involved in the wave generation or dueo errors associated to the spatial and/or temporal wind field res-lutions. Wave models are very sensitive to wind field variations,hich results in one of the main source of errors in wave predic-

ions [3–5]. Additionally, in small and semi-enclosed seas, waveodelling becomes cumbersome due to the complex bathymetry

nd the limited fetch. Surface waves are generated by the windlowing over the sea surface and any error in the input wind fieldill be reflected in the computation of wave conditions [3].

For instance, wave climate over the Balearic Sea has in gen-ral, a complex pattern as a result of the complex orography of theurrounding area. In the last years, some studies have attemptedo analyse the accuracy of different wave and wind models inhe NW Mediterranean Sea. Signell [6] analysed the surface winduality in the Adriatic Sea concluding that for a period of twoonths the limited area models LAMI-Limited Area Model Italy

7] and COAMPS-Coupled Ocean/Atmosphere Mesoscale Predic-ion System [8] provide better amplitude response than the coarserCMWF (European Center for Medium-range Weather Forecasts).rdhuin [9] analysed the accuracy of four atmospheric models and

hree wind-wave models concluding that quality of the wind fieldsegrades in the coastal areas. Ponce de León and Guedes Soares10], compared wave hindcast in the Western Mediterranean Seasing the reanalysis wind fields from HIPOCAS and ERA-40 findingystematic negative biases of significant wave height using ERA-40elds and positive biases for the HIPOCAS data.

In addition, there are several works pointing out the necessityf further improvement on the wind field quality as well as inhe increase of wind field spatial resolution especially in enclosedasins such as the Mediterranean Sea. Cavaleri and Bertotti [11]ound that large errors in wave height estimation were obtainedt short fetches (∼100 km); Bertotti and Cavaleri [12] pointed thathe reliability of forecasts may decrease when dealing with mete-rological situations characterized by strong temporal and spatialradients. In spite of these advances, to our knowledge, the effectsf local meteorological events around archipelagos in the Mediter-anean Sea have not been treated in detail.

The aim of this work is to further study the accuracy of a thirdeneration wave model forced by three different atmospheric windelds with different spatial resolutions (30, 25, 16, 6, 5 and 1.5 km)round the Balearic Islands (NW Mediterranean Sea) prior to theevelopment of a real time operational system for wave prediction

n the area. The present application was performed for November008 and July and August 2009 as these months are representativef the different wave climates of the area during winter and sum-er seasons. Moreover, during the summer the local sea breeze

eeds to be properly reproduced since it can drive an important

art of wave variability.

The paper is structured as follows: Section 2 presents the sin-ularities of the study area. Section 3 provides a description of thetmospheric models, in situ measurements as well as wave model

an Research 34 (2012) 1– 9

set-up. Section 4 presents the results and provides the discussionand finally Section 5 concludes the work.

2. Study area

The Western Mediterranean (WM) has a complex structure withnumerous peninsulas and islands that difficult numerical mod-elling. Moreover, the WM is an important cyclogenetic area wherethe main hydrodynamics is conditioned by the severe atmospheric-climatic forcing during winters [13]. Most of the strong windsobserved in the Mediterranean belong to the category of local windsand are originated as down slope flows by air-flow/mountain inter-action or due to channelling effects [14]. The WM area is forced bynortherly and north-westerly winds during most of the year, whileless intense cyclogenetic activity is observed during the rest of theyear [15,16]. The mountains range in the vicinity is a key factor in itsclimatic characteristics [17]. The role of the Pyrenees in the westernarea and the Alps in the north-eastern area are decisive boundariesfor the pressure and wind distribution over the WM basin. Windspeeds for a 100-year return period shows a maximum located inthe Gulf of Lion with winds up to 30 m/s [16]. The low-pressuresystems entering to the Mediterranean Sea from the Atlantic Oceantend to dissipate moving east, with the major storms taking placein the WM [10].

The Balearic Archipelago, near the eastern Mediterranean coastof Spain is formed by four major islands that may lead to a reductionof the wave energy in the basin during winter period due to theshadowing effect of the islands [18].

The present work has been centered in the southern coast ofMallorca Island for operational purposes where a high resolutionwave model (1500 m) has been implemented and nested to thegeneral wave model covering the entire Mediterranean. Resultingwave fields are compared with in situ data from deep and shallowwater wave buoys.

3. Data and methods

3.1. Atmospheric models

Three different sources of atmospheric models are used to testthe accuracy of wave fields within this work: HIRLAM-High Res-olution Limited Area Model [19], ECMWF [20] and WRF-WeatherResearch and Forecasting [21]. HIRLAM was chosen because it isthe atmospheric model that runs daily four times in the SpanishMeteorological Agency (AEMET); the WRF because it presents thehighest spatial resolution and the ECMWF because it is providedoperationally by the European Center for Medium-range WeatherForecasts.

