Diurnal coastal upwelling was previously observed when sea breezes were exceptionally strong, or when the process occurred close to critical latitudes (30°N, 30°S) where local inertial oscillations may be resonantly excited. Our data collected in the Adriatic show that the pronounced diurnal upwelling is also possible under milder wind-forcing and outside critical latitudes. It is found that the thermocline recorded in the summer of 2006 at the south coast of the island of Lastovo was subject to diurnal variability with a maximum range of about 30 m, and that the corresponding currents measured off the west coast of the island pointed to internal waves propagating around the island in a clockwise direction. We suggest that the summertime stratification occasionally promotes coastal waves that revolve daily around the island, creating the conditions needed for resonant excitation by sea breezes. Numerical modeling reveals that the 24-h waves are trapped around the island due to the influence of both the Coriolis force and bottom slope, and that the 12-h waves radiate away from the island. The biogeochemical data show that the diurnal upwelling may stimulate primary production in the area but may also adversely affect benthic organisms.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Upwelling occurs in oceans and seas when horizontal divergence in the surface layer creates water deficit that is filled by water brought up from a subsurface layer [Barber, 2001]. The divergence is mainly caused by winds, either open ocean or coastal. Winds blowing at a coast produce upwelling by forcing offshore transport in the surface layer, compensating onshore transport in a subsurface layer, and ascending motion close to the coast. Because the subsurface water is rich in nutrients, coastal upwelling, which brings those nutrients to the sunlit surface layer, supports intense organic productivity and is of great economic importance: though the areas occupied by upwelling constitute less than 1% of the world's oceans, they account for about 20% of the global fish catch [Pauly and Christensen, 1995]. The subsurface water transported to the surface layer is relatively cold and therefore coastal upwelling is also recognized as an important link in the processes affected by increased concentrations of greenhouse gases: due to global warming upwelling is presently intensifying [Bakun, 1990; Roemmich and McGowan, 1995]—although not everywhere [McGregor et al., 2007]—and is expected to feedback on the atmosphere and to influence the drawdown of atmospheric CO2.
 Early studies of coastal upwelling were mainly based on classical hydrographical surveys of four eastern boundary currents and one western boundary current found in subtropical ocean regions: the Canary Current off northwestern Africa, the Benguela Current off southwestern Africa, the California Current off western North America, the Peru Current off western South America, and the Somali Current off the Arabian Peninsula [Smith, 1968]. Upwelling in the Arabian Peninsula region is related to strong summer monsoon winds and is therefore highly seasonal; in the other four regions, upwelling is rather stable in lower latitudes and more seasonal in higher latitudes. Subsequent investigations capitalized on the then-new technology of continuous current and temperature measurements to reveal superseasonal, day-to-day variability of upwelling, and the term “upwelling event” came into use [Smith, 1981; Codispoti, 1981]. Long-term hydrographical measurements and analyses of sediment cores taken beneath upwelling systems made it possible to track subseasonal, i.e., interannual and climatic variability [Roemmich and McGowan, 1995; Summerhayes et al., 1995; McGregor et al., 2007].
 The collection of time series of currents and temperature enabled the documentation of not only day-to-day but also diurnal variability of coastal upwelling. The analyses of the latter were undertaken relatively slowly, probably because the diurnal wind-driven signal can be masked by the diurnal tidal signal. When empirical studies of diurnal upwelling started, they concentrated more on current variability [Rosenfeld, 1988; DiMarco et al., 2000; Lerczak et al., 2001; Rippeth et al., 2002; Woodson et al., 2007; Hunter et al., 2007; Zhang et al., 2009] than on temperature variability [Kaplan et al., 2003; Woodson et al., 2007; Zhang et al., 2009]. The studies were carried out not only in traditional regions where coastal upwelling occurs but also in some new regions (Texas-Louisiana shelf, Catalonian shelf, New York Bight). Analytical and numerical modeling developed at the same time as this empirical activity. Diurnal winds were supposed to blow over the sea bounded by a straight coast, and the sea was assumed to be homogeneous [Rosenfeld, 1988], two-layered [Rippeth et al., 2002] or continuously stratified [Lerczak et al., 2001]. The picture that emerged from these studies was of pronounced diurnal upwelling related either to strong sea breezes that develop due to temperature differences between the sea and land or to inertial oscillations resonantly excited close to critical latitudes (30°N, 30°S).
