Backhaus, Reference: J. O., H. Wehde, E. Nst Hegseth und J. Kmpf, (1999): Phyto Convection -On the role of Oceanic Convection in Primary Production-. Mar. Ecol. Prog. Ser., Vol. 189: 77-92.
Phyto-Convection - On the Role of Oceanic Convection in Primary Production -
Jan O. Backhaus , Henning Wehde , Else Nst Hegseth , and Jochen Kmpf
: Institute of Oceanography, University of Hamburg, Troplowitzstr.7, D-22529 Hamburg, Germany : The Norwegian College of Fisheries Science, University of Troms, N-9037 Troms, Norway email: [e-mail]
ABSTRACT: Typical sinking rates of marine phytoplankton cover a range from a few meters up to several hundred meters per day. If it were not for a process which maintains plankton near the sea surface, in the euphotic layer, it would sink to depths of thousands of meters in the deep ocean during a winter season. Consequently plankton would not be available for the next spring bloom. In shelf seas and coastal areas, as well as fjords, a deep sinking is prohibited. The mechanism which reliably initiates a spring bloom is generally not considered in models of marine primary production. They generally rely on the assumption that a very small background concentration of plankton is available to initiate a bloom. Penetrative oceanic convection in the open ocean forms the perennial thermocline in winter in mid and high latitudes. The thermocline is situated at depths of several hundred meters. On a shelf, or in a fjord, convection may penetrate to the seabed thereby affecting the entire water column. We argue that it is oceanic convection in winter which accounts for the availability of plankton in the euphotic layer in spring. In support of this hypothesis a coupled phytoplankton convection model was developed. Plankton, i.e. resting spores and vegetative cells, is simulated by Lagrangian tracers moving within the flow predicted by the convection model. For each tracer a simple phytoplankton model predicts growth in dependence of light conditions. Plankton spores sink with a prescribed velocity of 120 m/d. Growing vegetative cells have a sinking rate of only 1 m/d. The model operates in a vertical ocean slice covering the water column. The width of the slice is typically 1-3 km, and it is resolved by an isotropic grid size of 5m. The phyto-convection model was applied to a region in the Barents Sea shelf and to a coastal fjord in the north of Norway. It was run over winter/spring periods under realistic meteorological forcing. Tracers representing resting spores were initially offered in a thin bottom layer of the model domain which constitutes the worst case in terms of maximum sinking. The water column, apart from the bottom layer, was assumed to be void of plankton. In both cases convection eroded the initial stratification and dispersed plankton over the entire water column. The onset of a phytoplankton bloom coinciding with the establishment of a (weak) seasonal thermocline in spring was predicted which agrees with observations from both regions considered. The simulations support the hypothesised role of oceanic convection in primary production.
KEY WORDS: Oceanic Convection, Primary Production, Dispersion of Phytoplankton
2
1.INTRODUCTION
Phytoplankton spring blooms normally require a stable surface mixed layer to develop. The mixed layer may either be built up by thermal heating of water masses in the open ocean, and/or by fresh water input in near-shore regions. In northern Norwegian fjords and coastal areas such blooms have for a long time been known to take place in apparently unstratified, or only weakly stratified water masses since strong pycnoclines are not established until May/June when water temperature and fresh water input increase (Heimdal 1974, Eilertsen & Taasen 1984, Hegseth et al. 1995). Although spring blooms have extensively been studied during the past 25 years, the initialisation of a bloom has (surprisingly) received less attention. New investigations have demonstrated the probability of diatom resting spores functioning as the bloom inoculum (Eilertsen et al. 1995, Hegseth et al. 1995) and thereby playing an essential role in phytoplankton population dynamics. The idea that resuspended resting spores act as seed populations for blooms was presented by Gran (1912). He considered spores as an
overwintering stage, but later studies have shown that they may have a wider range of function and seed blooms on variable time scales and in variable ecological systems (Garrison 1984, Smetacek 1984, Pitcher 1991, Pitcher et al. 1992). In temperate areas it has proven difficult to verify that spores actually seed the spring bloom since vegetative cells are always present during winter (Garrison 1981, 1984). However, in northern Norway the winter situation is characterised by a long dark period with extremely low phytoplankton biomass. The few vegetative cells present do not belong to the spring bloom species (Hegseth et al. 1995, von Quillfeldt 1996), rather they represent benthic populations and thereby indicate resuspension from the bottom. Hence, the inoculum for the spring bloom has to come from somewhere else than the water masses, and resting spores are the most likely source. The soft sediment surface of northern Norwegian fjords contain large amounts of spores which have shown to germinate and develop into a spring bloom under experimental conditions (Eilertsen et al. 1995). Typical sinking rates of marine phytoplankton, mostly diatoms, cover a wide range from a few meters up to several hundred meters per day (Smayda 1970, von Bodungen et al. 1981, Billett et al. 1983, Platt et al. 1983, Passow 1991). Cells might sink during a winter season to depths of thousands of meters in the deep ocean and consequently would not be available for the next spring bloom. Chlorophyll-containing phytoplankton cells have in fact been reported from depths down to 4300 m (Kimball et al. 1963), indicating high sinking rates. In shelf seas 3