To assess the sensitivity of wave fields to the wind, datasets have been divided into two groups according to their spa-tial resolution. The first group consists of atmospheric modelswith a relatively coarse resolution HIRLAM(16), WRF(30) andECMWF(25) with 16, 30 and 25 km resolution, respectively. Thesecond group includes the atmospheric models with higher res-olution HIRLAM(5) with 5 km and two WRF(6/1.5) configurationswith 6 km and 1.5 km.

The HIRLAM is a hydrostatic, primitive-equation model, whichuses a three dimensional variational (3D-VAR) data assimilationscheme [22]. In the WM, HIRLAM is operated by AEMET provid-ing wind fields every 3 h at 16 km (low resolution) and 5 km (highresolution) twice a day cycle with a 72 h horizon [23]. In this config-

uration, HIRLAM takes the boundary conditions from the ECMWFatmospheric global model.

The ECMWF is a spectral model which incorporates a fourdimensional variational (4D-VAR) data assimilation procedure [24].

ed Ocean Research 34 (2012) 1– 9 3

F2at

hbtwD(wBucb(Cfi3

3

sauaWTW

tbdt

wdtt

S

wiS(o

qftfqa

tta

(vBb

Fig. 1. The coarse (top) and nested (bottom) grids geographical domain, the area of

S. Ponce de León et al. / Appli

orecasts were provided every 6 h with a horizontal resolution of5 km until the end of January 2010. The vertical structure of thetmosphere is solved by a multi-level hybrid sigma coordinate sys-em [20].

The WRF [21,25] model is a next-generation mesoscale non-ydrostatic numerical weather prediction system designed to serveoth operational forecasting and atmospheric research needs. Forhis work, the simulation includes three two-way nested domainsith horizontal grid spacing (�x ∼ �y) of 30 km, 6 km and 1.5 km.omain 1 (coarser mesh) covers Western Europe and North Africa

32.6◦N–44.8◦N, 5◦W–8.6◦E) whereas the nested domains cover theestern Mediterranean Sea (36.5◦N–42.2◦N, 0.6◦W–5.9◦E) and thealearic sea (38.6◦N–40.3◦N, 1.6◦W–4.1◦E). A total of 47 levels weresed in the vertical, half of them in the lowest 1.5 km. The modeloarse grid was spun-up and interpolated (using time-dependentoundary conditions) to the boundary of domain 1 from the NCEPNational Center for Environmental Prediction) – NCAR (Nationalenter for Atmospheric Research) reanalysis every 6 h. This con-guration provides forecasts with a temporal resolution of 1 and

h.

.2. Wave model

The WAM model [2,3] is a state-of-the-art third generationpectral wave model specifically designed for global and shelf seapplications. The third generation wave WAM model is nowadayssed by several institutions world wide in research activities as wells for operational applications. Some modifications of the originalAM cycle 4 code were developed to be applied in coastal areas.

his version of the WAM-model will hereafter be referred to as theAMPRO model [26].The WAMPRO latest version includes additional source terms

o deal with the shallow water physics; a depth-induced wavereaking and six new bottom friction formulations. The modelescribes the evolution of the directional wave spectrum F, throughhe energy balance equation [3],

∂F(f, �, �, �, t)∂t

+ ∇ · (CgF) = Stot(f, �, �, �, t)

here f is the wave frequency, � is the mean wave propagationirection, � and � are the latitude and longitude, respectively, Cg ishe group velocity and Stot is the source function. In general, Stot ishe sum of the three source functions,

tot = Sin + Snl + Sds

here Sin represents the wind input, Snl the non-linear wave-waventeractions and Sds the wave energy dissipation. For deep watersds is the dissipation induced by the wave breaking in deep waterswhitecapping) while in shallow waters the dissipation is the sumf the effects of the bottom friction and the wave breaking [3].

The wind input is a further development based on Janssen’suasi-linear theory of wind-wave generation [27,28]. The bottomriction dissipation term in WAMPRO is represented accordingo the JONSWAP formulation [29] and extended to the newormulations [26]. The nonlinear resonant interactions betweenuadruplets of wave components are calculated through anpproximation to the exact expression [30].

Despite that WAMPRO includes refraction due to the bottom andhe presence of currents, diffraction effects are missed. In addition,he triads non-linear interactions in shallow water regions are notccounted in the WAMPRO code.