 A recent experiment in the Adriatic Sea revealed that the pronounced diurnal upwelling is also possible under milder wind-forcing and outside critical latitudes. We aim in this paper to present empirical findings based on the Adriatic data set and to interpret them using analytical and numerical modeling.
 A portion of the Adriatic experiment relevant for the study of diurnal upwelling was carried out at stations shown in Figures 1a and 1b between February and September 2006. During the experiment (1) acoustic Doppler current profiler (ADCP) measurements were performed at the station shown using a trawl-resistant bottom mount, (2) sea temperature was recorded with a series of sensors deployed along a steep cliff on the south side of the island of Lastovo, (3) sea level was monitored at the permanent Dubrovnik tide-gauge station, and (4) meteorological conditions were documented by routine stations in the area (Dubrovnik and Vis).
 We used an RDI Workhorse Sentinel ADCP operating at 300 kHz with a 15 min sampling interval for current measurements. The bin size was 3 m and the bottom depth 95 m, implying 28 measurement layers that extended from the 8.4 to 89.4 m depths. While the current measurements were performed following standard practice in oceanography, the temperature measurements were done in a somewhat unusual way: thermistors were not deployed on a mooring but instead on a steep island cliff. Because fishing activities in the area would limit mooring use to a month or two, this approach was necessary to enable data collection for the entire study period. As it turned out, practical necessity proved to be scientifically beneficial, because cliff measurements provided information on temperature variability at the coast where the diurnal upwelling is most pronounced rather than offshore where the variability is attenuated. This measurement strategy has rarely been used, and if so, it has been utilized primarily to document the internal tides [Wolanski et al., 2004]. We used TidbiT type thermistors to record sea temperature with a 10 min temporal resolution and an accuracy of ±0.2°C at 21°C. Ten sensors were distributed between depths of 4 and 40 m with a 4 m vertical resolution, along an almost vertical cliff reaching to the 80 m depth. All sensors were successfully recovered at the end of the stratified season.
 The first step in the data analysis was to remove barotropic tides from the current data by calculating vertically averaged currents, performing harmonic analysis on the averages [Pawlowicz et al., 2002] and subtracting the synthesized tidal signal from the data measured at each level in order to obtain the residual current. The procedure appears to be suitable in our case of weak barotropic tidal currents (at most 3 cm/s and thus contributing not more than 20% to the total current variance), in a rather deep sea (ca. 100 m), with the surface and bottom boundary layers (extending over some 8 and 5 m, respectively) being excluded from the analysis; when applied on stronger currents in a shallower sea the procedure may result in the boundary layer structure related to barotropic tides to be mistakenly interpreted as the internal tidal variability [Edwards and Seim, 2008]. From the detided current time series and the simultaneous temperature data we computed the spectra [Jenkins and Watts, 1968] for the whole measurement interval and for all the depths (Figures 1c and 1d). There is considerable low-frequency variability visible in the spectra, probably related to the changing east-coast inflow to the Adriatic and to the synoptic-scale wind events. Higher frequencies are dominated by inertial oscillations of a 17.6-h period and by diurnal oscillations comprising 24-, 12- and 8-h periods. The peaks are significant at the 95% level at the depths at which they attain maxima (not shown). The temperature variability is largest at a depth of 10–25 m, i.e., at the thermocline level, while the currents are surface- and bottom-intensified, which implies their baroclinicity.