and coastal areas, as well as fjords, deep sinking is prohibited by the proximity of the seabed. There must, however, be processes bringing up and maintaining phytoplankton near the sea surface both in open ocean and in near-shore areas. The physical mechanism which initiates a plankton bloom is generally not considered in detail in models of marine primary production (Moll, 1995, Aksnes and Lie, 1990). These models rely on the assumption that a very small background concentration of plankton is always available to initiate a spring bloom. Observations appear to justify this assumption, but the actual mechanism accounting for the presence of plankton in the euphotic layer in spring remains in the dark. Barkmann and Woods (1996) simulated a vertical migration of plankton by means of a Monte Carlo approach in which estimates of ambient turbulence (caused by wind and waves) were determined from a turbulence closure scheme. Their Lagrangian study was confined to the upper 200 m of the water column and they did not try to explain dispersion of plankton during the winter spring season. Several mechanisms have been suggested to bring spores from the bottom layers to the euphotic zone: turbulent mixing (Eilertsen et al. 1995) or upwelling (Malone et al. 1983, Pitcher 1991), being the most common. We propose that convection is the mechanism which accounts for a transport of spores to the upper layers during winter, and in the early phase of a spring bloom. Convection in high and mid latitudes is driven by a cooling of the sea surface in winter. It is only recently (i.e. in the past 10-15 years) that convection in the ocean has been studied in detail because of its importance as a process relevant to understand climate dynamics. A comprehensive overview of the present knowledge about oceanic convection, comprising observations and model simulations, is given by Marshall and Schott (1998). Generally, a discrimination is made between convection in the open water column and bottom arrested slope-convection (also called: cascading). Detailed studies of oceanic convection in the open water column of the Greenland Sea, in the presence of sea ice, were conducted by Backhaus (1995), Kmpf and Backhaus (1998), and Backhaus and Kmpf (1999). Convection in the water column of polar shelves plays an important role in forming dense bottom water masses which would finally cascade down the continental slope to account for a ventilation of deep water masses in the Arctic Ocean (Jungclaus et al. 1995, Backhaus et al. 1997). More information about convection in the context of this investigation is given below (see: synopsis under materials and methods), where our experimental set up is described. Penetrative open ocean convection, in contrast to wind induced and tidal dynamics, is a turbulent process which generally is not confined to the marine boundary (Ekman) layers. Oceanic convection in the open ocean, in mid and high latitudes, forms the perennial 4
thermocline during winter. This thermocline may be situated at depths of several hundred meters, i.e. well beneath the surface Ekman layer. In the North Atlantic, for instance, Subpolar Mode Water masses are formed in winter by convection in the vicinity of the polar front (van Aken and Becker 1997, McCartney and Telly 1982). These waters may cover (on a seasonal average) the upper 200 - 600 m of the water column. On a shelf, or in a fjord, convection may penetrate to the seabed, thereby affecting the entire water column. We argue that it is convection in winter which accounts for the availability of plankton in the euphotic zone in spring. In support of this hypothesis, and due to lacking observations from winter, a coupled phytoplankton convection model was developed. In this model plankton is simulated by Lagrangian tracers which move within the flow predicted by an oceanic convection model. For each tracer a simple phytoplankton model is solved, thus predicting the growth of plankton in both time and space. The phyto-convection model was applied to the Barents Sea shelf and to a coastal fjord in northern Norway by considering a winter/spring period under realistic meteorological forcing.
2. MATERIALS AND METHODS
2.1 Observations and Data
Salinity and temperature data were sampled regularly by CTD casts in fjords and coastal areas of northern Norway by the University of Troms (see Norman 1990-1993, or Havmiljdata database on webside of UIT/NFH (University of Troms, Norges Fiskeri Hoyskole)). In this work we have used CTD observations from 1990 - 1995 from Balsfjord. For a shelf region in the Barents Sea T,S-data predicted by the HAMSOM model of the Barents Sea (HAMSOM: HAMburg Shelf-Ocean Model, Backhaus 1985, Harms 1994) was provided by Harms (pers. com. 1998). Six-hourly meteorological data (wind speed, air temperature, humidity, and cloudiness), provided by the Norwegian Meteorological Institute (NMI) and the ECMWF (European Centre for Medium Range Weather Forecast), were used as model forcing.
2.2 The Coupled Phytoplankton-Convection Model (PCM)
The main focus of this modelling process-study is devoted to support our hypothesised relationship between oceanic convection and phytoplankton succession, and its important 5
support for the success of a spring bloom. In addressing this interesting interdisciplinary problem we hope that this paper will be read by both marine physicists as well as marine biologists. However, disciplines have their specific jargon, in particular when it comes to modelling. We want to avoid that a reader, who just wants to be informed about the process which we call phyto-convection, is distracted by an extensive use of (modelling) jargon. We, therefore, included a short synopsis in which information about both convection and our model is provided. In this synopsis we have avoided any jargon. It is followed by a more detailed description of the model which would require some knowledge about both modelling and convection.