A coarse mesh covering the whole Mediterranean Sea

30◦N–46◦N and 6◦W–37◦E) was implemented in order to pro-ide the boundary conditions – hereinafter WAMPRO1 (Fig. 1, top).athymetry was obtained from the ETOPO1, which is a 1 min oceanathymetry from a relief model of NOAA-National Oceanic and

study and locations of the buoys.

Atmospheric Administration. Bathymetry was linearly interpolatedto the final resolution of 0.25◦ in a 173 × 65 mesh. A high resolutionmesh covering the area of 38.75◦N, to 39.75◦N and 1.75◦E–4.0◦Ewas nested to the former – hereinafter WAMPRO2 (Fig. 1, bot-tom). Depth was obtained from nautical charts of the SpanishHydrographical Institute. This domain includes the southern partof Mallorca Island as well as the Cabrera Archipelago. Bathymetrywas interpolated to the final mesh of 181 × 81 nodes with a resolu-tion of 1/80◦. The WAMPRO1 mesh was arranged to match with theresolution provided by ECMWF(25) model in the whole Mediter-ranean basin while the WAMPRO2 mesh arranged to coincide withthe high resolution WRF(1.5) model in the Balearic Sea. Data fromHIRLAM(16/5) were internally interpolated to both WAMPRO1 andWAMPRO2 meshes. Both, integration time step and propagationtime step were set as 600 and 40 s for WAMPRO1 and WAMPRO2,respectively. The energy balance equation was integrated for 24directions and 25 frequencies. The lowest resolved frequency is0.0418 Hz.

3.3. Buoy data

Numerical simulations are validated at four locations withmeasurements from moored buoys. The first location is westernDragonera Island at the western part of WAMPRO2 domain (B1 inFig. 1). This mooring consists of a directional wave buoy from theSpanish Harbour Authority (Puertos del Estado, hereinafter PE) in135 m water depth. The second location, Capdepera is a scalar shal-low water buoy in the north-eastern side of the numerical domain,4 km offshore in 45 m water depth (B2 in Fig. 1). The third loca-tion, Cabrera is a scalar buoy operated from IMEDEA moored in70 m depth in the channel between Mallorca and Cabrera islands(B3 in Fig. 1). Additionally, a deep-water buoy located in frontof Maó (Menorca Island) from PE (B4 in Fig. 1) has been used tovalidate numerical simulations for the coarse model implementa-

tion. The geographical coordinates of the wave buoys are given inTable 1.

4 S. Ponce de León et al. / Applied Oce

Table 1Geographical location of buoys and depths.

Dragonera (B1) Cap de Pera (B2) Cabrera (B3) Maó (B4)

Latitude (◦N) 39.55 39.75 39.23 39.72Longitude (◦E) 2.10 3.50 2.96 4.42Depth (m) 135 45 70 300

Table 2Experiments performed for November 2008.

Experiment WAMPRO1(dx = dy = 25 km)

WAMPRO2(dx = dy = 1.5 km)

Wind input time step (h)

1 HIRLAM(16 km)

HIRLAM(16 km)

3

2 HIRLAM(16 km)

HIRLAM (5 km) 3

3 ECMWF(25 km)

ECMWF(25 km)

6

4 ECMWF(25 km)

WRF (1.5 km) 6

5 ECMWF(25 km)

WRF (6 km) 6

6 ECMWF(25 km)

WRF (30 km) 6

4

4

wfif

FN

7 ECMWF(25 km)

WRF (1.5 km) 1

. Results

.1. Wind forcing experiments

Different experiments were carried out in order to compareind input models according to Table 2. The low resolution windelds – HIRLAM(16) and ECMWF(25) – were used as the forcing

or WAMPRO1 which in turn provided the boundary conditions for

ig. 2. Scatter plots for the HIRLAM(16) and ECMWF(25) wind speeds at B4 (top panels)

ovember 2008. Number of observations: 119.

an Research 34 (2012) 1– 9

WAMPRO2. Moreover, WAMPRO2 was forced with all wind datasets.

During November 2008, four isolated storms characterized bywind speeds over 15 m/s crossed the area. These storms pro-duced significant wave heights (Hs) at B1 over 5.5 m (recorded atNovember 29th at 15 UTC). These storms are used to first study theaccuracy of the wave model against the different wind fields.

Besides, during summer month (e.g. data corresponding to Julyand August 2009) the main characteristic around Mallorca is thegeneration of sea breeze that has some effect in the generation andgrowth of waves.

