Characterization of surface heat fluxes in the Mediterranean Sea from a 44-year high-resolution atmospheric data set

Global and Planetary Change 63 (2008) 258–274

Contents lists available at ScienceDirect

Global and Planetary Change

j ourna l homepage: www.e lsev ie r.com/ locate /g lop lacha

Characterization of surface heat fluxes in the Mediterranean Sea from a 44-yearhigh-resolution atmospheric data set

Simón Ruiz a,⁎, Damià Gomis a, Marcos G. Sotillo b, Simon A. Josey c

a IMEDEA (CSIC-UIB), Mallorca, Spainb Puertos del Estado, Madrid, Spainc NOC, Southampton, United Kingdom

⁎ Corresponding author.E-mail address: [email protected] (S. Ruiz).

0921-8181/$ – see front matter © 2007 Elsevier B.V. Aldoi:10.1016/j.gloplacha.2007.12.002

a b s t r a c t

Article history:

Keywords:heat fluxseasonal cycleinterannual variabilityMediterranean Sea

8–2001) of model data with the aim of characterizing the low frequency (theseasonal cycle and lower) variability of surface heat fluxes. The data set was produced in the framework of theHIPOCAS project through a dynamical downscaling (1/2˚×1/2˚) from theNCEP/NCAR global reanalysis using theatmospheric limited areamodel REMO. The added value of this data set is the better representation of regionaland local aspects related to thermal and dynamical effects resulting from its higher resolution.The basinmean values of the heatfluxes have been estimated in 168W/m2 for the solar radiation (QS), 73W/m2

for the longwave net radiation (QB), 8 W/m2 for the sensible heat (QH) and 88 W/m2 for the latent heat (QE),giving a total heat budget of about −1 W/m2. The main differences with respect to previous results are thereduced QS and QE terms. The seasonal cycle accounts for a significant fraction of the variability (75%, 20% and10% for QS, QE and QH) except for QB (less than 1%). The total heat budget has an amplitude of 164 W/m2 andpeaks by middle June, in agreement with previous works and observations.The interannual variability of each component has been first quantified by the standard deviation of the annualmean values, obtaining ±2.0 W/m2 for QS, ±1.1 W/m2 for QB, ±4.7 W/m2 for QE and ±1.1 W/m2 for QH. Thedominant modes have been obtained through an EOF analysis, which is shown to be robust with respect to theanalysis domain. The correlation between the amplitudes of the radiation terms (QS and QB) and MOI wintervalues is higher than 0.7 (in absolute value) in the Eastern basin. For the other flux components the correlationwith the MOI is less than 0.7 everywhere. The correlation between the heat flux terms and the NAO is smallerthan 0.7 for all terms. From the evaluation analysis, HIPCOAS fluxes show stronger correlations with theobservation based NOC fields than are obtained with the original NCEP/NCAR fluxes for the full set ofinterannually varying heat flux estimates. Thus, the downscaling has led to an improved representation of theinterannual variability when compared with observations.

© 2007 Elsevier B.V. All rights reserved.

1. Introduction

Improving our knowledge of the heat fluxes between ocean andatmosphere is essential for understanding the climate system andpotential climatic changes. The surface fluxes play a key role in crucialprocesses such as the deep water formation, and therefore a betterdescription of their variability will help to understand the results ofnumerical simulations. The total heat budget QT consists of tworadiation components and two turbulent components. The first arethe solar net radiation flux QS absorbed by the sea (shortwave) and thenet terrestrial flux radiation QB emitted by the sea (longwave).The turbulent components are the latent heat flux QE and the sensibleheat QH, which are related to energy losses of the sea by evaporationand convection, respectively. Hence, the budget can be represented by

l rights reserved.

QT=QS+QB+QE+QH. Positive values denote a gain of heat by the ocean.Semi-enclosed basins such as the Mediterranean Sea are ideal for thecharacterization of heat fluxes, since they make possible to investigatethe closure of the heat budget. The key hypothesis is that at longtimescales, the heat transport through the strait of Gibraltar must bebalanced by the ‘vertical’ heat fluxes (exchanged with the atmospherethrough the ocean surface) integrated over the basin. Using in-situobservations (current and temperature) from moored instruments,Macdonald et al. (1994) estimated an average heat transport troughthe Strait of Gibraltar of 5.2±1.3 W/m2, where the positive signdenotes a net heat transport from the Atlantic into the Mediterranean.Others authors have obtained the long-term heat flux throughGibraltar from estimates of the volume transport and the tempera-tures of the inflow and outflow. Results range from 8.5 W/m−2

(Bethoux, 1979) to 5 W/m2 (Bunker et al., 1982). According to Bunkeret al. (1982), and Garrett et al. (1993), the uncertainty of theseestimates is b2 W/m2, therefore we can use it as a reference for theevaluation of the surface heat flux budget. The Black Sea heat flux is

259S. Ruiz et al. / Global and Planetary Change 63 (2008) 258–274

neglected in all computations, since it is less than 1 W/m2 (Garrettet al., 1993).

Several studies (Bunker et al., 1982; Garrett et al., 1993; Schianoet al., 1993; Gilman and Garrett, 1994) have compared long termaverages of vertical heat fluxes with the heat transport though thestrait of Gibraltar, obtaining discrepancies of up to 30 W/m2. Thereasons given for the disagreement are the different periods coveredby the data sets or the different bulk formula parameterizations. Inparticular, Garrett et al. (1993) suggested that the solar radiation termis usually overestimated and that the latent and sensible heat isunderestimated by most parameterizations. Implementing somecorrections for the latent and sensible heat they obtained that theamplitude of the seasonal cycle of the total heat increased a 20%, butthe lack of reliable observations prevented a deeper insight in thebenefits of the correction.

Gilman and Garrett (1994) adopted the formulation of Garrett et al.(1993) but moreover they corrected the parameterization of theisolation including the effects of the aerosols. Later, Castellari et al.(1998) determined the set of heat flux bulk formulas providing thebest estimates of the heat balance of the Mediterranean basinaccording to the available data sets. Checking different formulationsthey found total heat flux values ranging between −17 W/m2 and+67 W/m2. Using the May formula for QB (May, 1986) and the Kondoscheme (Kondo, 1975) for QH and QE they obtain a long-term meanvalue of −11 W/m2, which is in close agreement with theirobservations.

In this context our study intends to evaluate a new heat flux dataset produced in the framework of the EU-funded HIPOCAS (Hindcastof Dynamic Processes of the Ocean and Coastal Areas of Europe)Project (Ratsimandresy et al., in press-a). The 44 years (1958–2001)of model data were produced through a dynamical downscaling(1/2˚×1/2˚) of the NCEP/NCAR global reanalysis. The added value ofthis data set is the better representation of local aspects related tothermal and dynamical effects. The downscaling technique ensuresthat large scale information reliably simulated by global climatemodels is transferred to smaller scales (Sotillo et al., 2005). Thisfeature is particularly important in the Mediterranean Sea, wheresynoptic and mesoscale phenomena play an important role in boththe atmosphere and the ocean. The horizontal and vertical heattransfers (in particular the latent heat flux) strongly depend on acorrect resolution of temperature, humidity and wind structures,and therefore the new data set constitutes a good opportunity toimprove the estimation of the heat budget. In this work, theevaluation of the HIPOCAS data is done from a comparison withindependent ship based datasets from the National OceanographyCentre of Southampton.

Hence, a first aim of this work will be the quantification of each ofthe components of the long-term mean heat budget. Results will becompared with previous estimates and the total budget will becompared with heat transport values through the Strait of Gibraltar.

A second aim will be characterizing the seasonal cycle of the fourheat flux components. Deseasoning the series is a compulsory stepprior to the modal decomposition, but the information obtained onthe seasonal cycle is also valuable by itself.

The third and last aim is to obtain a modal decomposition of theinterannual variability of heat fluxes. In particular wewill focus on theeventual relations between the observed variability and climaticindices. These relations have already been suggested in previousstudies, often through the impact of heat fluxes on sea level variability.Hence, the clear link between North Atlantic sea level variability andthe NAO is apparently due to wind forcing effects, but a smallerthermosteric contribution has also been suggested (Tsimplis et al.,2006). In the Mediterranean, sea level variability has been related tothe NAOmainly through the atmospheric pressure anomalies, but alsothrough changes in the evaporation-precipitation budget (Tsimplisand Josey, 2001).