 There are two obvious causes of the diurnal oscillations: tidal forcing and wind-forcing. We performed a wavelet analysis [Torrence and Compo, 1998] to distinguish between the two. In this analysis tidal forcing was represented by sea level measured at Dubrovnik and by barotropic currents recorded at station ADCP, whereas wind-forcing was represented by data collected at meteorological stations Dubrovnik and Vis. Temporal variability of the two forcing mechanisms was compared to fluctuations of the 20°C isotherm, roughly corresponding with the thermocline at Lastovo. Figure 2a shows that the beats of 24-h tides are visible in both the sea level and current time series and that they closely followed each other. In June and late July 2006, when the tidal forcing was strongest, the 24-h isotherm variability was apparently related to it. Two-input one-output cross-wavelet analysis, documenting relationship between barotropic tidal currents and wind stress on one hand and thermocline variability on the other, confirmed the finding and made it possible to consider internal tides in some detail [Mihanović et al., 2009]. In the middle of July 2006, however, when the 24-h isotherm oscillations were most pronounced, the tidal forcing was at a minimum. Figure 2b shows that the 24-h wind variability culminated at the time, strongly suggesting that the wind-forcing was responsible for the maximum diurnal oscillations of isotherm elevation. Figure 2c reveals that the typical range of thermocline variability is about 10 m, but it increased to 30 m in mid-July 2006.
 A note on the wind field is in order here. Most useful for the present analysis would be wind data recorded on the island of Lastovo. Since, however, such data were not available, we were forced to use time series originating from the nearby anemographic stations, Dubrovnik and Vis. The observed time series could be compared with the predictions obtained by Aladin, a mesoscale meteorological model run by the Croatian Meteorological and Hydrological Service, not only for Dubrovnik and Vis but also for Lastovo. The model results suggested that the diurnal wind variability at Lastovo, being pronounced not only in the NE-SW direction but also in the NW-SE direction, resembles more closely that at Vis than in Dubrovnik, reflecting the fact that Lastovo and Vis are open-sea locations whereas Dubrovnik is a coastal station. On the other hand, the model also indicated that the long-term modulation of diurnal winds at Lastovo is similar to the modulation in Dubrovnik while it differs from the modulation at Vis, probably due to a curvature of the coastline to the north of Lastovo (Figure 1). We have therefore decided to use the wind data originating from both Dubrovnik and Vis in the comparison of meteorological and oceanographic time series. The meteorological model results were not utilized in the comparison because the analysis of model errors, which would be a necessary prerequisite for the utilization, was beyond the scope of the present paper.
 The wavelets shown in Figure 2 suggested that the interval from 14 to 19 July 2006 was the most significant for diurnal upwelling analysis. Temperature data collected over that time interval at the south coast of the island of Lastovo are depicted in Figure 3c. The range of thermocline variability increased over the interval, suggesting resonant excitation. In the days with the largest variability, the thermocline was lowest at about 6 a.m. and highest at about 4 p.m. with a secondary maximum occurring at about 10 p.m. As shown in Figure 3b, the diurnal thermocline oscillations were phase-locked to the sea breezes revolving clockwise throughout the day; a relatively weak morning breeze, blowing in the open sea (station Vis) from the southeast, corresponded to the thermocline reaching its deepest point, and a relatively strong afternoon breeze, blowing in the open Adriatic from the northwest, coincided with the thermocline being close to the sea surface. The sea breezes were superimposed on a quasi-steady airflow directed southwestward at the coast (station Dubrovnik) and southeastward in the open sea (Figure 3a); the latter combined with the sea breeze in the afternoon to produce a pronounced wind known locally as maestral. In Figure 3d, the elevations of the 20°C isotherm are compared with the northwestward currents measured off the west coast of the island of Lastovo. It is clear that when the thermocline was rising the currents tended to flow northwestward in the surface layer and to be of opposite direction in the bottom layer, and that the current directions were reversed when the thermocline was falling, which confirms the baroclinic nature of the process.
 To illustrate the relationship between various parameters more vividly, their mean daily course was determined for the 14–19 July 2006 interval and is shown in Figure 4. It confirms the main characteristics of diurnal variability noticed in the original time series, but also points to some complexities. Thus, for example, the currents averaged in the surface (bottom) layer are directed northwestward (southwards) when the thermocline is rising, and are therefore not exactly opposed to each other during this part of diurnal cycle. This probably signals that the two-layer interpretation should be regarded as approximate and that eventually one should allow for vertical propagation as well. Otherwise, Figure 4 clearly shows that the northwestward surface currents measured at the station located off the west coast of the island of Lastovo coincide with the rising thermocline recorded at the south coast of the island. This convincingly points to a thermocline trough being first recorded at the south and then at the west coast and therefore to an internal wave propagating around the island in a clockwise sense. From the fact that about 6 h is needed for the wave to propagate from the south to the west coast, it may be concluded that the wave circumnavigates the island in about a day and that therefore it may be resonantly excited by a forcing possessing daily periodicity.