4.2. Performance of WAMPRO1 in providing boundary conditions

In this section we are going to present first, the performance ofthe WAMPRO1 configuration in providing the boundary conditionsfor the nested grid. Prior to evaluating the accuracy of differentwind models in wave field, both wave resolutions (WAMPRO1and WAMPRO2) are compared at specific locations. The windinput at B4 for both HIRLAM(16) and ECMWF(25) are comparedwith the wind measurements through scatter plots (Fig. 2 toppanels). The measured wind speeds have been adjusted to 10 musing the same neutral logarithmic wind profile as described inBidlot [31,32], with a Charnock constant parameter of 0.018. Cor-relation coefficients between u10 measured at B4 and from themodel are 0.82 and 0.65 for HIRLAM(16) and ECMWF(25), respec-tively.

Wave hindcast was done for WAMPRO1 domain for November2008 with the low resolution winds HIRLAM(16) and ECMWF(25).

The resulting time history for the significant wave height wascompared with recorded data at B4 buoy. Even though some dis-crepancies in wind speed for both wind data sets, they are notespecially relevant for the prediction of the significant wave height

and scatter plots for the Hs from WAMPRO1, respectively (bottom panels). Period:

S. Ponce de León et al. / Applied Ocean Research 34 (2012) 1– 9 5

Table 3Bias, the best-fit scatter index and slopes between measured and modelled U10 and Hs at B4 from the WAMPRO1 for November 2008. The number of observations for theHIRLAM(16) is 119 and for ECMWF(25) is 119.

U10 (m/s) U10 (m/s) Hs (m) Hs (m)Models and parameters HIRLAM(16) ECMWF(25) WAMPRO1 & HIRLAM(16) WAMPRO1 & ECMWF(25)

Bias −0.57 0.20 0.04 0.08SI 0.31 0.41 0.19 0.20Slope 0.98 0.90 0.95 0.95

Table 4Bias and the best-fit line scatter index between measured Hs (m) at B1, B2, B3 and modelled with WAMPRO1 and WAMPRO2 for November 2008. The mean buoys Hs valuesare 1.53 m for B1, 1.45 m for B2 and 0.71 m for B3. The number of observations for the HIRLAM(16) is 119 and for ECMWF(25) is 119.

Location B1 B2 B3

Parameters Bias S.I. Bias S.I. Bias S.I.

HIRLAM(16)(WAMPRO1) 0.18 0.29 −0.15 0.43 −0.30 0.75HIRLAM(16)(WAMPRO2) 0.17 0.30 0.064 0.33 −0.05 0.55ECMWF(25)(WAMPRO1) 0.26 0.18 −0.08 0.30 −0.39 0.63

(sEftm

FB

ECMWF(25)(WAMPRO2) 0.26 0.18

Fig. 2 bottom panels). The correlation coefficients between mea-ured and modelled Hs at B4 are 0.92 and 0.94 for HIRLAM(16) and

CMWF(25), respectively. As seen, both winds provide good resultsor the predicted Hs. The bias (defined as the difference betweenhe mean observation and the mean prediction) between B4 and

odel results is presented in Table 3. Both wind models yield low

ig. 3. Scatter plots of the results obtained using the low resolution winds: WRF(30)(top p3 (right). Period of simulation: November 2008. Number of observations: 119.

0.13 0.28 −0.12 0.48

biases for the prediction of Hs at large scale. The scatter index (SI)defined as the standard deviation of the predicted data with respect

to the best-fit line, divided by the mean observations, is also of thesame order. The low values of the biases at B4 indicate that small-scale wind fields are not required for the wave field prediction atthis part of the Archipelago. Despite that the SI for the atmospheric

anels), HIRLAM(16)(middle) and ECMWF(25)(bottom) at B1 (left), B2 (middle) and

6 S. Ponce de León et al. / Applied Ocean Research 34 (2012) 1– 9

F .5), HN

mfi

4

iatwI

BgetBbwt

4(

NH

ig. 4. Scatter plots of the results obtained using the high resolution winds: WRF(1ovember 2008. Number of observations: 119.

odels is different the resulting wave field is similar for both windelds (Table 3).

.3. Performance of WAMPRO2 vs. WAMPRO1

To solve adequately the effects of the islands, a fine mesh resolv-ng small geographical indents has to be implemented although thispproach is unaffordable when forecasting large areas in an opera-ional way. To overcome this problem we have nested WAMPRO2hich has 1.5 km grid resolution in the southern part of Mallorca

sland in WAMPRO1.The biases and the scatter indexes between measured data at

1, B2 and B3 and WAMPRO1 (coarse) and WAMPRO2 (nestedrid) using ECMWF(25) and HIRLAM(16) are given in Table 4. Asxpected, increasing the grid resolution results in general in a bet-er adjustment of Hs at locations B2 and B3. The improvement at3 is explained due to the fact that Cabrera Archipelago is slightlyetter resolved by increasing the resolution in WAMPRO2. At B1,hich is not influenced by the coast, no evident differences from

he two mesh resolutions were found.