Less attention has been paid to the Mediterranean Oscillationindex (MOI), originally defined as the pressure difference betweenthe Mid-North Atlantic and the Southeast Mediterranean (Supicet al., 2004). More recently Sušelj and Bergant (2006) have shownthat the MOI actually corresponds to the time amplitude of theleading mode resulting from a principal component analysis appliedto monthly surface atmospheric pressure anomalies over the region(between 30°E and 40°W and between 30°N and 60°N). In thenegative (low pressure) phase the MOI is linked to intensecyclogenesis over the Western Mediterranean, then yielding anom-alously wet conditions over most of the basin except in the south-eastern Mediterranean. The situation is inverted in the positive(high pressure) phase. The MOI has been reported to reflectMediterranean variability (Supic et al., 2004) including the flowexchanges trough Gibraltar (Gomis et al., 2006) better than the NAOindex.

The above mentioned references to sea level variability pointtowards the ultimate aim of this work: in the near future the HIPOCASfluxes (altogether with precipitation/evaporation and other forcingvariables) will be used to carry out a 44 year run of a baroclinic modelaimed to reproduce the steric variability of sea level. Hence,conducting this previous study on the characterization and quantifi-cation of the different components of the heat flux is essential tounderstand the results of the run.

The paper is organized as follows: Section 2 gives a detaileddescription on the HIPOCAS data set, focusing on the downscalingprocess and model set up. The climatic indices and the data analysisare also described. Details on the parameterization of the fluxcomponents are given in an Appendix. Section 3 presents the resultsin four subsections: the mean heat budget, the seasonal cycle, themodal decomposition and the evaluation of the HIPOCAS dataset. Thediscussion and conclusions are presented in Section 4.

2. The data set and data analysis

The heat flux data set covers the domain shown in Fig. 1(Mediterranean Sea, Black Sea and the Northeastern Atlantic area ofthe Iberian Peninsula). In this study we will focus only on theMediterranean basin (referred hereafter as the MED domain). For theEOF analysis we additionally considered two sub-domains, theWesternMediterranean (WMdomain) and the EasternMediterranean(EM domain).

2.1. The downscaling

The dynamical downscaling from the NCEP/NCAR global reanalysis(Kalnay et al., 1996) was carried out by Puertos del Estado (Sotillo et al.,2005) using the atmospheric limited area model REMO (Jacob andPodzun, 1997). The resolution of the HIPOCAS 44-year Mediterraneanhindcast is 0.5°×0.5° in space. Hindcast outputs were hourly storedover the whole 44-year period.

A spectral nudging technique, following the one proposed by VonStorch et al. (2000), was applied to the long-term run in order to keepthe modelled atmospheric flow close to the realistic time-variablelarge-scale atmospheric states provided by the NCEP forcing. At thesame time, the model is free to resolve the smaller scales (i.e. regionalfeatures and surface processes) independently of the global forcingdata, which is not adequate to realistically reproduce such regionalfeatures. The spectral nudging is achieved by adding nudging terms inthe spectral domain with maximum efficiency for smaller wavenumbers and higher altitudes. It has only been applied to zonal andmeridional wind components, leaving the rest of the variables,including the ones related to heat and water processes, free to berebalanced attending to the new flow regime, ruled by the large scaleconditions provided by the forcing, but additionally conditioned bythe improved regional configuration.

Fig. 1. Map of the Mediterranean basin. The main regional sub-basins are indicated.

260 S. Ruiz et al. / Global and Planetary Change 63 (2008) 258–274

Several variables of the Mediterranean HIPOCAS data set havealready been validated against satellite and in-situ observations(Sotillo, 2003; Sotillo et al., 2005; Sotillo et al., 2006; Sotillo et al.,2007; Ratsimandresy et al., in press-a), enhancing the confidence onthe realism of the hindcast. This is the case of sea level pressure, 2-mtemperature, 10-m wind and precipitation. Additionally, someHIPOCAS atmospheric fields such as mean sea level pressure and10-m winds have been used as driving fields in long-term hindcastruns of ocean (waves and sea level) models (Ratsimandresy et al., inpress-a,b). The good agreement between the oceanographic outputsand observations provided an indirect proof of the quality of theHIPOCAS atmospheric forcing. In fact, this oceanographic validation ofthe atmospheric forcing has allowed to validate HIPOCAS data overoffshore areas, where the lack of meteorological measurements is ashortcoming.

Some of the previous works focussed on HIPOCAS validationhave verified a significant regional improvement in the Mediterra-nean for variables such as 10-m wind speed and direction, andprecipitation, in comparison to global reanalysis (NCEP/NCAR andERA). Since surface heat fluxes are highly conditioned by theprevailing boundary layer conditions, one can expect that theimproved Mediterranean surface thermal, wind and water condi-tions lead to a consequent improvement of the surface heat fluxes.In this work we carried out a first comparison between the HIPOCASheat fluxes and those provided by global reanalysis using anindependent ship based datasets from the National OceanographyCentre of Southampton.

2.2. Model set-up

In order to generate the long-term Mediterranean data set, thehydrostatic REMOmodel was set-up in its climatic mode. This specificconfiguration implies that whereas the dynamical scheme used in themodel set-up is analogous to the one used by the DWD EM/DMregional forecast operational system (Doms et al., 1995), theparameterization of physical processes, including those related toshort and longwave radiation, surface fluxes and vertical diffusion, aswell as stratiform and convective cloud processes, was accomplishedby means of the code used in the general circulation model ECHAM4(Roeckner et al., 1996).

The ECHAM4 radiation scheme was adopted from the onepreviously tested in the ECMWF model (Fouquart and Bonnel, 1980;Morcrette et al., 1986) with some modifications for inclusion of:additional greenhouse gases such as methane, nitrous oxide and 16CFCs; 14.6 μmozone band and some types of aerosols, according to theGADS dataset (Koepke et al., 1997). The details on the parameteriza-tion of the four heat flux components are given in Appendix A.

A number of validation studies, using both satellite and in-situmeasurements, have evaluated the performance of ECHAM4 physicsin reproducing different aspects, such as water vapour distribution

(Chen et al., 1996), cloud cover (Chen and Roeckner, 1996), and surfaceradiation fluxes (Wild et al., 1996).

2.3. Data analysis

The hourly outputs of the model were first averaged to producedaily values at each grid point. Hence, for each variable we have a totalof 16068 daily fields spanning the 1958–2001 period. The seasonalcycle of each heat flux component was determined by fitting twoharmonic functions to each grid point series in the form:

Y tð Þ ¼ A1cos w1t � F1ð Þ þ A2cos w2t � F2ð Þ ð1Þ

where A1, w1 and F1 (A2, w2 and F2) are the amplitude, frequency andphase of the annual (semi-annual) signal. Additionally, a ‘mean’seasonal cycle was computed by fitting Eq. (1) to the spatial meanvalues (averaged over the MED domain) of each component.

Once the seasonal cycle at grid point was determined, it wassubtracted from each grid point series. The detrended series werethen used to characterize the interannual variability through anEmpirical Orthogonal Function (EOF) analysis. This techniquerequires the resolution of a linear algebra problem that from thenumerical point of view can be more or less stable depending onthe strategy used. Toumazou and Cretaux (2001) compared threedifferent approaches: the Singular Value Decomposition (Goluband Kahan, 1965; Golub and van Loan 1996), a backward stablealgorithm (Andersson et al., 1992; The MathWorks, 1992) and theirown (Lanczos) strategy, which uses an iterative eigensolver basedon a Krylov-type method. The Lanczos approach consists ofcomputing only the N largest singular values and associatedvectors, the advantage of this technique being that it provides thesame quality of results as the SVD strategy but with a lowercomputational cost. This allows the application of the EOF analysisto large dataset (our case) and therefore we followed this approach.The spatial modes obtained from the eigenvectors were normalizedin such a way that all had a unity variance. Because EOFdecompositions are dependent on the selected domain, we carriedout the process for the MED domain and also for the WM and EMsub-domains.