 We first used a simple analytical model, which addressed the response of a reduced-gravity f-plane sea bounded by a straight vertical wall to a periodic wind stress, to interpret the empirical findings. The model, originally formulated by Cushman-Roisin , was extended by allowing not only for alongshore but also cross-shore wind-forcing and by considering both subinertial and superinertial frequencies [Orlić and Pasarić, 2011]. Wind data collected at Vis between 14 and 19 July 2006 had been utilized to compute pseudo wind stress, which was then subjected to harmonic analysis. The results, shown in Figure 5, revealed a dominance of wind rotating clockwise at the 24-h period, but also a presence of variability at other periods. Subsequently, wind stress was determined by assuming the drag coefficient equal to 1.5 × 10−3. The other parameters used were the proportional density defect (1.3 × 10−3) and the thermocline depth (20 m) estimated from our temperature time series averaged over the 14–19 July 2006 interval and assuming a uniform salinity equal to 38, as well as the acceleration due to gravity (9.81 m/s2) and the Coriolis parameter (10−4 s−1). The reduced-gravity model then enabled the amplitudes of the thermocline variability to be determined for a number of frequencies and to be compared with the amplitudes of the 20°C isotherm variability as observed on the island of Lastovo between 14 and 19 July 2006. As is obvious from Figure 6a, the model reproduced well the 12-h signal, but underestimated the 24- and 8-h signals and the variability at largest periods. While the discrepancy at large periods is probably due to phenomena that are not related to the diurnal wind-forcing and are therefore not reproduced by the present model, the difficulties encountered at the shorter periods indicate that the model possibly failed to capture a wind-related process relevant for the Adriatic. The fact that diurnal isotherm variability gradually increased between 14 and 18 July 2006 (Figure 3) and that there were indications of an internal wave circumnavigating Lastovo in about a day (Figure 4) suggested that resonance was the missing process and that the assumption of a straight coast should be abandoned.
 The next simplest model assumed that a circular island is embedded in a two-layer f-plane sea. A series of analytical and numerical investigations [Longuet-Higgins, 1969; Wunsch, 1972; Hogg, 1980; Brink, 1999] indicated that in the Northern Hemisphere the model allows internal waves to travel around the island in a clockwise direction at subinertial frequencies and radiate away from the island at superinertial frequencies. If the island is large relative to the internal Rossby radius of deformation [gɛhh′/(h + h′)]1/2/f (where g is the acceleration due to gravity, ɛ is the proportional density defect, h is the upper-layer depth, h′ is the lower-layer depth, and f is the Coriolis parameter), the waves trapped around the island closely resemble internal Kelvin waves. Although the mean radius of Lastovo (5200 m) is close to the Rossby radius of deformation, it is still useful to make a first-order estimate of the time the trapped internal waves need to circumnavigate the island by taking into account the speed of internal Kelvin waves [gɛhh′/(h + h′)]1/2: with the parameters given above and the bottom depth equal to 80 m, the speed equals 0.44 m/s, implying that about 20 h would be needed for the internal waves to travel around the island. The time is smaller than the diurnal period, but—as shown by Longuet-Higgins —an underestimation may be expected since the speed of internal Kelvin waves is larger than the speed of internal waves traveling around a relatively small island. It therefore appears that not only the data analysis but also the simple calculation suggests that the pronounced diurnal upwelling observed at Lastovo between 14 and 19 July 2006 could be interpreted as the internal waves resonantly driven around the island by clockwise-rotating sea breezes. Let it be mentioned that the speed of internal Kelvin waves considerably varied over the 8-month measurement interval. Thus, for example, our temperature time series indicate that the speed was as small as 0.3 m/s in May 2006, and that it surpassed 0.5 m/s in September 2006 due to a deepening of the thermocline and an increase of the proportional density defect. This implies that the stratification supported internal coastal waves traveling around the island in about a day in mid-summer and that the period was larger (smaller) and therefore the conditions not so favorable for resonant excitation by diurnal winds earlier (later) in the year.