2.3 Synopsis
Convection, in contrast to predominantly horizontal ocean dynamics, is characterised by substantial vertical, both up- and downward motions (typical extremes of vertical velocities: up to 20 cm/s). Here, we consider convective dynamics driven by a negative buoyancy of waters at the sea surface which results from a cooling of the ocean. However, convection can also be induced by heating from beneath as it occurs, for instance, in hydro-thermal vents. Spatial scales of convection are generally much smaller, and vertical accelerations are much larger than for predominantly horizontal flow. This is characteristic for nonhydrostatic dynamics. Unless convection covers a region large enough to feel the effect of earth rotation, it occurs on spatial scales which are well below those of baroclinic eddies (< internal Rossby radius of deformation). The nonhydrostatic dynamics require a special approach in modelling which differs from the commonly used hydrostatic models applied for predominantly horizontal flow. (In non-hydrostatic physics (models) both vertical accelerations and velocities play an important role and require an equation of motion for the vertical coordinate. This is not the case for hydrostatic physics where vertical velocities are diagnosed from the equation of continuity; the equation of motion for the vertical co-ordinate is omitted.) The convection model utilised in this investigation has previously been applied for simulations of convection in the Greenland Sea. Hence, a well-tested tool was available. Convection, originating from a cooled ocean surface, is characterised by energetic downward motions. The sinking of these heavier water masses occurs in narrow funnels, or plumes. Conservation of mass requires an upwelling in the vicinity of sinking regions. The upward motion is generally less energetic and occurs on larger spatial scales. The aspect ratio of convection (Kmpf and Backhaus 1998), i.e. the relation between the horizontal separation of 6
plumes and their vertical extent, has been found to be in the order of 1 : 2.5 (vertical versus horizontal scale). Both downward and upward motion form a convective cell which may have, in dependence to ambient turbulence, a more or less well organised orbital shape. According to our hypothesis, plankton is dispersed with the orbital motions of convective cells in winter. To simulate this we considered individual water parcels, i.e. Lagrangian tracers, which are moved around by convective dynamics within the water column. Each tracer in our coupled Phyto-Convection Model (PCM) carries a simple phytoplankton model which predicts plankton growth in dependence to environmental conditions (light, nutrients, grazers etc.). Whenever a tracer approaches the euphotic layer near the sea surface it may receive an amount of light-quanta before it descends again into the oceanic darkness. Therefore, a slow plankton growth is expected which would largely depend on the frequency of the cellular motion. This, in turn, depends on the applied meteorological forcing and may cover a wide range of time scales. The meteorological forcing of the model comprises winds, air temperature and humidity, cloud coverage, and long and short wave radiation. These parameters are needed to compute a) the heat (and buoyancy) fluxes which account for the surface cooling/heating, and b) the light conditions for the phytoplankton model. As a result of the applied forcing the model predicts the temporal evolution of currents and changes in both temperature and salinity (from which the density is computed). The model requires start conditions for both flow and temperature and salinity. The latter may be obtained from observations. An ocean at rest is usually assumed as dynamical initialisation. The model is defined in a vertical ocean slice by ignoring any gradients normal to the slice. Advection of matter, and of water mass properties, from outside the slice is avoided by pasting the lateral boundaries of the slice together. This way, the model domain becomes a self-contained micro-cosmos: disturbances which may, for instance, leave the ocean slice on the right hand side re-enter the domain via the left hand side (modellers jargon: cyclic boundary conditions).
2.4 Details of the Coupled Phyto-Convection Model For the dynamical part, i.e. the convection model, we make reference to already published papers. For convenience, we here only denote the equations for the biological model which is carried around by each tracer. A nonhydrostatic, rotational convection model (Backhaus 1995, Backhaus and Wehde 1996) was coupled to a Lagrangian phytoplankton model (Wehde 1996). The convection model has 7
previously been applied to investigate convection, resulting water mass formation and iceocean interactions in the Greenland Sea (Backhaus 1995, Kmpf and Backhaus 1998, 1999; Backhaus and Kmpf 1999). The model ignores large-scale advective transports and works on sub-meso spatial scales, i.e. on scales well below the internal Rossby-radius of deformation.