2.4. Climatic indices

The NAO index can be defined in slightly different ways, a commonone being the normalized pressure difference between Gibraltar andRejkiavik stations. We used monthly values spanning the modelperiod (1958–2001), which were obtained from the Climate ResearchUnit (http://www.cru.ac.uk/; Jones et al., 1997). Monthly series of theMediterranean Oscillation Index were kindly provided by Kay Sušelj.The two indices are obviously not independent; as an example, theMediterranean cyclogenesis that largely determines the MOI, strongly

Fig. 2. Spatial distribution of the temporal mean (left panel, in W/m2) and variance (right panel, in [W/m2]2) computed from gridpoint daily series. From top to bottom: sensible heat (QH), latent heat (QE), solar radiation (QS) and net longwaveradiation (QB). The black lines delimitate the zero level. 261

S.Ruizet

al./Globaland

PlanetaryChange

63(2008)

258–274

Table 1Mediterranean long term mean heat budget estimated from the HIPOCAS data set

Hipocas data 1958–2001 QS QB QH QE QT

Mediterranean Sea 168 73 8 88 −1Western Mediterranean basin 162 71 7 79 5Eastern Mediterranean basin 172 74 9 93 −4

262 S. Ruiz et al. / Global and Planetary Change 63 (2008) 258–274

depends on the activity of North Atlantic fronts, which are in turnrelated to the NAO.

3. Results

3.1. Mean values and variance of surface heat fluxes

Each grid-point daily series of the four heat flux components wasfirst averaged to obtain the spatial distributions of the temporal meanand its variance the modelled period (1958–2001). The time-meansensible heat QH is negative over the whole domain except in theAlborán Sea, where some positive values are found (Fig. 2). Themaximum variance of the daily series corresponds to the Gulf of Lionsand the northern Adriatic Sea, with values of about 1.2×103 (W/m2)2,and to the Aegean Sea, where it reaches 2.5×103 (W/m2)2. Conversely,the grid-point series were also spatially averaged day by day, in orderto obtain the temporal evolution of the spatial mean. The basin meanvalue for the whole period is −8.4 W/m2 with a standard deviation of15.6W/m2.When the spatial mean series is yearly averaged in order toeliminate the seasonal cycle and higher frequencies, the standarddeviation reduces to 1.1W/m2. This can be considered a representativefigure of the interannual variability and therefore of the dependenceof the mean value on the considered time period.

Fig. 3. Time series of the yearly averaged heat flux components and of the total heat budgetheat transport through Gibraltar. (For interpretation of the references to colour in this figur

The latent heat QE presents spatial distributions that resemblethose of QH, but with higher (absolute) values. The temporal meanranges from −20 W/m2 in the Alborán Sea to −120 W/m2 in theLevantine basin, Aegean Sea and Gulf of Lions. Maximum variances forthe daily series are also observed in the Aegean Sea and the Gulf ofLions (up to 9×103(W/m2)2), and to a less extent in the Levantine basin(7.5×103(W/m2)2). Relatively high variances are also observed in theIonian Sea,where they reach6.5×103 (W/m2)2. The basinmeanvalue forthewhole period is −87.9W/m2with a standard deviation of 48.9W/m2.The variability reduces to 4.7 W/m2 for the annual mean series.

Both the temporal mean and the variance of the solar radiation fluxQS show a clear South–North gradient. QS values range from 140W/m2

in the northern Adriatic Sea to 190 W/m2 in the Levantine basin. Thehighest variance of the daily series corresponds to the Adriatic sea, theAegean sea and the Gulf of Lions, where it reaches 9×103 (W/m2)2. Thebasin mean value for the whole period is 168.2 W/m2, with a standarddeviation of 76.1 W/m2. The variability reduces to 2.0 W/m2 for theannual mean series.

Finally, the spatial distribution of the net longwave radiation QB

time mean is rather uniform over all the Mediterranean Sea, the meanvalue being −72.7 W/m2. The variance of the daily series is lesshomogeneous, showingmaximumvalues of about 600 (W/m2)2 in theGulf of Lions, Tyrrhenian Sea and Adriatic Sea. The standard deviationof the basin mean is 8.1 W/m2 when computed for the daily series and1.1 W/m2 for the annual mean series.

Hence it results that for the total heat budget we obtain apractically neutral budget: the four components give a total value of−0.7 W/m2. The mean values of the four components have also beencomputed for each sub-basin (Table 1). The total heat budget of theWM is positive (4.6 W/m2), whereas for the EM the budget is negative(−4.0 W/m2).

for the modelled period. The red lines delimitate the range of previous estimates for thee legend, the reader is referred to the web version of this article.)

263S. Ruiz et al. / Global and Planetary Change 63 (2008) 258–274

In order to investigate the dependence of these numbers on theconsidered period, we plotted the basin-mean annual series for eachcomponent and for the total heat budget (Fig. 3). The plot shows thatthe annual heat budget fluctuates between ±10 W/m2, though formost of the period it keeps within ±5 W/m2. More precisely, the totalbudget reaches the lowest values at the beginning and end of themodelled period, whereas the highest values are reached in the formof successive peaks in 1972, 1978, 1985 and 1992.

Looking at the four components it seems clear that the one drivingthe interannual variation of the total heat budget is the latent heat.

Fig. 4. The seasonal cycle (thick line) obtained from fitting an annual and semi-annual hacomponents and to the total head budget. The thin line corresponds to a sample year of the 1heat (QE), solar radiation (QS), net longwave radiation (QB) and total heat budget (QT).

And looking at the seasonal distribution of this component (notshown) we found that most of the observed variability occurs inautumn, suggesting that it could be related with important anomaliesin the deep water formation.

3.2. The seasonal cycle

The two harmonic functions shown in (1) were first fitted to thebasin mean daily series of each component (Fig. 4). As expected, QS isthe only component for which the harmonic seasonal cycle clearly

rmonic functions to the temporal evolution of the spatial mean of the four heat flux958–2001 period covered by the data set. From top to bottom: sensible heat (QH), latent

264 S. Ruiz et al. / Global and Planetary Change 63 (2008) 258–274

Fig. 5. Spatial distribution of the amplitude (left, in W/m2) and phase (right, in day) of the annual harmonic cycle. From top to bottom: sensible heat (QH), latent heat (QE), solar radiation (QS), net longwave radiation (QB) and total heatbudget (QT).

265S.Ruiz

etal./

Globaland

PlanetaryChange

63(2008)

258–274

266 S. Ruiz et al. / Global and Planetary Change 63 (2008) 258–274

dominates over higher frequency variability. For QE and QH the highfrequency variability is larger, but still of the same order than theseasonal cycle, whereas for QB the seasonal cycle is almost non-significant compared to high frequency variability. In all cases thesemi-annual cycle (not shown) is much weaker than the annual cycle,and therefore its impact on the results is negligible.

For the sensible heat flux QH the cycle is positive from May toAugust, reachingmaximumvalues of about 6W/m2 by the end of June.The cycle is negative for the rest of the year, reaching minimumvaluesof about −22 W/m2 by the end of December. The amplitude of thecycle is about 14W/m2, which implies that the seasonal cycle accountsfor a variance of about 100 (W/m2)2 (a half of the squared amplitude).This represents more than 10% of the total QH daily variance, which isless than 1000 (W/m2)2 over most of the basin (see Fig. 2).

For the latent heat flux QE the minimum value (−125 W/m2) isdetected by mid November and the maximum (−51 W/m2) by midMay. The amplitude of the seasonal cycle is 37W/m2, which translatesin a variance of 685 (W/m2)2. This represents about 20% of the total QE

daily variance, which is larger than 3000 (W/m2)2 over most of thebasin (Fig. 2).

The harmonic seasonal cycle of the shortwave radiation QS has anamplitude of about 105 W/m2 and obviously reaches the maximumvalue (273 W/m2) by the summer solstice and the minimum value(63 W/m2) by the winter solstice. The variance associated with theseasonal cycle is about 5500 (W/m2)2, which represents as much as75% of the typical daily QS variance (see Fig. 2).

The net longwave radiation QB is the component which has a lessevident harmonic cycle, the amplitude being only 2.6 W/m2. Theminimum value (−75 W/m2) is reached in October (Fig. 4), whereasthe maximum (−70 W/m2) is found in May. The variance associatedwith the seasonal cycle is about 3.5 (W/m2)2, i.e., just 1% of the total QB

daily variance, which is larger than 300 (W/m2)2 over most of thebasin (Fig. 2). Such small values put in question the statisticalsignificance of the seasonal cycle of this component.

Summarizing, the net longwave radiation and the sensible heatcontribute less than the latent heat and the solar radiation to theseasonal cycle of the total heat budget (Fig. 4). As expected, thedominant role at the annual frequency clearly corresponds to the solarradiation, whereas the latent heat component is the one showing alarger variability at higher frequencies. The seasonal cycle of the totalheat budget can in fact be computed by fitting the function (1) to thesum of the four spatial mean components. Fig. 4 shows that the cyclepeaks by mid of June and a minimum value in December. Theamplitude of the total cycle is about 165 W/m2.