.4. Influence of wind spatial resolution on the nested gridWAMPRO2)

To assess the performance of the wind models a hindcast forovember 2008 was done with all model inputs. HIRLAM(16) andIRLAM(5) were nested with HIRLAM(16) in WAMPRO1 while

IRLAM(5) and WRF(6) at B1 (left), B2 (middle) and B3 (right). Period of simulation:

ECMWF(25) and WRF(1.5, 6, 30) were nested with ECMWF(25) (seeTable 2 for details of the different spatial and temporal resolutions).

The scatter plots between Hs hindcasted in WAMPRO2 and mea-sured data at B1 (left), B2 (center) and B3 (right) for the WRF(30)(top panels), HIRLAM(16) (middle) and ECMWF(25) (bottom) areshown in Fig. 3. At B2, which is oriented to the dominant fetchbetter statistics in terms of slope, are obtained with the three mod-els. However, in terms of the scatter of the data, better results areobtained at B1, which as already stated is not affected by topo-graphical effects. At B2 results are worse both in terms of slope andscatter of the data. At B1 and B2, WAMPRO1 underestimates the sig-nificant wave height using the three atmospheric models whereasat B3 WAMPRO1 overestimate measured Hs. The three analysedmodels provide very similar predictions at this spatial resolution.The correlation coefficients between measured and modelled Hsare over 0.89, 0.90 and 0.84 at B1, B2 and B3, respectively, withall three atmospheric models analysed. At this point we want toremark that the temporal resolution for the analysed data from theECMWF(25) and WRF(30) is 6 h in front of the 3 h of HIRLAM(16).

For the high-resolution wind models the scatter plots are dis-played in Fig. 4. The hindcast of Hs at the three locations presenta similar behaviour than the coarse resolution wind models. Thebest fit is obtained at B2 with slope of 0.94 for WRF(1.5), 0.95 for

(HIRLAM5) and 0.93 for WRF(6) but with smaller scatter index atB1. The larger scatter indexes obtained at B3 indicate that the wavemodel does not properly resolve the small-scale effects betweenislands (less than 10 km). The three models display statistically

S. Ponce de León et al. / Applied Ocean Research 34 (2012) 1– 9 7

Table 5Averaged normalized deviations between modelled Hs forced with WRF(30),HIRLAM(16) and ECMWF(25) and B1, B2 and B3 buoys (top) and with WRF(1.5),HIRLAM(5) and WRF(6) and B1, B2 and B3 buoys (bottom) for November 2008.

B1 B2 B3

Coarse resolution wind fieldsWRF(30) 0.124 0.075 −0.154HIRLAM(16) 0.085 0.052 −0.038ECMWF(25) 0.184 0.104 −0.146

Fine resolution wind fieldsWRF(1.5) 0.086 0.008 −0.355

sta

eNofiWrm

ibdcpeptIttb

glwBEtmirsttfW

4

fratsfitfiW

Fig. 5. Scatter plots at B1 for July and August 2009. Top panel – WAMPRO2 computedHs (m) versus buoy data using the high temporal (hourly) and spatial resolution(1.5 km) WRF winds. Bottom panel – simulations performed using the ECMWF(25)winds. Number of observations: 248.

HIRLAM(5) 0.049 0.072 −0.074WRF(6) 0.122 0.032 −0.294

imilar behaviour but for the data analysed the temporal resolu-ion of wind input was 1, 3 and 6 h for the WRF(1.5), HIRLAM(5)nd WRF(6), respectively.

Table 5 summarizes the normalized deviations between mod-lled and measured Hs at B1, B2 and B3 buoys averaged forovember 2008. For the coarse wind data, the best agreement isbtained with the HIRLAM (16). Although the best correlation coef-cient between measured and modelled Hs was obtained fromRF(1.5) with a value of 0.93, 0.91 and 0.87 at B1, B2 and B3,

espectively, no significant differences are obtained with the threeodels.In the Balearic Sea and especially around Mallorca Island, an

mportant source of variability in the wind is induced by the seareeze. During summer, sea breeze blows inland (outland) due theifferential warming of land (sea) with a 12 h period. The impli-ations of the use of wrong wind field information (omission oroor representation of breezes) for the wave forecast in the south-rn coast of Mallorca, will lead to inaccurate forecast of wavearameters mainly during summer months. To properly reproducehis process, high frequency (less than 3 h) winds are required.n fact, sea breeze circulation systems have been recently inves-igated by Papanastasiou [33] who applying the WRF model inhe coastal zone of Greece found that it is able to reproduce thereeze.