2.4.1 The Ocean Model
The non-linear Bousinesq equations for an incompressible fluid of the nonhydrostatic, rotational model and further details about its numerical scheme are described in detail in Kmpf and Backhaus (1998). Therefore only a brief description is given here. The model utilises an equidistant numerical grid (Arakawa C) and grid sizes in the order of meters. In contrast to previous convection studies, which considered the rotational phase of convection (see overview by Marshall and Schott 1998), our grid is isotropic to avoid any distortions of convective dynamics by a non-isotropic grid. The model domain is a vertical ocean slice with vanishing normal gradients of all variables. With a prescribed grid size of 5 m the model time step, in the below described applications, varied between 5 and 45 seconds, in dependence to convective dynamics. The time step depends on the courant number of the applied explicit numerical advection scheme for both momentum and water mass properties. Cyclic boundary conditions at the lateral boundaries of the model domain are applied, thereby excluding lateral advection by the large scale flow. Independent conservation equations for heat and salt are linked to the momentum equations via a non-linear equation of state (UNESCO). The model predicts the spatial and temporal evolution of T,S (density), nonhydrostatic pressure and flow fields from an initial (horizontally homogeneous) temperature and salinity profile. The turbulent eddy viscosity, and diffusivity, in the ocean model are parameterised by a simple diagnostic one-equation turbulence closure scheme (Backhaus 1995, Kochergin 1987). Coefficients for turbulent exchange of momentum and diffusion of water mass properties are assumed equal. The validity of this concept is confirmed by our previous convection studies (Kmpf and Backhaus 1998; Backhaus and Kmpf 1999). The high spatial and temporal resolution of the model allows to resolve a good deal of the turbulent spectrum which is usually parameterised in larger scale models. The ocean model is forced with fluxes of momentum and heat computed from bulk formulae (Gill 1982). For the computation of the heat flux the sea surface temperature (SST), predicted with the model, and prescribed atmospheric data (air temperature, humidity, wind speed, and
8
cloudiness) are used. The thermodynamic forcing comprises sensible and latent heat-fluxes and short and long-wave radiation (Friehe and Schmitt 1976).
2.4.2 The Lagrangian Transport Model
The simulation of the motion of phytoplankton in the vertical ocean slice of the coupled PCM is accomplished by Lagrangian tracers, i.e. marked water parcels which follow convective dynamics. We preferred to use a Lagrangian approach, as opposed to an Eulerian, because it allows to follow single tracers in time (and space). This way a process-oriented simulation of the anticipated dispersion of phytoplankton is achieved. Moreover, Eulerian schemes generally suffer from numerical diffusion, which we wanted to avoid in this application. With each newly predicted flow field, i.e. at each model time step, positions of all Lagrangian tracers are updated. Convective dynamics may change rapidly on small spatial and temporal scales. Any temporal interpolation in a strongly variable flow field might result in erroneous trajectories of tracers. Therefore, a temporal interpolation of flow fields, sampled from the model output with a time step larger than the (dynamical) model time step, was avoided. Note, that in coarser scale tracer-simulations tracer positions are often updated from flow fields which are interpolated in time. This is only justified if the flow would not vary on both small temporal and spatial scales.
2.4.3 The Phytoplankton Model
In this study we want to pay special attention to the hypothesised fundamental relationship between oceanic convection and primary production. Therefore, model components were reduced to the most simple case. We made use of the phytoplankton model described by Moll (1995, 1998). We created a process-study void of any second order effects by reducing our model to essential mechanisms only. For this reason a phytoplankton model is used in which the stock of phytoplankton biomass A (mg C m-3) changes according to: Error! = A (rP min (rI , rN) - rR - rM - rZ ) (1)
The right hand terms of (1) express gross primary production, respiration, mortality and grazing. Where rR , rM , and rZ are respiration, mortality and grazing rates, respectively, and rp is the optimal growth rate for phytoplankton. In our simple model respiration, mortality and grazing are assumed proportional to the phytoplankton stock A. The term min (rI , rN) in 9
equation (1) indicates the minimum of variables in brackets. All constants used in this model are given in table 1 below. Following Liebigs law the gross primary production is calculated from the minimum of the limitation functions for light rI (2) and nutrients rN (4). We used Steeles (1962) formulation for underwater light intensity I (W m-2) and optimal light intensity (I1) including photoinhibition as limitation function for light.
rI =
I (1 - I/I ) e 1 I1
(2)
The selfshading effect of phytoplankton is ignored in the model because our interest in phytoconvection ends with the onset of a phytoplankton bloom. Therefore, the available underwater light intensity (3) is modelled by the simple relationship: I(z,t) = Qswav (t) * e ( - k0 * z ) z (3)
where the amount of incoming short wave solar radiation Qswav (W m-2), depending on time, latitude, and observed local cloudiness, is calculated after Dobson and Smith (1988). Here, k0 is the extinction coefficient which describes a background concentration of suspended matter within the water column, and z (m) is the actual water depth at which the estimate for available underwater light intensity is needed. The Michaelis-Menton relationship (4) is applied to describe a nutrient limitation. However, in our experiments (cf. table 1) nutrient limitation will have only a small effect because, according to available observations, there are always enough nutrients available until the peak of a phytoplankton bloom (Hegseth 1995, Mann and Lazier 1991). The limitation of nutrients, entering in (1), is given by the function:
rN =
P P + kS
(4)
where kS is the half - saturation constant, and P (mmol m-3) the available phosphate. The sinking of tracers representing spores was set to a constant rate of 120 m/d, in order to simulate observed sinking rates of diatom spores (Hegseth 1995, Degens 1968). With the onset of plankton growth a self-induced buoyancy of growing diatoms, i.e. vegetative cells, was simulated by reducing the sinking velocity of tracers to 1 m/d (Moll 1995).