In order to examine the regional variability of the seasonal cyclewealso fitted the harmonic functions to each grid point time series andplotted the spatial distribution of the obtained amplitudes and phases(Fig. 5). For QH the amplitude of the annual harmonic is quite uniform,keeping between 10 and 15 W/m2 over most of the basin, but itincreases to 25–30 W/m2 in the Aegean Sea. The phases show anoverall Southwest–Northeast gradient, peaking by the beginning ofJune (day 155) in the African and Spanish shores and bymiddle August(day 195) in the Aegean Sea.

For QE the variability of the amplitude is higher: in the Levantinebasin it reaches 55W/m2, but it reduces to 25W/m2 to the south of theIonian Sea and the Adriatic and to 10 W/m2 in the Alborán Sea. TheBalearic–Ligurian basin has amplitudes slightly lower than theLevantine basin (40 W/m2). The phases are all included in a twomonth period, as for QH, though in this case they do not show a simplegradient structure. Another difference is that the cycle is about onemonth advanced to that of QH: the phases range from day 125(beginning of May) in the WM to day 165 (middle July) in the Adriaticand Aegean seas.

As expected, the amplitude of QS exhibits a clear South–Northgradient with minimum values (around 95 W/m2) along the Africancoast and maximum values (around 120 W/m2) in the northern

shores. Conversely, the phase shows a weak but clear West–Eastgradient: it ranges from middle June (day 165–170) in the WM to theend of June (day 180) in the EM.

Finally, the amplitude of QB is about 2–3 W/m2 over most of thebasin with the only exception of the Eastern Levantine basin, where itincreases to 10–15 W/m2. The phase of QB is the one showing thelargest variability: while the mean cycle peaks in May, it advances tolate March (day 55) in a wide area covering the Ionian and Aegean Seaand delays until middle July (day 190) in the Spanish shores. It must bestressed, however that for a cycle as small as the one obtained for thiscomponent, the uncertainty of the phases is significantly larger thanfor the other components.

Fig. 5 also shows the distribution of the amplitude and phase of thetotal heat budget. Maximum amplitudes (almost 200 W/m2) corre-spond to the Aegean Sea, while minimum values (about 120 W/m2)are obtained in the Alborán Sea. The phases fall all within June: fromday 160–165 in the WM to day 170–175 in the Adriatic and AegeanSea.

3.3. Modal decomposition

The EOF decomposition of the detrended series was undertaken forthe whole Mediterranean basin (shown in Figs. 6 and 7) and for theWM and EM sub-basins (not shown). The three decompositions showconsistent results. For the MED domain, the leading EOFs of QH, QE, QB

and QS are all basin-wide modes that explain 36.8%, 35.7%, 21.2% and15.6% of the variance, respectively. The second EOFs are sub-basinmodes (the nodal lines run more or less along the separation betweenthe two sub-basins) that explain 25.8%, 26.1%, 16.2% and 12.6% of thevariance. In the WM and EM decompositions (and in particular for QS

and QB), these two modes appear altogether as a single leading EOFexplaining 62.8% (WM) and 50% (EM) of the QH variance, 67.1% and47.3% of QE variance, 42.5% and 25.8% of QB variance and 32.6% and21.3% of QS variance.

The third modes of the MED domain explain 7.6%, 8.0%, 8.9% and8.4% of the variance of QH, QE, QB and QS, respectively; this numbersare close to the secondmodes of theWM (EM) domains: 11.4% (18.4%),11.2% (20.3%), 15.4% (17%) and 14.9% (14.5%). Hence the first threemodes of the MED domain (accounting altogether for 70.2%, 69.8%,46.3%, 36.6% of the variance) would be equivalent to the two leadingEOFs of the WM (EM) sub-basins (accounting for 74.2 (68.4), 78.3(67.6), 58.1 (42.8) and 47.5% (35.8%) of the variance of QH, QE, QB andQS, respectively).

Combining the temporal amplitudes with the spatial pattern of themodes allows the description of the main features of the heat fluxvariability. Hence, the leading EOF of QH has minimum amplitude intheWesternMediterranean and to the East of the Levantine basin, andmaximum amplitude in the Aegean Sea and to the North of the IonianSea (Fig. 6). The variance of the temporal amplitude is 27.3 (W/m2)2,and therefore the variance of this first QH mode across the basin willrange from a fraction of this value in the Western Mediterranean andto the East of the Levantine basin (where the EOF amplitude is lessthan 1) to five times this values in the Aegean Sea. For QE, the varianceof the leading amplitude is 181.2 (W/m2)2. The spatial mode showsminimum values (below 1) in the Alboran Sea and to the East of theLevantine basin, and maximum values (up to four units) in the Ionianand Tyrrhenian Sea and in the Balearic–Ligurian basin (Fig. 6). Thesecond modes of QH and QE are clearly sub-basin modes in the sensethat theWM and the EM oscillate in opposite phase. The third mode ismore complicated, with two nodal lines located in the middle of theWM and EM.

The variance of the leading amplitude of QS is 116.1 (W/m2)2, andthe spatial mode shows minimum values in the Levantine basin andthe Aegean and Alboran Sea (around 0.5 units) and a clear maximumin the Tyrrhenian and the Balearic–Ligurian basins, where it reaches 3units (Fig. 7). The regions of maximum and minimum values are

Fig. 6. Leading EOFs of the Mediterranean basin for sensible heat (QH) and latent heat (QE). The percentage of variance accounted by each mode is indicated at the top right corner of each map. The black lines delimitate the zero level.

267S.Ruiz

etal./

Globaland

PlanetaryChange

63(2008)

258–274

Fig. 7. Leading EOFs of the Mediterranean basin for shortwave solar radiation (QS) and net longwave radiation (QB). The percentage of variance accounted by eachmode is indicated at the top right corner of each map. The black lines delimitatethe zero level.

268S.Ruiz

etal./

Globaland

PlanetaryChange

63(2008)

258–274

Fig. 8.Winter correlation between theMOI and the four heat flux components with a 95% significance level (values lower than 0.251 have beenmasked). From top to bottom and from left to right: sensible heat (QH), latent heat (QE), shortwavesolar radiation (QS) and longwave radiation (QB).

269S.Ruiz

etal./

Globaland

PlanetaryChange

63(2008)

258–274

Fig. 9. Values for the correlation coefficient squared (r2) between individual monthlymean net heat flux values for a.) HIPOCAS and NOC; b.) NCEP/NCAR and NOC.

Table 2Mediterranean long term mean heat budget estimated by different authors

Authors QS QB QH QE QT

Bethoux (1979) 195 68 13 120 −6Bunker et al. (1982)(1) 202 68 13 101 20Bunker et al. (1982)(2) 202 68 11 130 −7May (1986) 193 68 11 112 2Garrett et al. (1993) 202 67 7 99 29Gilman and Garrett (1994) 183 77 7 99 0Castellari et al. (1998) 202 78 13 122 −11

270 S. Ruiz et al. / Global and Planetary Change 63 (2008) 258–274

similar for the leading mode of QB, but in this case maximum valuesare restricted to the Tyrrhenian Sea (Fig. 7). The variance of the leadingamplitude of QB is 30.5 (W/m2)2. For the second, sub-basin modes weobtained amplitude variances of 174.7 (46.2) (W/m2)2 for QS(QB), andfor the third mode we obtained 100.4 (23.4) (W/m2)2.

In order to investigate eventual relations between the EOFamplitudes and climatic indices, the amplitudes were monthlyaveraged and correlated with monthly values of the NAO index andthe MOI. Results were below 0.5 in all cases. In order to eliminate anexcessive averaging in time, we split the series in seasons. Correlationshigher than 0.5 were obtained only between the (three amplitudes ofthe) two radiation terms and the winter MOI index. Finally, in order toeliminate also an excessive averaging in space, we computed thecorrelation between every grid point and the climatic indices, againfor seasonal series. Regional correlations higher than 0.5 were thenobtained for all the heat flux components, but only for winter valuesand for the MOI. In fact, MOI and NAO correlations are clearly notindependent: they show a rather similar pattern, with opposite sign,but NAO values are always smaller in magnitude. We only present theresults obtained for the MOI (Fig. 8).