In order to study the ability of WRF(1.5) to reproduce the waveenerate by the local sea breeze in the southern coast of Mal-orca, a hindcast covering July and August of 2009 was performed

ith the high temporal resolution (hourly) data in (WAMPRO2).oundary conditions were obtained from WAMPRO1 forced withCMWF(25) interpolated to 1 h. These results are compared withhe hindcast obtained using the coarser ECMWF(25) atmospheric

odel. The comparison is presented for the significant wave heightn Fig. 5 at B1. For consistency between the two data set, theesults for WRF(1.5 km) are presented every 6 h. The best fit-linelopes and scatter index show that the WRF(1.5 km) provides bet-er results in the predicted Hs (slope 0.9) than the obtained withhe ECMWF(25). The value for the SI is not comparable since thator every prediction using ECMWF(25) we get 6 prediction for

RF(1.5).

.5. Spatial variability

The spatial variability of wind fields is a key factor especiallyor the small-scale features that can only be resolved by the highesolution models. As an example, Fig. 6 shows the wind speednd direction for WRF(1.5) (top panel) and for ECMWF(25) (bot-om panel) for November 14th 2008 at 12 UTC. As seen, for theame event (Tramuntana-intense winds from the north), both wind

elds largely differ not only in the small scale features, but inhe large scale which leads to significant difference in the waveeld. The wind fields (left) and the forecast wave fields (right) forRF(1.5)(top) and for ECMWF(25)(bottom) are displayed in Fig. 7

Fig. 6. Wind fields for 14th of November 2008 at 12 UTC according high (top) andlow (bottom) wind resolutions: WRF(1.5) and ECMWF(15), respectively.

8 S. Ponce de León et al. / Applied Ocean Research 34 (2012) 1– 9

Fig. 7. WRF(1.5) wind field (left) and the corresponding Hs (m) (right) from WAMPRO2

panels).

Fm

fi

ao[isatw

5

hcfI

models WRF(1.5) shows a good performance in predicting the sum-

ig. 8. WRF(1.5) wind field for 29th of November 2008 at 09 UTC and ASCAT(12.5)easured data from the orbit at 08:24 UTC (white flags).

or November 4th at 12 UTC. As seen, small scale pattern are presentn the WRF(1.5).

To show the importance of the small scale structure of windround the Archipelago, Fig. 8 shows a snapshot of November 29thf WRF(1.5) and winds from ASCAT – Advanced SCATterometer34,35]. Despite the orbit of the EUMETSAT’s (European Organ-sation for the Exploitation of Meteorological Satellites) MetOpatellite Scatterometer is at 8:24 a.m. and not exactly at 9:00 UTCs on the case of the map of WRF(1.5), it can be seen a similar pat-ern between the ASCAT wind speeds and directions and WRF(1.5)ind.

. Concluding remarks

The impact of different spatial resolution atmospheric modelsas been evaluated using a third generation wave model. The hind-

ast has been carried out nesting a coarse grid (�x = �y = 0.25◦)or the whole Mediterranean Sea with a finer mesh for the Balearicslands (�x ∼ �y = 1.5 km). Wave fields obtained using the different

for November 4th 2008 at 12 UTC (top panels). The same for ECMWF(25) (bottom

input forcings are validated against four different wave moorings(deep and shallow waters) around the coast.

For the coarse wind inputs the correlation between measuredand forecast Hs during winter conditions are similar (0.90) forWRF(30), HIRLAM(16) and ECMWF(25) at locations B1 and B2 beingthe modelled Hs underestimated. At B3, correlation between modeland buoy data falls to 0.87 for ECMWF(25) and to 0.84 and 0.88 forWRF(30) and HIRLAM(16), respectively.

For the fine resolution winds, correlation between modelledand measured Hs at B1 and B2 are over 0.90 with the threeatmospheric models being the results slightly better at B2 withWRF(1.5) (correlation of 0.93). Similarly to the coarse winds, pre-dicted Hs are underestimated at these locations. For both, coarseand fine resolution set of wind models, the predicted significantwave heights at B3 is overestimated being the correlation coef-ficient between predicted and measured Hs significantly lower(0.67) than in the other two locations. This indicates that theresolution implemented in WAMPRO2 (1.5 km) is not the appro-priate to solve properly the effects of the small islands of CabreraArchipelago.