10
Table 1: List of Constants used in the Phytoplankton Model
Quantity optimum light intensity Extinction coefficient Maximum growth rate of phytoplankton sinking velocity of spores mortality rate grazing rate Respiration rate Phosphate half saturation constant
Symbol I1 k0 rp ws rM rZ rR kS
Value 46 0,09 1,5 120 0,05 0,5 0,06 0.06
Unit W m -2 m 1 day 1 m day -1 day 1 day 1 day 1 Mmol PO4-P m-3
3. Experimental Set-up
Experiments with the PCM, simulating different conditions in polar coastal seas were conducted to study phyto-convection in winter and spring. Regions for which data for an initialisation of T,S properties were available from either observations or larger scale modelling were chosen. The first experiment concerned the Balsfjord (water depth: 160 m; latitude: 69 30 N) near Troms (henceforth indicated by FJORD), the second a region in the south-western Barents Sea (water depth: 240 m; latitude: 73 N) which remains ice free during winter (henceforth indicated by SHELF). In this first attempt to understand phytoconvection supported by a process-oriented simulation we wanted to exclude sea ice. The horizontal dimension of the ocean slice must be chosen in dependence to the expected convective aspect ratio, in order to account for a process-oriented spatial resolution of convection. In considering a maximum vertical extent of convection of 160 and 240 m, respectively, i.e. the depth of the considered water columns, we defined a horizontal dimension of 1.25 km and 2 km, respectively for our model domain. With this choice we made sure that at least 3 convection cells are always included within the model domain. The ocean slice is resolved by an isotropic grid size of 5 m. Two records of six-hourly meteorological forcing from the years 1990 and 1994 were available for experiment FJORD; for experiment SHELF only data from 1990. Hence, two model runs were conducted in the course of experiment FJORD. Both runs were initialised 11
with the same observed T,S-profile from Balsfjord (figure 1) obtained during winter. The prescribed profile served as initialisation for a process-study conducted with the intention to simulate typical conditions in stratification. Making use of one and the same initial stratification allows a judgement of the influence of the meteorological forcing on convection without considering the dependence on different initial conditions. (our results below show that the final stage of convection in all our experiments is independent of the initial condition, because convection always penetrated through the entire water column accounting for a total homogenisation). Experiment SHELF was initialised by a T,S-profile (figure 2) extracted from results of a baroclinic shelf model (Harms 1994, 1995). The latter predicted circulation, water mass, and ice formation for the entire Barents Sea shelf. Its results were validated (Harms 1994) against field observations from Loeng (pers. com. 1996). The T,S-profile (figure 2) is typical for the ice free region in the south-western Barents Sea with a dominating influence of the Atlantic inflow. In all experiments the model was dynamically initialised from a state of rest. In regard to Lagrangian transport the experiments were initialised by a thin (15 m) homogeneous bottom layer in which tracers representing resting spores were evenly distributed. A number of 45000 tracers was considered in experiment FJORD. In SHELF 67000 tracers were used, accounting for the larger model domain. A high number of tracers was chosen to ensure a high resolution in our dispersion study. Based upon our pervious experience in convection modelling (Backhaus 1995, Kmpf and Backhaus 1998) we expected a high temporal and spatial variability of tracer distributions as a consequence of the complex convective dynamics. By initially placing all tracers in a thin layer at the seabed we considered the most extreme case in terms of sinking. It implies that plankton spores have accumulated at the seabed. The water column above the layer was assumed to be void of spores (and vegetative cells). The existence of such a layer can be inferred from observed bottom nepheloid layers which are created with the bottom Ekman layer, for instance, by tidal stirring, or mixing due to internal waves (McCave, 1986).
4. RESULTS 12
In support of the hypothesised phyto-convection mechanism it is necessary to provide a detailed insight on both tracer dispersion and plankton growth during a winter/spring period. Since, in our model, dispersion and growth results from the combined effect of oceanic convection, thermodynamic forcing, and light conditions, we also need to describe the physical environment. A detailed description of the results of our simulations can largely be confined to experiment FJORD because it also exemplifies the results obtained for case SHELF. However, the most important information emerging from the latter, i.e. plankton production in relation to the stratification predicted for spring, will be presented.
4.1 Meteorological Forcing for Experiments FJORD and SHELF
Experiment FJORD consists of two separate model runs, each covering a time span of 120 days. The runs were conducted for two different winter/spring seasons (1990 and 1994) in which the model was forced by six-hourly atmospheric data obtained from the Norwegian Meteorological Institute and the ECMWF. The data (figure 3) were chosen because they represent the forcing for a comparatively mild (1994) and a colder winter (1990), respectively. This allows to judge the dependency of phyto-convection on the applied forcing. For experiment SHELF (figure 4), located in the Barents Sea, only one model run with an atmospheric forcing from 1990 was conducted. In experiment SHELF, located further north, oceanic heat losses during the winter/spring period of 1990 were much higher (figure 4b), even as compared with the cold FJORD forcing from 1990 (cf. figure 3b).