For the turbulent terms QH and QE correlations are positive in theEM and negative in the WM (opposite sign for the NAO). Values areslightly higher for QH than for QE, but they do not go beyond 0.7. Thetwo radiation terms show higher correlation values and a differentpattern. For QS correlations are negative over most of the basin,reaching −0.8 in the Tyrrhenean, Adriatic, Ionian and Aegean seas andin the Levantine basin; similar values are obtained in wide sectors ofthe Atlantic. A similar, though slightly more complex pattern isobtained for QB, but with positive values. Correlation values reach 0.8mainly in the Ionian and Aegean seas and in the Levantine basin.

3.4. Evaluation of HIPOCAS using the NOC flux dataset

In the preceding sections we have presented results from the newdownscaled HIPOCAS flux dataset. We now consider whether thedownscaling has led to an improvement in the original NCEP/NCAR

fields. A full evaluation would ideally include comparison againstmeasurements from air-sea flux reference buoys and high quality insitu observation based datasets (primarily voluntary observing shipsand buoys) following the methodology of Josey and Smith (2006).However, although research buoys with air-sea flux sensors have beendeployed at a limited number of locations in the Mediterranean Seathis data has not as yet been fully assessed and made generallyavailable. Hence, we focus here on results from a comparisonwith theNOC flux dataset (Josey et al., 1999; Josey 2001; Josey, 2003) which isprobably the most accurate of the currently available ship baseddatasets. We present results obtained from a correlation analysiswhich enable us to establish confidence in the downscalingmethod asa means for improving the original NCEP/NCAR fields.

Both the downscaled HIPOCAS and the original NCEP/NCAR netheat flux fields have been compared with the NOC flux dataset on agrid cell by grid cell basis for the period of overlap (1980–1993)between the NOC fields and HIPOCAS. For this analysis the HIPOCASfields were re-gridded from the 0.5o×0.5o grid onto the 1o×1o grid onwhich the NOC fields are available. Following the re-gridding, thecorrelation coefficient between the individual monthly net heat fluxvalues for each grid cell fromHIPOCAS and NOCwere determined (seeFig. 9a). The same approach was applied to determine the correlationbetween NCEP/NCAR and NOC with the exception that, as the NCEP/NCAR fields are on a coarser grid than NOC, the nearest NCEP grid cellto the corresponding NOC value was used for the analysis (see Fig. 9b).The level of correlation is relatively high (r2N0.85) across the basin inboth cases with typically better agreement away from coastal margins.However, the HIPOCAS fields exhibit a stronger correlation with NOCthan NCEP/NCAR, particularly in the centre of the basin, indicatingthat the downscaling process has led to improved agreement betweenthe model fields and the purely observation based dataset.

4. Discussion and conclusions

The long-term averages obtained for each component of the heatflux must be first compared with previous estimates (see Table 2). Thevalue obtained for the positive, shortwave radiation termQS (168W/m2)is a 17% lower than the value given by Bunker et al (1982) and Garrettet al. (1993), and about 10% lower than the results of Gilman andGarrett (1994). The values obtained for the longwave radiation termQB (−73 W/m2) and the sensible heat QH (−8 W/m2) are in betweenthe range of values given by the different authors. Conversely, for thelatent heat QE the obtained value (−88 W/m2) is about 20–25% lower(in absolute value) than most of the literature values (between −99and −122 W/m2).

The differences in the solar radiation QS can hardly be attributed tothe different period considered for the study, since the interannualvariability is only ±2.0 W/m2. Although the higher resolution of theHIPOCAS data set has demonstrated to improve significantly theprecipitation with respect to lower resolution data sets, the improve-ment is more in the location than in the total amount. Therefore,although it could be argued that HIPOCAS cloudiness distribution isprobably more accurate, this does not necessarily hold for averagevalues. The remaining source for the differences is the differentparameterization scheme.

271S. Ruiz et al. / Global and Planetary Change 63 (2008) 258–274

The differences in the latent heatQE can neither be attributed to theperiod covered by the model. The interannual variability of this term(±4.7 W/m2) is higher than for QS, but still smaller than the observeddisagreement. Only at the very end of the modelled period the valuesof QE approach the upper boundary of those obtained in previousstudies (Fig. 3). Conversely, the observed differences in the latent heatQE could well be due to the higher spatial resolution of the HIPOCASdata set. In particular, the fact that local winds, temperature andhumidity are clearly better than for global reanalysis points in prin-ciple towardsmore accurate values for the latent heat. The distributionof QE variability (Fig. 2) actually shows well resolved structures in theGulf of Lions and in the Levantine basin. Both are well known areas ofdeep convection driving the Western Mediterranean Deep Water andLevantine Intermediate Water formation, respectively.

In terms of the total heat budget, the smaller values (with respectto previous studies) obtained for QS compensate the smaller values (inabsolute terms) obtained for QE, so that the total heat budget is withinthe range of previous studies. Although for the whole period we haveobtained a practical cancellation of the heat flux components, theyearly total budget fluctuates between ±10W/m2. The fluctuations arenot random: the total budget is clearly negative before 1975 and after1995, and mostly positive between 1975 and 1995. Therefore, thedifferences between previous estimates could well be due to thedifferent time periods covered by the data sets.

Regarding the estimates of heat transport through Gibraltar, theserange between 8.5 W/m−2 (Bethoux, 1979) and 5 W/m2 (Bunker et al.,1982). Therefore, when combined with the obtained ‘vertical’ fluxeswe obtain positive total values for most of the covered period. This is,the total heat content of the Mediterranean Sea would have increasedat a rate of a fewW/m2 during the last decades, in agreement with theincrement of deep water temperature reported by different authors(López-Jurado et al., 2005; Font et al., 2006).

For the seasonal cycle we have found that for the shortwaveincoming radiation QS the cycle accounts for 75% of the variabilitycomputed from daily values. Conversely, for the longwave radiationQB the seasonal cycle could not be statistically significant, since it onlyaccounts for 1% of the daily variability. For the turbulent fluxes QE andQH the percentages are 20% and 10%, respectively.

The amplitude obtained for the total cycle is about 164 W/m2,which is an intermediate value of the amplitudes obtained by Garrettet al. (1993). They obtained 140 W/m2 when they used the option ofdecreasing the solar radiation term and 180W/m2when they followedthe option of increasing the evaporation with respect to previousparameterizations. In all cases the semi-annual cycle is much weakerand its impact on the annual harmonic cycle can be neglected. Thephase obtained for the total heat flux is also in agreement with thatobtained by Garrett et al. (1993), peaking by middle June.

The results obtained for the seasonal cycle of the total heat budgetcan also be compared with the steric sea level cycle. This can beinferred from tide gauge data and has been described in severalprevious works (e.g., García-Lafuente et al., 2004; Fenoglio-Marc et al.,2006; García et al., 2006). To carry out the comparison, the heat fluxmust be integrated in time, in order to obtain the ocean heat content.The steric component is the response of the ocean to changes in theheat content per unit area H (in J/m2) and the response will depend onhow the heat distributes along the water column. The heat flux Q isthe time derivative of the heat content: Q=dH /dt. If we assume aharmonic function for the annual cycle of Q, then H will also have aharmonic shape, but delayed k/2 (3 months) with respect to H. Inpractice, this implies to shift forward the harmonic function fitted tothe total heat budget (shown in the last panel of Fig. 4) by k/2(3 months). The result is that the ocean heat content would peak bymiddle September and reach a minimum value by middle March, inagreement with previous works. Conversely, the spatial distribution ofthe amplitudes of the total heat budget (Fig. 5) is rather different fromthe amplitudes of the steric sea level cycle reported by García et al.

(2006). These authors obtained maximum values in the Levantinebasin, Tyrrhenian and Ionian seas and in the Ligurian–Balearic basin,and minimum values in the Adriatic and Aegean Sea. In our case wehave obtained maximum amplitudes in the Aegean Sea and minimumvalues in the WM, in particular in the Alborán Sea. This indicates thekey role of horizontal advection in relocating the heat contents of theannual cycle. The forthcoming use of a baroclinic model forced by theHIPOCAS heat fluxes will hopefully solve this question. The resultsdescribed above on the comparison between heat fluxes and the stericsea level can be considered as a first qualitative attempt that can beuseful for future deeper analysis.

We have characterized the interannual variability through a modaldecomposition based on the computation of the leading spatial EOFsand their corresponding temporal amplitudes. To our knowledge, this isfirst time that a modal decomposition of the four components of theheat budget is attempted and therefore no comparison can be madewith previous works. Similarly to the analysis of the atmospheric con-tribution to sea level (Gomis et al., this issue) the EOFs have shown to beconsistent when the decomposition has been undertaken in differentdomains (the whole Mediterranean, and the Western and EasternMediterranean basins). The similarity holds regarding the existence of abasin-wide leading EOF. However, for the heat flux component thatleading EOF accounts for a smaller fraction of the variance than for theatmospheric contribution to sea level. Namely it accounts for 36.8%,35.7%, 21.2% and 15.6% for QH, QE, QB and QS, respectively.