 In order to check the possibility of resonant excitation of internal waves by diurnal winds, we used a numerical model—the Princeton Ocean Model (POM). In the earlier numerical modeling studies, the generation of island-trapped waves by tides [Brink, 1999; Wolanski et al., 2004] and by synoptic-scale winds [Brink, 1999; Merrifield et al., 2002] had been considered. Our modeling aimed to demonstrate how a stratified sea in a wider area around Lastovo responds to diurnal wind-forcing. The POM is a three-dimensional, nonlinear model based on the traditional, primitive equations for conservation of momentum, mass, heat and salt, combined with the equation of state [Blumberg and Mellor, 1987]. The model used for the Adriatic study has a 2 km resolution and either a constant depth of 300 m or realistic bathymetry. The bathymetry is based on the 7.5 s resolution depth field derived during the Dynamics of the Adriatic in Real-Time (DART) project at the NATO Undersea Research Centre (NURC) using inverse distance weighted interpolator, with the fine resolution NURC data being bin averaged on the 2 km model grid and smoothed by Shapiro filter. The model starts from the state of rest with the initial vertical temperature profile deduced from our temperature time series averaged over the 14–19 July 2006 interval in the upper part of the water column combined with summertime middle Adriatic climatology [Artegiani et al., 1997] in the lower part and a uniform salinity equal to 38. The model is forced by the wind stress observed at Vis between 13 and 19 July 2006; the day before our interval of interest is used to spin up the model, with the wind stress linearly increasing from zero to the observed value over an inertial period. The elevations of the 20°C isotherm computed for the interval extending from 14 to 19 July 2006 were subjected to harmonic analysis. In Figure 6 the modeling results are compared to measurements carried out at Lastovo, whereas in Figure 7 the modeling results are depicted for a wider middle-Adriatic area.
 The 24-h isotherm oscillations are closer to observations in the flat-bottom Adriatic (Figure 6b) than in the previously utilized analytical model (Figure 6a) and are even better approximating the data in the real-bathymetry case (Figure 6c). This suggests that the 24-h internal waves were resonantly driven by sea breezes around the island of Lastovo in mid-July 2006. The same improvement is observed at the 8-h period but not at the 12-h period: surprisingly, the latter oscillation is better reproduced by the analytical model than by either version of the numerical model. A possible explanation is inadequacy of wind-forcing, since the winds observed on the island of Vis are not fully representative of the winds in the area of Lastovo. The other possibility is that the numerical model errs at this period, e.g., due to some numerical problems or an overestimation of the nonlinear effects that accompany the resonant generation of island-trapped waves. The exact nature of the nonlinearity is not clear. It may be that advection and entrainment result in a reduced proportional density defect when the thermocline rises. As shown by Orlić and Pasarić , an increase of the thermocline displacement could then be expected. This may lead to a further decrease of the proportional density defect, etc. However, a thorough analysis of the positive feedback loop is beyond the scope of this work. As for the variability at the largest periods, the numerical model parallels the analytical model in underestimating it, thus confirming that the variability is not related to the diurnal wind-forcing considered here.
 As for the conditions in a wider middle-Adriatic area, the 24-h isotherm oscillations in the flat-bottom Adriatic are found to be pronounced close to the islands of Vis and Lastovo, with the phase propagating in a clockwise direction around them (Figure 7a). Surprisingly, the amplitude is not largest near Lastovo unless real bathymetry is included (Figure 7b), indicating that a topographic Rossby effect is important for resonant excitation of subinertial island-trapped waves. Both the amplitude and phase obtained for the 12-h thermocline oscillations show that the oscillations are related to waves radiating away from the topography rather than being trapped by it (Figure 7c). These modeling results agree with the thermistor measurements that were carried out not only on the island of Lastovo but also on the much smaller islands of Sušac and Biševo (Figure 1); their spectral analysis revealed that, whereas the 24-h variability at Lastovo was much larger than at the other two islands, the 12- and 8-h variability at all three islands was similar [Mihanović et al., 2009]. It should be stressed again that the pronounced variability at Lastovo depended not only on island dimensions but also on stratification (which in mid-July 2006 supported internal waves circumnavigating the island in about a day) as well as on diurnal wind-forcing (which was most pronounced in mid-July 2006). Different stratification conditions would reduce the variability at Lastovo and probably strengthen it at some other islands, whereas weaker diurnal wind-forcing would obviously reduce variability everywhere.