Results indicate that the large scale atmospheric models ana-lysed provide similar results in terms of forecast wave heightand that they are accurate enough to provide wave forecast inopen seas being suitable for large scale operational wave mod-els at time interval of 6 h. On the other hand, these models aretoo coarse to reproduce the complex wave field produced aroundthe islands. Near the coast or at the lee of Islands, resolving smallscale topographical features will result in a better forecast of windfields. Our results show that high resolution wind models pro-vide a good spatial representation of waves around Islands (notcharacterized by large fetch). In insular systems characterized bya complex topography as the case studied, the grid size has to bereduced to forecast properly the whole wave spectra and to solveaccurately the shallow water processes. From our results we con-clude that all high resolution models analysed HIRLAM(5), WRF(6)and WRF(1.5) provide similar results in terms of slope and scatterindex when compared with the coastal buoys. Among the above

mer variability of Hs induced by the local sea breeze phenomena.WRF(1.5) has the appropriate spatial and temporal resolution tobe applied in small islands such as in Balearic. These results are in

ed Oce

aHo

aanfi

A

Habbfftf

R

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[

[[

[

S. Ponce de León et al. / Appli

ccordance with a recent analysis performed in the leeward side ofawaii (Pacific Ocean conditions) were it is concluded that the usef hourly WRF(6) is sufficient for modelling the wave pattern [36].

The best forecasts are obtained in locations B1 and B2 (deepnd shallow waters, respectively). At B3 low values for the slopesnd high SI between modelled and observed Hs pointed out theecessity of increasing the grid resolution to solve better the waveeld in a complex configuration of the Cabrera Archipelago.

cknowledgments

The authors would like to thank to AEMET for providing theIRLAM data and to ECMWF for the wind field. We gratefullycknowledge Puertos del Estado for supplying the Dragonera waveuoy data. We are grateful to Dr. Heinz Günther from GKSS, Ham-urg for his comments to this paper and two anonymous refereesor comments on the manuscript. S. Ponce de León is supported byunding from MICINN-JDC (Spanish Ministry of Science and Innova-ion). A. Orfila and L. Gomez-Pujol acknowledge financial supportrom MICINN through Project CTM2010-16915/MAR.

eferences

[1] WAMDI groupHasselmann S, Hasselmann K, Bauer E, Janssen PAEM, Komen GJ,et al. The WAM model – a third generation ocean wave prediction model. J PhysOceanogr 1988;18:1775–810.

[2] The WISE GroupCavaleri L, Alves JHGM, Ardhuin F, Babanin A, Banner M, et al.Wave modelling. The state of the art. J Prog Oceanogr 2007;75:603–74.

[3] Komen GJ, Cavaleri L, Donelan M, Hasselmann K, Hasselmann S, Janssen PAEM.Dynamics and modelling of ocean waves. Cambridge University Press; 1994.

[4] Ponce de León S, Ocampo-Torres FJ. Sensitivity of a wave model to wind vari-ability. J Geophys Res 1998;103(C2):3179–201.

[5] Bolanos-Sanchez R, Sanchez-Arcilla A, Cateura J. Evaluation of two atmo-spheric models for wind-wave modelling in the NW Mediterranean. J Mar Syst2007;65:336–53.

[6] Signell RP, Carniel S, Cavaleri L, Chiggiato J, Doyle JD, Pullen J, et al. Assessmentof wind quality for oceanographic modelling in semi-enclosed basins. J MarSyst 2005;53:217–33.

[7] Cacciamani C, Emiliani P, Ferri M, Minguzzi E. In: Doms G, Shatter U, editors.High resolution verification of hydrostatic and non-hydrostatic LAM precip-itation forecasts in Italy, COSMO newsletter, vol. 2. Offenbach: DeutscherWetterDienst (DWD); 2002. p. 176–86.

[8] Hodur RM. The naval research laboratory’s Coupled Ocean/AtmosphericMesoscale Prediction System (COAMPS). Mon Weather Rev 1997;125:1414–30.

[9] Ardhuin F, Bertotti L, Bidlot JR, Cavaleri L, Filipetto V, Lefevre JM, et al.Comparison of wind and wave measurements and models in the WesternMediterranean Sea. J Ocean Eng 2007;34:526–41.

10] Ponce de León S, Guedes Soares C. Sensitivity of wave model predictions towind fields in the Western Mediterranean Sea. J Coast Eng 2008;55(11):920–9.

11] Cavaleri L, Bertotti L. The accuracy of modelled wind and waves fields inenclosed seas. Tellus 2004;56A:167–75.

12] Bertotti L, Cavaleri L. Large and small scale wave forecast in the Mediterranean

Sea. Nat Hazards Earth Syst Sci 2009;9:779–88.

13] Astraldi M, Gasparini P, Sparnocchia S. The seasonal and interannual variabilityin the Ligurian-Provenc al Basin. In: La Violette P, editor. Seasonal variabilityin the western Mediterranean Sea, coastal and estuarine studies, WashingtonAGU, vol. 46. 1994. p. 93–113.