A surprising feature is that the amplitudes of the leading EOFs arenot significantly correlated with climatic indices such as the NAOindex or the MOI. Only the correlation between the amplitudes of thetwo radiation terms QS and QB are the winter MOI index are above 0.5.However the correlation values increase when they are computed forindividual gridpoints, the highest values being obtained for thewinterperiod and the MOI index.

The shortwave radiation term QS is negatively correlated with theMOI practically over the whole domain (Fig. 8), reaching values of 0.8in the NE sector of the basin and lower values in thewestern basin. Forthe NAO the correlation is positive, though the values are smaller. Thisis consistent with the fact that the Atlantic storm track shifts to thenorth during a positive NAO phase, leaving the Mediterranean areafree of clouds. The correlation pattern obtained for the longwaveradiation QB is almost opposite to that of QS: positive correlation withthe MOI (except in thewestern basin) and negativewith the NAO. Thisis, during positive phases of the NAO the oceanwould be loosing moreheat in form of longwave radiation, likely in response to the largeincoming shortwave radiation.

For the turbulent terms QH and QE we have obtained weakercorrelations, indicating that turbulent heat fluxes would be moreinfluenced by local phenomena (less by the large-scale atmosphericpattern) than the two radiation terms. For both terms correlations arenegative in the WM and positive in the EM (opposite sign for thecorrelationswith the NAO index), with a clear NW–SE gradient (Fig. 8).However, for QH correlations are weaker in the WM than in the EM,whereas for QE correlations are weaker in the EM than in the WM.

Finally, concerning the evaluation of the HIPOCAS fluxes, theyshow stronger correlations with the observation based NOC fieldsthan are obtainedwith the original NCEP/NCAR fluxes for the full set ofinterannually varying heat flux estimates. Thus, the downscaling hasled to an improved representation of the interannual variability whencompared with observations. We plan to extend this analysis furtherusing data from research buoys to fully quantify the increase inaccuracy obtained through the downscaling but reserve this analysisfor a separate paper.We also note that preliminary investigations haveshown that major atmospheric forcing events such as the severecooling of winters 1991–1992 and 1992–1993 which played a key rolein the Eastern Mediterranean Transient (Josey, 2003) are apparent inthe HIPOCAS fields and that an analysis of these is underway whichwill also be reported separately.

272 S. Ruiz et al. / Global and Planetary Change 63 (2008) 258–274

As a general conclusion of this work, we can state that we havedrawn a consistent picture of the heat budget in the MediterraneanSea. The obtained results are in agreement with previous worksexcept in some figures, as the lower radiation term and the lower (inabsolute value) latent heat term. A consideration in favour of theHIPOCAS data set with respect to previous estimates is thecoherence: HIPOCAS data do not contain any artificial jump derivedfrom changes in the parameterizations. Moreover, not only keyvariables such as 2-m temperature, 10-m wind field and precipitationare better resolved. The better characterization of the land-sea maskby the high resolution hindcast ensures a better representation oflocal aspects related to thermal and dynamical effects. Finally, surfaceatmospheric dynamics are better reproduced by HIPOCAS than byglobal reanalysis.

The potential of the HIPOCAS heat fluxes to investigate sea levelvariability is promising. After the present overview, one could focus forinstance on examining the interannual variability of specific processin which the heat fluxes play a key role, such as the deep water for-mation. Also detailed studies on the interannual variability of theseasonal cycle or the occurrence of extreme events are worth attempt-ing. However, the major step forward will be to use the surface fluxes(altogether with precipitation/evaporation and other forcing variables)to carry out a 44 year run of a baroclinic model with the aim of re-producing and quantifying the long-term steric variability of sea level.

Acknowledgments

We gratefully acknowledge the comments and computational helpprovided by Drs. M. Marcos, G. Jordà, A. Pascual and B. Garau. We alsothank the anonymous reviewers of this work and in particularDr. Mikis Tsimplis for their constructive comments. This study hasbeen carried out in the framework of the VANIMEDAT project(CTM2005-05694-C03/MAR) funded by the Spanish Marine Scienceand Technology Program. The HIPOCAS data set has been provided byPuertos del Estado and is available to the scientific communitythrough that institution. The code used for EOF computation has beenkindly provided by Vincent Toumazou.

Appendix A. Parameterization of the four heat flux components

A.1. Longwave radiation QB

The long-wave radiation, that finally produces the net longwaveradiation parameter (QB), is derived in themodel from the upward anddownward components of the monochromatic flux Fν at wavenumberν, assuming a non-scattering atmosphere:

Fum ¼ Bm Tsð Þ −Bm Tað Þ½ �tm ps;p; rð Þ þ Bm Tp� �

−Z p

pstm p V;p; rð ÞdBm ðA:1Þ

Fdm ¼ Bm Ttop� �

−Bm T∞ð Þ� �tm p;0; rð Þ þ Bm Tp

� �−Z 0

pstm p; p V; rð ÞdBm ðA:2Þ

where Bν is the Planck function, T is the temperature refers to surface(s), surface air (a), specific pressure level (p) and top of the atmosphere(top) with the respective subscripts, tν(p,p′;r) is the monochromatictransmission function of the flux between pressure levels p and p′,evaluated in a direction θ (angle with the vertical) such r=secθ is thediffusivity factor. Eqs. (A1) and (A2) are integrated for different wavenumbers, dividing the longwave spectrum into six spectral regionscorresponding to centres of absorption bands related to atmosphericwindows, and the bands associated to rotation and vibration-rotationof water, CO2 and O3 absorption. The aforementioned extra minortrace gases effects are also included.

The treatment of clouds follows the method of Washington andWilliamson (1977), calculating separately for every level the fluxes

for clear and overcast skies and later proportionally combine infunction of the total cloud amount. Some factors that play a keyrole in the different longwave cloud optical properties, such asdroplet absorption, different liquid/ice absorption, droplet numberconcentration are taken into account, resulting in a differenttreatment of ice/warm, low/high altitude level and continental/maritime clouds.

A.2. Shortwave radiation QS

The shortwave radiation scheme integrates the fluxes (wavenumber comprises between nu=0.2 and nu=4 m) resolving thefollowing transfer equation for diffuse radiation (Lν) in a directiongiven by the azimuth angle ϕ and μ=cos ϑ, being ϑ the zenith angle:

μddδ

Lm δ; μ;/ð Þ½ � ¼ Lm δ; μ;/ð Þ − ωm δð Þ4

P1 − P2f g ðA:3Þ

with

P1 ¼ Pm δ; μ;/; μ0;/0ð ÞF 0m e

−δ=μ0 ðA:4Þand

P2 ¼Z 2k

0

Z 1

−1Pm δ; μ;/; μ V;/ Vð ÞLm δ; μ V;/ Vð Þdμ Vd/ V ðA:5Þ

where Fv0 is the solar irradiance in the direction μ0, δ the optical depth,

ωv the single scattering albedo and Pm δ; μ;/; μ V;/ Vð Þthe scatteringphase function, which defines the probability that radiation comingfrom the direction (μ V;/ V) is scattered in the direction (μ;/).

Solar radiation is attenuated by absorbing gases, mainly watervapour, carbon dioxide, oxygen and ozone, as well as scattered bymolecules, aerosols and cloud particles. Assuming an atmospheredivided into N homogeneous layers, the upward and downward fluxesat a layer interface j can be modelled using the reflectance at the layertop and the transmittance at the bottom.

Likewise, total optical thickness (δ), total single scattering albedo ofthe layer (ω) and total asymmetry factor (g) are calculated taking intoaccount the given optical thickness of the cloud (δc), the aerosol (δa),and the molecular absorption (δg).

δ ¼ δc þ δa þ δgω ¼ δc þ δað Þ=δc þ δa þ δgg ¼ gcδc= δc þ δað Þ þ gaδa= δc þ δað Þ

ðA:6Þ

The shortwave cloud optical properties are considered, being thesingle scattering properties determined on the basis of Mie calcula-tions using idealized size distribution for both cloud droplets andspherical ice crystals (Rockel et al., 1991). Since it is well known thatMie theory tends to overestimate the asymmetry factor for ice clouds(Stephens et al., 1990), a correction factor of 0.91 has been applied toadjust gi to a more realistic value for a wide range of effective radii ofcloud and ice crystals (Francis et al., 1994).