 The data collected in the Adriatic in mid-July 2006 show that the thermocline was subject to a 30-m diurnal variability at a steep cliff of the south coast of Lastovo, being lowest at about 6 a.m. and highest at about 4 p.m. with a secondary maximum occurring at about 10 p.m. The diurnal currents measured off the island west coast flowed northwestward (southeastward) in the surface layer when the thermocline was rising (falling), thus pointing to internal coastal waves propagating around the island in a clockwise sense. Comparison of the data with the winds simultaneously measured in the area suggests that the diurnal waves were driven by the sea breezes turning clockwise throughout the day. A simple analytical model confirms that over a part of the summer the stratification promotes internal coastal waves that revolve daily around the island in a clockwise direction, creating the conditions needed for resonant excitation by the diurnal winds. Numerical modeling reveals that the 24-h waves are trapped around the island due to the influence of both the Coriolis force and bottom slope, and that the 12-h waves radiate away from the island.
 In the present analysis we have concentrated on the mid-July 2006 episode because it was clearly wind driven. In a previous paper the three cases with prevalent tidal forcing were studied [Mihanović et al., 2009]. It should be mentioned that the diurnal thermocline variability was present over a greater part of our measurement interval, but it was usually less pronounced and the two forcing mechanisms could not be readily disentangled. It therefore seemed appropriate to focus in these, first analyses of the Adriatic data set on simpler cases, and to put off a more complete statistical analysis to a latter occasion.
 How important is this, apparently new type of diurnal upwelling for climate-related and biological issues? The overall effect of periodic upwelling is a cooling of coastal zones above which time-variable winds blow. It is not known, however, whether global warming would intensify this process by strengthening sea breezes and altering stratification in a way that would change the conditions at Lastovo and other similar islands to more closely resemble resonant ones and whether the process is widespread enough to be of global importance. As for the productivity of coastal waters, the diurnal upwelling may influence the generation of phytoplankton characterized by a near-daily scale [Walsh et al., 1977] and therefore may also influence the generation of zooplankton and nekton at much larger temporal scales. The well-known fact that the larger areas of the islands of Lastovo and Vis are relatively productive ones in the Adriatic [Zavatarelli et al., 1998] supports the proposed mechanism and suggests that these islands represent the natural laboratories in which the generation times of various members of the food web can be studied. The diurnal upwelling at Lastovo may also influence benthic organisms. There is less diversity of bryozoans and other organisms populating the cliffs down to a 40 m depth at Lastovo than at other Adriatic locations [Novosel et al., 2004], and this may be partially due to the fact that organisms at Lastovo are often exposed to large temperature changes over short time intervals.
 We thank the scientists and crews of RVs Bios and Palagruža for their skilled contribution to the field phase of the experiment. In particular, we are indebted to Maja Novosel and Anđelko Novosel for leading the team of divers who deployed and recovered the thermistors. Meteorological data and Aladin modeling results were provided by the Meteorological and Hydrological Service of the Republic of Croatia. We thank Paola Nardini and Michel Rixen for preparing and sharing the NATO Undersea Research Centre bathymetry data. We also thank two anonymous reviewers for providing constructive comments on the manuscript. The work was supported by the U.S. Office of Naval Research and Naval Research Laboratory (project ITHACA, grant N00014-05-1-0698) and the Croatian Ministry of Science, Education, and Sports (grants 119-1193086-3085, 001-0013077-1122, and 022-0222882-2823).