4.2 Experiment FJORD 4.2.1 Predicted Water Properties
The Mass evolution of the thermal stratification in experiment FJORD-1990, resulting from convective activity, is given by both a series of consecutive temperature profiles (figure 5) and a T,S-diagram (figure 7). The forcing for the winter/spring season of 1990 accounted for a rapid cooling of near surface waters which de-stabilised the stratification (figure 5a). The final stage of the convective erosion of the initially stable stratified water column was a total
13
homogenisation of water masses. For spring 1990 (time > day 100) the onset of a seasonal thermocline was predicted (figure 5b). Temperature profiles from the final simulation days of experiment FJORD-1994 (figure 6) show a similar establishment of a seasonal thermocline as predicted for 1990. However, as a result of the milder forcing bottom temperatures (figure 6), at the end of the winter, are higher than predicted for 1990 (cf. figure 5b). Also for the winter/spring period of 1994 convection had accounted for a total homogenisation of the water column (not shown). Convective water mass transformation for both cooling and warming phases are best described in a T,S-diagram (figure 7). In the diagram for simulation FJORD-1990 a homogenisation of the haline stratification together with a cooling of the upper part of the water column is evident. This is seen from a comparison of the TS-properties at 12 h (close to the initial T,S properties) and the T,S characteristics obtained after 15 days (figure 7). Thereafter, convection penetrates down to the bottom and, finally, a total homogenisation in terms of both temperature and salinity is obtained. The resulting T,S-cluster of the homogeneous water mass appears as one T,S-point (cf. results after 20 d in figure 7) with T 4.3C, and S 33.3 psu. Salinity, in our simulation, behaves like a passive tracer due to the thermal character of convection, as opposed to haline convection which occurs when ice is formed (Backhaus et al. 1997). With an ongoing cooling of the system, for the remaining winter period, the point cluster approaches a minimum temperature at 2.7C. The onset of the seasonal thermocline is seen by T,S-clusters approaching higher temperatures, starting at the lowest winter temperature (cf. results after 120 d in figure 7). The T,S diagram for experiment FJORD-1990 (figure 7) is a typical example for a convection which, during a winter, penetrated through the entire water column. The predicted temperature profiles of experiment FJORD may be compared with a typical seasonal cycle of the observed thermal stratification in Balsfjord (figure 8). Predicted bottom temperatures at the end of a winter agree very well with the observed temperatures (cf. figures 5, 6, and 8).
4.2.2 Convection and Tracer Dispersion
Convective dynamics penetrating towards the seabed, thereby collecting tracers from the thin bottom layer, are depicted by instantaneous temperature contours (figure 9) within the ocean slice. Predicted tracer positions are superimposed on the temperature contours to illustrate convective activity. Well before convection has proceeded all the way down to the 14
seabed (time < 15 days) the shape of the bottom layer of tracers attains a wavy structure (figure 9a, b). These motions near the seabed are driven by nonhydrostatic pressure fluctuations which are generated by convection and transmitted through the water column despite of the remaining stratification (Backhaus 1995, Backhaus and Kmpf 1999). Apparently, convective dynamics have an influence also on dynamics which are well beneath the actual penetration depth of convection. However, soon after simulation day 16, tracers leave the bottom layer and start to move into the water column (figure 9c). At first the instantaneous tracer positions remain close together (figure 9d), resulting in almost isolated trajectories. However, with time proceeding (days 15 25), tracers increasingly cover the entire water column and their distribution (figure 9e, f) attains a marble-like pattern (note, that temperature contours were omitted in figure 9e,f because of the thermal homogenisation of the water column). Up until circa 40 days of simulation, a thin region near the sea surface remained remarkably free of tracers (figure 9e, f). The tracer-free surface layer shows significant changes in thickness. During later stages of the simulation, however, tracers were also found close to the sea surface (not shown), and a nearly homogeneous distribution within the entire water column was obtained. Figure 10 illustrates orbital motions of two selected tracers during 100 hours of simulation starting at day 16.5. It is evident that the trajectories are better described by a drunkards walk rather than by a well defined orbital cell. This is caused by ambient turbulence which is induced by convection itself. The final fate of two other tracers, randomly selected from the ocean slice, is illustrated by time series of their vertical positions (figure 11) for the entire simulation. With decaying convective activity, which coincides with the establishment of a seasonal thermocline (cf. figure 5b), vertical displacements of tracers become very small. One tracer, by chance, ended up near the sea surface (figure 11a). The second, however, ended up at depth, far away from the euphotic zone (figure 11b).