Finally, it is worth to mention that the surface background albedoused in the run was derived from the dataset compiled by Claussenet al. (1994). An albedo value of 0.07 is assigned to all water surfaces,whereas the snowand ice surfaces albedo is a function of temperature.Over land, fractional forest areas as described in Roeckner et al. (1992)are taken into account.

A.3. The turbulent fluxes: QE and QH

The surface heat fluxes are calculated as a turbulent flux of avariable χ in the lowest model level, according to a bulk transferrelation:

w Vχ V� �

S¼ −CχjVLj χL −χSð Þ ðA:7Þ

273S. Ruiz et al. / Global and Planetary Change 63 (2008) 258–274

where Cχ is the transfer coefficient. The subscripts L and S refer tovalues at the lowest model level and the surface, respectively. VL is thehorizontal wind speed at this lowest level. The transfer coefficient iscalculated using the Monin–Obukhov similarity theory, being the Cχdependant of the existing atmospheric stability conditions. Thecoefficient for momentum and heat are calculated following theapproximate analytical expression proposed by Louis (1979):

Cm;h ¼ CN � fm;h RiB;zLz0m

þ 1;zLz0h

þ 1� �

ðA:8Þ

where CN is the transfer coefficient for neutral conditions that isexpressed as follows:

CN ¼ k2

ln zLz0m

þ 1�

� ln zLz0h

þ 1� ðA:9Þ

being k the Von Karman constant, zL the height of the lowest modellevel, z0m and z0h the roughness length for momentum and heat,respectively.

The stability functions for momentum and heat (fm and fh), thatrepresents the ratio of Cm,h in relation to the values of CN underneutral conditions, are calculated also following Louis (1979). Themoist bulk Richardson number of the surface layer (RiB), calculated interms of cloud conservative variables such as the total water contentand the liquid water potential temperature, provides a stability index.

Roughness length is specified over land as a function of thesubgrid-scale orography and vegetation (parameters taken from thephysiographic database compiled by Claussen et al (1994), whereasover open waters is computed from the Charnock formula:

z0m ¼ max 0:032u24=g;1:5x10

−5�

ðA:10Þ

being u⁎ the friction velocity and g the acceleration of gravity.For the transfer of heat and vapour over sea, this Charnock relation

is modified following the suggestion of Large and Pond (1982). Theirempirical study, based on observational data, showed that the transfercoefficients for heat and water vapour are quite independent of windspeed, suggesting the following empirical formula to calculate it insubstitution of a Charnock formula

z0h ¼ z0mexp 2−86:276z0:3750m

� ðA:11Þ

Parameterization of surface process is carried out by means of a5-layer soil model that is included in order to take into account heat andwater budgets within the soils (DKRZ, 1994). With regards to the seasurface temperature, the data set used as boundary condition along the44-yr REMO run comes from the one used in the NCEP reanalysis. Theanalysis and climatologies used in the generation of such NCEP SSTboundary field were the following ones: starting from 1982, whenAVHRR (Advanced Very High Resolution Radiometer) data becameavailable, it was used an optimal interpolation of Reynolds SST re-analysis (Reynolds and Smith, 1994). From earlier periods, it was usedthe UKMO global ice and SST (GISST) reanalysis (Parker et al., 1993).

References

Andersson, E., et al., 1992. LAPACK User's Guide. SIAM. 325 pp.Bethoux, J.P., 1979. Budgets of the Mediterranean Sea: their dependence on the local

climate and on the characteristics of Atlantic waters. Ocean. Acta 2, 157–163.Bunker, A.F., Charnock, H., Goldsmith, R.A., 1982. A note on the heat balance of the

Mediterranean and Red Seas. J. Mar. Res. 40, 73–84 Suppl.Castellari, S., Pinardi, N., Leaman, K., 1998. A model study of air–sea interactions in the

Mediterranean Sea. J. Mar. Syst. 18, 89–114.Chen, C.T., Roeckner, E., 1996. Cloud simulation with the Max Planck Institute for

Meteorology general circulation model ECHAM4 and comparison with observa-tions. Max Planck Institute fur Meteorologie, Report N 193. Hamburg. 49 pp.

Chen, C.T., Roeckner, E., Soden, B.J., 1996. A comparison of satellite observations andmodel simulations of column-integrated moisture and upper tropospherichumidity. J. Climate 9 (7), 1561–1585.

Claussen, M., Lohmann, U., Roeckner, E., Schulzweidad, U., 1994. A global data set of landsurface parameter. Max Planck Institute fur Meteorologie, Report N 135, Hamburg.34 pp.

DKRZ-Modelling group, 1994. The ECHAM3 atmospheric General circulation model.Technical Report 6, Deutsches KlimaRechenZentrum, Hamburg, Germany.

Doms, G., et al., 1995. Documentation of EM/DM-Systems. Deutscher Wetterdienst,Technical Report. Offenbach, Germany.

Fenoglio-Marc, L., Kusche, J., Becker, M., 2006. Mass variations in the Mediterranean Seafrom GRACE and its validation by altimetry, steric and hydrologic fields. Geophys.Res. Lett. 33, L19606, doi:10.1029/2006GL026851.

Font, J., Salat, J., Emelianov, M., Puig, P., Palanques, A., Julià, A., Lopez-Jurado, J.L.,González-Pola, C., Flos, J., 2006. A new DeepWater formed in the NW Mediterra-nean in 2005. Geophys. Res. Abstr. 8, 03081.

Fouquart, Y., Bonnel, B., 1980. Numerical prediction of convectively driven mesoescalepressure systems. Part I. Convective parameterization. J. Atmos. Sci. 37, 1722–1733.

Francis, P.N., Jones, A., Saunders, R.W., Shine, K.P., Slingo, A., Sum, Z., 1994. Anobservational study of the radiative properties of cirrus: some results from ICE'89.Quart. J. Roy. Meteor. Soc. 120, 809–848.

García, D., Chao, B.F., Del Río, J., Vigo, I., 2006. On the steric and mass-inducedcontributions to the annual sea level variations in the Mediterranean Sea. J.Geophys. Res. 101, C09030, doi:10.1029/2006JC002956.

García-Lafuente, J., Del Río, J., Alvarez-Fanjul, E., Gomis, D., Delgado, J., 2004. Someaspects of the seasonal sea level variations around Spain. J. Geophys. Res. 109,C09008, doi:10.1029/2003JC002070.

Garrett, C., Outerbridge, R., Thompson, K., 1993. Interannual variability in Mediterra-nean heat and buoyancy fluxes. J. Climate 6, 900–910.

Gilman, C., Garrett, C., 1994. Heat flux parameterization for the Mediterranean Sea: therole of atmospheric aerosols and constraints from thewater budget. J. Geophys. Res.99 (C3), 5119–5134.

Golub, G.H., Kahan, W., 1965. Calculating the singular values and pseudo-inverse of amatrix. SIAM J. Numer. Anal. 2, 205–224.

Golub, G.H., van Loan, C., 1996. Matrix Computations, 3d ed. Johns Hopkins UniversityPress. 694 pp.

Gomis, D., Ruiz, S., Sotillo, M.G., Álvarez-Fanjul, E., Terradas, J., accepted, this issue. Lowfrequency Mediterranean sea level variability: the contribution of atmosphericpressure and wind.

Gomis, D., Tsimplis, V., Martín-Míguez, B., Ratsimandresy, A.W., García-Lafuente, J.,Josey, S.A., 2006. Mediterranean Sea level and barotropic flow through the Strait ofGibraltar for the period 1958–2001 and reconstructed since 1659. J. Geophys. Res.111, C11005, doi:10.1029/2005JC003186.

Jacob, D., Podzun, R., 1997. Sensitivity studies with the regional climate model REMO.Meteor. Atmos. Phys. 63, 119–129.

Jones, P.D., Jónsson, T., Wheeler, D., 1997. Extension to the North Atlantic Oscillationusing early instrumental pressure observations from Gibraltar and south-westIceland, Int. J. Climatol. 17, 1433–1450.

Josey, S.A., 2001. A comparison of ECMWF, NCEP/NCAR and SOC surface heat fluxes withmoored buoy measurements in the subduction region of the North–East Atlantic. J.Climate, 14 (8), 1780–1789.