[

[

an Research 34 (2012) 1– 9 9

14] Lionello P, Bhend J, Buzzi A, Della-Marta PM, Krichak SO, Jansá A, et al. Cyclonesin the Mediterranean region: climatology and effects on the environment. In:Lionello P, Malanotte-Rizzoli P, Boscolo R, editors. Mediterranean climate vari-abilità. Amsterdam: Elsevier; 2006. p. 325–72.

15] Trigo RM, Osborn TJ, Corte-Real JM. The North Atlantic Oscillation influenceon Europe: climate impacts and associated physical mechanisms. J Clim Res2002;20:9–17.

16] Canellas B, Orfila A, Méndez FJ, Menéndez M, Gomez-Pujol L, Tintoré J. Applica-tion of a POT model to estimate the extreme significant wave height levelsaround the Balearic Sea (western Mediterranean). J Coast Res Spec Issue2007;50:329–33.

17] Orfila A, Álvarez A, Tintoré J, Jordi A, Basterretxea G. Climate teleconnectionsat monthly time scales in the Ligurian Sea inferred from satellite data. J ProgOceanogr 2005;66:157–70.

18] Ponce de León S, Guedes Soares C. The sheltering effect of the Balearic Islandsin the hindcast wave field. J Ocean Eng 2010;37(7):603–10.

19] Undén P, Rontu L, Jarvinen H, Lynch P, Calvo J, Cats G, et al. HIRLAM-5 scientificdocumentation. Hirlam sci. report, DMI; 2002, 132 pp.

20] Simmons A. Observation, assimilation and the improvement of Global WeatherPrediction – some results from operational forecasting and ERA-40. In: Pre-dictability of weather and climate. Cambridge University Press; 2006. p.428–528.

21] Skamarock WC, Klemp JB, Dudhia J, Gill DO, Barker DM, Duda MG, et al. Adescription of the advanced research WRF version 3. NCAR technical note 475;2008, 113 pp.

22] Yang X, Kmit M, Petersen C, Nielsen NW, Sass BH, Amstrup B, et al. The DMI-HIRLAM upgrade in November 2005. DMI technical report; 2005, 17 pp.

23] Sotillo MG, Alvarez Fanjul E, Castanedo S, Abascal AJ, Menendez J, EmelianovM, et al. Towards an operational system for oil-spill forecast over Span-ish waters: initial developments and implementation test. Mar Pollut Bull2008;56:686–703.

24] Jung T, Klinker K, Uppala S. Reanalysis and reforecast of three major Europeanstorms of the twentieth century using the ECMWF forecasting system, Part II:ensemble forecasts. J Meteorol Appl 2005;12:111–22.

25] Skamarock WC, Klemp JB, Dudhia J, Gill DO, Barker DM, Wang W, et al. Adescription of the advanced research WRF version 2. NCAR technical note 468;2005, 88 pp.

26] Monbaliu J, Padilla-Hernandez R, Hargreaves J, Carretero-Albiach JC, Luo W,Sclavo M, et al. The spectral wave model WAM adapted for applications withhigh spatial resolution. J Coast Eng 2000;41:41–62.

27] Janssen PAEM. Wave induced stress and the drag of air flow over sea waves. JPhys Oceanogr 1989;19:745–54.

28] Janssen PAEM. Quasi-linear theory of wind-wave generation applied to waveforecasting. J Phys Oceanogr 1991;19:745–54.

29] Hasselmann K, Barnett TP, Bouws E, Carlson H, Cartwright DE, Enke K, et al.Measurements of wind-wave growth and swell decay during the Joint NorthSea Wave Project (JONSWAP). Deutsche Hydrographische Zeitschrift A8 (12);1973, 95 pp.

30] Hasselmann S, Hasselmann K, Allender JH, Barnett TP. Computations andparameterizations of the nonlinear energy transfer in a gravity wave spectrum.Part II: Parameterizations of the nonlinear energy transfer for application inwave models. J Phys Oceanogr 1985;15:1378–91.

31] Bidlot JR, Holmes DJ, Wittmann PA, Lalbeharry R, Chen HS. Intercomparisonof the performance of operational ocean wave forecasting systems with buoydata. Weather Forecast 2002;17:287–310.

32] Charnock H. Wind stress on a water surface. Q J R Meteor Soc 1955;81:639–40.33] Papanastasiou DK, Melas D, Lissaridis I. Study of wind field under sea breeze

conditions; an application of WRF model. Atmos Res 2010;98:102–17.34] Portabella M, Stoffelen A. On scatterometer ocean stress. J Atmos Ocean Technol

2009;26:368–82.35] Verhoef A, Stoffelen A. ASCAT wind product user manual. EUMETSAT/OSI SAF;

2011, 26 pp.36] Stopa JE, Cheung KF, Chen YL. Assessment of wave energy resources in Hawaii.

Renew Energy 2011;36:554–67.