4.2.3
Predicted Primary Production
The evolution of the phytoplankton stock for the winter/spring period of 1990 is characterised by a slow production which, with light becoming available after about 30 days (cf. figure 3), lasted throughout the larger part of the simulation. Production commenced after tracers were dispersed towards the sea surface (cf. figure 9), in coincidence with available light. Predicted concentrations increased very slowly throughout the water column (time > 30 days), but stayed at values below 0.01 Chla (mg/m). The high initial concentration of plankton spores 15
within the bottom layer (figure 12a) is rapidly reduced between days 14 and 18. This coincides with penetrative convection reaching the seabed (cf. figure 9). Thereafter, very low concentrations of spores are predicted (figure 12a) which, however, cover the entire water column. Note, that initially the water column, apart from the bottom layer, was void of any plankton tracers. With the onset of a stable stratification (cf. figure 5b), i.e. after about 100 days of simulation, predicted plankton concentrations (figure 12b) show a rapid increase in the upper 40 m of the water column with a peak at depths around 20 m. Finally, at the end of the simulation (i.e. day 120), predicted concentrations (figure 12b) arrive at a value of about 0.9 Chla (mg/m). A very similar evolution of predicted production (figure 13) emerged from the FJORD-1994 run. However, one noteworthy difference occurred. After the development of a (weak) thermocline (cf. figure 6a), and a resulting increase in biomass (figure 13a), a convection event accounted for a new homogenisation of the water column (cf. figure 6). As a result vegetative plankton cells which had previously grown in the young thermocline were again distributed over the entire water column. Thereafter, the stratification stabilised again and a subsequent bloom occurred. Hence, plankton concentrations below 50 m (figure 13b) represent both plankton spores (concentrations: 0.01 mg/m) and vegetative cells. The latter have concentrations which are higher by almost one order of magnitude ( 0.1 mg/m).
4.2 Experiment SHELF; Predicted Stratification and Primary Production
Convective activity in experiment SHELF, as a result of higher oceanic heat losses for polar conditions (figure 4b), turned out to be more energetic compared to results of FJORD. Nevertheless, very similar results in terms of both stratification and plankton production were obtained for this experiment. We confine the description of experiment SHELF to predicted plankton production in relation to the timing of the thermocline, allowing for a comparison with results of experiment FJORD. The onset of a seasonal thermocline (figure 14), emerging from a water column which was homogenised during the preceding winter, coincides with the start of a plankton bloom in spring (figure 15). The bloom is confined to the upper 40 m. Peak concentrations occur around depths of 30 m. Note, that temperatures are much lower than in experiment FJORD. The slight warming of temperature profiles in experiment SHELF (cf. figure 14) is not the result of convection but the consequence of an additional positive heat flux which we prescribed throughout the water column. This way we simulated the effect of
16
an advection of Atlantic heat with the Barents Sea inflow which, in the region considered, prevents a formation of sea ice.
5. DISCUSSION
According to general knowledge oceanic convection in a global setting, i.e. the conveyor belt circulation (Broecker 1991), implies a sinking of water masses. However, conservation of mass requires that any sinking of water masses is compensated by an upward motion with the same magnitude in terms of water mass transport. As outlined in our synopsis this upward convective motion is less energetic and it generally covers larger spatial scales than the more energetic downward motion. In the context of primary production, according to our hypothesis, it is this less spectacular, weaker upward component of convection which plays the most important role. The upward motion provides the mechanism which brings sinking phytoplankton spores towards the sea surface during winter. The presence of downward motions, occurring simultaneously in the vicinity of upward motions, disperses plankton throughout those parts of the water column which are affected by convection. In shallow, tidally dominated shelf seas, for instance in the North Sea, convective mixing in winter is supported by, or superimposed on, tidal mixing. Any observed mixed layer deepening, for instance the formation of a perennial thermocline in the deep ocean, can only be explained by the existence of convection. In high and mid latitudes convection regularly destroys the seasonal thermocline in autumn, and ventilates underlying waters during winter. This explains the reliability of the transport mechanism which we call phyto-convection. The development of the phytoplankton spring bloom in northern Norwegian waters generally occurs in two critical phases. During the prebloom phase conditions for the subsequent bloom are set by providing the necessary inoculum of phytoplankton cells. In the bloom phase the actual growth takes place (Hegseth et al. 1995). In the latter a bloom dominated by sporeforming diatoms starts to develop during last part of March. At this time irradiance is sufficient to maintain growth in the euphotic zone (30-40 m), and day length has exceeded 12 hours. In the prebloom phase, that is before the vernal equinox, few of the typical spring bloom species are found among the vegetative cells. Spores, however, have been found both in shallow and deep waters (Eilertsen et al. 1995). Very little increase of the phytoplankton 17