Josey, S.A., 2003. Changes in the heat and freshwater forcing of the EasternMediterranean and their influence on deep water formation. J. Geophys. Res. 108(C7), 3237, doi:10.1029/2003JC001778.

Josey, S.A., Kent, E.C., Taylor, P.K., 1999. New insights into the ocean heat budget closureproblem from analysis of the SOC air–sea flux climatology. J. Climate 12 (9),2856–2880.

Josey, S.A., Smith, S.R., 2006. Guidelines for Evaluation of Air–Sea Heat, Freshwater andMomentum Flux Datasets, CLIVAR Global Synthesis and Observations Panel (GSOP)White Paper, July 2006, p. 12.

Kalnay, E., et al., 1996. The NCEP/NCAR 40-year reanalysis project. Bull. Amer. Meteor.Soc. 77, 437–471.

Koepke, P., Hess, M., Schult, I., Shettle, E., 1997. Global Aerosol Data Set. Max-PlanckInstitute fur Meteorologie. Report N 243. Hamburg. 44 pp.

Kondo, J., 1975. Air–sea bulk transfer coefficients in diabatic conditions. Boundary-LayerMeteorol., 9, 91–112.

López-Jurado, J.L., González-Pola, C., Vélez-Belchí, P., 2005. Observation of an abruptdisruption of the long-term warming trend at the Balearic Sea, WesternMediterranean Sea, in summer 2005. Geophy. Res. Lett. 32, L24606, doi:10.1029/2005GL024430.

Large,W.G., Pond, S., 1982. Sensible and latent heat fluxmeasurements over the ocean. J.Phys. Oceanogr. 12, 464–482.

Louis, J.F., 1979. A parametric model of vertical eddy fluxes in the atmosphere. BoundaryLayer Meteorological 17, 187–202.

MacDonald, A.M., Candela, J., Bryden, H.L., 1994. In: LaViolette, P.E. (Ed.), An Estimate ofthe Net Heat Transport Through the Strait of Gibraltar. . Seasonal and Interannualvariability of the Western Mediterranean Sea, Coastal Estuarine Stud., vol. 46. AGU,Washington, D.C., pp. 12–32.

May, P.W., 1986. A Brief Explanation of Mediterranean Heat and Momentum FluxCalculations. NORDA code 322, NSTL, MS 39529.

Morcrette, J., Smith, L., Fouquarat, Y., 1986. Pressure and temperature dependence of theabsorption in long wave radiation parameterizations. Beitr. Phys. Atmos. 59,455–469.

Parker, D.E., Folland, C.K., Ward, M.N., Maskell, K., Bevan, A., Jackson, M., 1993. Ocean–atmosphere climatic fluctuations on interannual to century time scales. Publica-tion of the Hadley Centre for Climatic Prediction and Research UKMO, Brackneel,Berks.

Ratsimandresy, A.W., Sotillo, M.G., Carretero, J.C., Álvarez, E., Hajji, H., in press-a. A 44-year-high-resolution ocean and atmospheric hindcast for the Mediterranean Basin

274 S. Ruiz et al. / Global and Planetary Change 63 (2008) 258–274

developed within the HIPOCAS Project. Coastal Engineering. special issue onHIPOCAS Project (in press).

Ratsimandresy, A.W., Sotillo, M.G., Álvarez, E., Carretero, J.C., Pérez, B., Hajji, H., inpress-b. A44-year (1958–2001) sea level residual hindcast over the Mediterranean Basinwithinthe HIPOCAS Project framework. Physics and Chemistry of the Earth (in press).

Reynolds, R.W., Smith, T.M., 1994. Improved global sea surface temperature analysisusing optimal interpolation. J. Climate, 7, 929–948.

Rockel, B., Raschke, E., Weyres, B., 1991. A parameterization of broad band radiativetransfer properties of water, ice and mixed clouds. Beitr. Phys. Atmos. 64, 1–12.

Roeckner, E., Arpe, K., Bengtsson, L., Brinkop, S., Dumenilm, L., Esch, M., Kirk, E., Lunkeir,F., Ponater, M., Rockel, B., Sausen, R., Schlese, U., Schubert, S., Windelband, M., 1992.Simulation of the present day climate with the ECHAM model: Impact of modelphysics and resolution. Max-Planck-Institute fur Meteorologie, Report N 93.Hamburg. 171 pp.

Roeckner, E., Arpe, K., Bengtstsson, L., Chistoph, M., Claussen, M., Dumenil, L., Esch, M.,Giorgetta, M., Schlese, U., Schulzweide, U., 1996. The atmospheric generalcirculation model ECHAM4: model description and simulation of present dayclimate. Technical Report 218. Max Plank Institute fürMeterologie, Hamburg, p.102.

Schiano, M.E., Santoleri, L., Bignami, F., Leonardo, R.M., Marullo, Bohm, 1993. Air–seainteractions measurements in the best Mediterranean Sea during the ThyrrhenianEddy Multi-Platform Observations Experiments. J. Geophys. Res. 98 (C2), 2461–2474.

Sotillo, M.G., 2003. High-resolution multi-decadal atmospheric reanalysis in theMediterranean Basin. PhD Dissertation. Universidad Complutense deMadrid, Madrid,pp 209.

Sotillo, M.G., Ratsimandresy, A.W., Carretero, J.C., Bentamy, A., Valero, F., González-Rouco, F., 2005. A high-resolution 44-year atmospheric hindcast for theMediterranean Basin: contribution to the regional improvement of globalreanalysis. Clim. Dyn, doi:10.1007/s00382-005-0030-7.

Sotillo, M.G., Martín, M.L., Valero, F., Luna, Y., 2006. Validation of a homogeneous 41-year (1961–2001) winter precipitation hindcasted dataset over the IberianPeninsula: assessment of the regional improvement of global reanalysis. ClimateDyn, doi:10.1007/s00382-006-0155-3.

Sotillo, M.G., Aznar, R., Valero, F., 2007. The 44-year Mediterranean HIPOCAS wind database: A useful tool to analyze offshore extreme wind events from a long-termregional perspective. Coastal Engineering. special issue on HIPOCAS Project(in press).

Stephens, G.L., Tsay, S.C., Stackhouse, P.W., Flateau, P.J., 1990. The relevance of themycrophysical and radiative properties of cirrus clouds to climate and climatefeedback. J Atm. Sci. 47, 1742–1753.

Supic, M., Grbec, B., Vilibic, I., Ivancic, I., 2004. Long-term changes in hydrodynamicalconditions in Northern Adriatic and its relationship to hydrological and atmo-spheric processes. Ann. Geophys. 22 (3), 733–745.

Sušelj, K., Bergant, K., 2006. Mediterranean oscillation index. Geophys. Res. Abstr. 8,02145 European Geosciences Union.

The MathWorks, 1992. MATLAB Reference Guide. The MathWorks Inc., Natick, MA.548 pp.

Toumazou, V., Cretaux, J.F., 2001. Using a Lanczos Eigensolver in the computational ofempirical orthoghonal functions. Mon. Weather Rev. 129 (5), 1243–1250.

Tsimplis, M., Josey, S.A., 2001. Forcing of the Mediterranean Sea by atmosphericoscillations over the North Atlantic. Geophys. Res. Lett., 28 (5), 803–806.

Tsimplis, M.N., Zervakis, V., Josey, S., Peneva, E., Struglia, M.V., Stanev, E., Lionello, P.,Malanotte-Rizzoli, P., Artale, V., Theocharis, A., Tragou, E., Oguz, T., 2006. Changes inthe oceanography of the Mediterranean Sea and their link to climate variability. In:Lionello, P., Malanotte-Rizzoli, P., Boscolo, R. (Eds.), Mediterranean ClimateVariability. Elsevier. ISBN: 0-444-52170-4, p. 438.

Von Storch, H., Langenberg, H., Feser, F., 2000. A spectral nudging technique fordynamical downscaling purposes. Mon. Weather. Rev. 128, 3664–3673.

Washington, W.M., Williamson, D.L., 1977. A description of the NCAr GCM's. In: Chang, J.(Ed.), General Circulation Models of the Atmosphere. . Methods Comput. Phys., 17.Academic Press, pp. 111–172.

Wild, M., Ohmura, A., Gilgen, H., Roeckner, E., Giorgetta, M., 1996. Improvedrepresentation of surface and atmospheric radiation budgets in the ECHAM4general circulation model. Max Planck Institute fur Meteorologie, Report N 200,Hamburg. 32 pp.