biomass is registered in this early phase which may be partly due to the photoperiodic dependence of spore germination. Most spores require a day length of at least 12 hours to germinate (Eilertsen et al. 1995). Sediments in fjords and coastal areas are found to contain up to 5 million spores ml -1 of diatom species (Eilertsen et al. 1995). Samples from the thin surface layer of the sediments function as a spring bloom inoculum under laboratory conditions at any time of the year (Hegseth et al. 1995). Hence, spores are a potential inoculum for any phytoplankton bloom under natural conditions, given that a sufficient amount could be brought up from the bottom sediments to the euphotic zone. This constitutes a critical phase in terms of the preconditioning of a plankton bloom. Spores sink faster than vegetative cells (Davis et al. 1980, Bienfang 1981). Vegetative cells are able to regulate their buoyancy physiologically (Smayda 1970, Bienfang 1981). Spores with a large negative buoyancy must be maintained in the upper layers of a weakly stratified water column in early spring in order to allow for germination and subsequent growth. The maintenance of spores with negative buoyancy near the surface constitutes the second critical phase in primary production. Both critical phases are explained by the predictions of our phyto-convection model. The results of the simulations are self-explanatory. They help to understand why both vegetative plankton cells and spores can be present in the water column and, in particular, in the euphotic layer. The existence of both stages at the same time is a result of the randomness of convective dispersion. In spring, however, all cells within the water column have finally become vegetative. We arrived at this conclusion because the model predicted a weak growth during the preceding winter. The randomness of phyto-convection, according to our modelling process-study, suggests a comparison with a gamble. The gamble is won by those vegetative plankton cells which, by chance, are within the euphotic layer at a time when convection finally ceases and a stable (though weak) stratification is built up. This is a result of both decreasing atmospheric forcing and increasing irradience. A further establishment of the young thermocline may be disrupted by isolated atmospheric cooling events which frequently occur in spring. In that case convection again disperses plankton cells over the water column (cf. figure 13) and the gamble, governed by convective activity, commences again. Those cells which are within the euphotic layer after the cooling event would initiate a subsequent bloom. Hence, rapid weather changes in spring may induce a series of (little) blooms which eventually, with a net warming, would finally form a main bloom. 18
Observations of a series of brief blooms preceding a main bloom are discussed in Townsend et al. (1994). They highlight the role of minor blooms in regard to carbon export and suggest that they may considerably increase the estimated spring production as compared to (commonly applied) estimates based on just the main bloom. In this context it is interesting to note that Hirche (pers. com. 1999) reported about a pre-bloom phase observed at weathership M in the NorthAtlantic in 1997. It lasted for more than one month and female copepodes (Calanus finmarchicus) produced as many eggs in the pre-bloom phase as in the following main bloom. Reported concentrations of phytoplankton during the pre-bloom phase were in a similar order of magnitude ( 0.2 mg/m) as predicted by our model. The model predicted vegetative cells dispersed throughout the water column (figs. 12 and 13). We, therefore, concluded that plankton beneath an established seasonal thermocline would sink very slowly. These cells may be lost for the respective production season. Given favourable (light and nutrient) conditions they might however, later, induce a secondary plankton bloom beneath the thermocline. The predicted plankton dispersion within the water column (figure 9) shows an interesting feature during the first 40 hours of simulation: tracers, which are spores at this time, do not reach the surface but are dispersed throughout the underlying water column. The surface layer, void of tracers, shows a thickness which is variable in time. This is explained by a competition between the sinking of spores and convective action. Once convection weakens as a consequence of variations in forcing, the sinking in relation to convective motions, becomes more dominant (prescribed sinking rate was 120 m/d). The sinking affects all tracers in the water column, but it is only visible for those near the interface between the (clear) surface layer and underlying waters (cf. figure 9 e, f). Once convective dynamics become strong enough as a result of both forcing and eroding stratification (time > 40 days) tracers are also moved into the euphotic zone. Very near the sea surface (O(10-30 m)) predicted vertical velocities of convection are generally smaller than at greater depth. Convective plumes need first be accelerated before the sinking and, consequently, also the rising of waters attains higher speeds. Therefore, with weaker vertical velocities, it is easier to remove plankton cells from near surface waters. Provided there is enough light available for growth near the surface, the sinking rate of plankton in our model was reduced to 1 m/d. This simulates growing vegetative cells with a largely reduced negative buoyancy. However, the reduction only applies for cells which have been within the euphotic layer and which have received enough light for growth. It might be questionable to immediately change the sinking rate from 120 m/d to just 1 m/d once cells 19
become vegetative. A slower transition might be a more appropriate description of changes in buoyancy as a consequence of vegetative growth. However, since a quantitative description of changes in sinking rates in dependence to received irradience was not available, we decided to simulate the transition as crude as described. A slower change of sinking rates would retard the vegetative growth of cells because they can be removed more frequently from the euphotic layer. The predicted growth until the onset of a seasonal thermocline was very small anyway (cf. figures 11, 12 and 14) for growing cells were always removed from the euphotic layer by convective action. Therefore, the applied simulation of sinking rates has apparently very little influence on the final result of our simulations. Typical predicted magnitudes of velocities in a developed convective regime are in the order of 1-10 cm/s. A (comparatively) large sinking rate of plankton of 200 m/d cor...