ICES Journal of Marine Science, 55: 756–766. 1998 Article No. jm980378

Data needs for ecosystem modelling J. W. Baretta, J. G. Baretta-Bekker, and P. Ruardij Baretta, J. W., Baretta-Bekker, J. G., and Ruardij, P. 1998. Data needs for ecosystem modelling. – ICES Journal of Marine Science, 55: 756–766. To test, verify, and improve marine ecosystem models there is an urgent need for time series of those variables that reflect the state of the system in terms of productivity, nutrient cycling, and oxygen dynamics. Not only time series of verification data are needed, but also time series for the forcing functions (wind, irradiance, river inflows, and nutrient loads), as well as for the boundary conditions. Especially in open systems like the North Sea, where nutrient imports across the boundaries by far exceed river inputs, the boundary conditions are of crucial importance. The existing data coverage of the open boundaries is so sketchy that inter-annual variability has to be ignored and only climatological time series can be constructed for all variables. Model results are presented to demonstrate the linkage between realistic forcing and realistic system response, indicating that coupled hydrodynamical ecosystem models can reproduce and explain the observed variability in aquatic systems when provided with forcing conditions at appropriate time scales. The paper presents arguments for measuring fluxes and rates rather than standing stocks as a means for verifying ecosystem models.  1998 International Council for the Exploration of the Sea

Key words: coupled hydrodynamical/ecological models, ecosystem modelling, long-term time series, mathematical models, model verification, nutrient dynamics simulation models. J. W. Baretta and J. G. Baretta-Bekker: Ecological Modelling Centre, Joint Department of VKI and DHI, Agern Alle´ 5, DK-2970 Hørsholm, Denmark. P. Ruardij: Netherlands Institute for Sea Research, PO Box 59, NL-1790 AB Den Burg, Texel, The Netherlands. Correspondence to J. W. Baretta: tel. +45 4517 9124, fax. +45 4517 9200, e-mail: [email protected]

Introduction During the past 25 years, simulation models have gradually become established tools for integrating and testing knowledge on the different components of marine ecosystems. The first models only included parts of ecosystems, and were either steady-state (Steele, 1974; Billen, 1978; Mommaerts et al., 1984) or dynamical systems (Fransz and Verhagen, 1985; Billen and Lancelot, 1988). From the end of the 1970s onwards, more complex dynamical models of estuarine and marine ecosystems appeared, which included pelagic and benthic processes as well as advection and dispersion processes (Kremer and Nixon, 1978; Radford and Joint, 1980; Baretta and Ruardij, 1988; Baretta et al., 1995). To test and apply dynamical models, a mathematical description embodying established concepts of marine ecosystem structure and function does not suffice. A physical model is required in addition, representing the advective and diffusive processes that transport and redistribute the particulate and dissolved constituents in 1054–3139/98/040756+11 $30.00/0

the model domain. Last, but not least, time-series data are needed to describe the boundaries of the area modelled, to force the model, as well as for calibration and validation. The aim of this paper is to discuss the present availability of long-term data sets and their potential for evaluating the diagnostic and prognostic power of ecosystem models, and to give suggestions on how these data series could be improved. The European Regional Seas Ecosystem Model (ERSEM) is used as frame of reference. The long-term data series used so far have been aggregated into ‘‘climatological’’ monthly means within spatial boxes or regions. If possible, ranges around the means were derived as well (Radach and Pa¨tsch, 1997). The term ‘‘climatological’’ is commonly used in oceanography (Levitus, 1982; Damm, 1989) to indicate the long-term average and ranges. Any inter-annual variability is only apparent in the range. Whether a ‘‘climatological’’ mean has much meaning in the real world is doubtful, but its use implicitly assumes that temporal variability is strictly seasonal and that spatial variability  1998 International Council for the Exploration of the Sea

Data needs for ecosystem modelling 0°

5°

10°E

60°N

60°N 1/11

757 0°

5°

60°N

10°E

60°N

2/12

3/13

6

10

4/14

58 59 60 61

55°

55°

5/15

55°

65 66 67 68

9

55°

74 75

7 8

50° 0°

5°

50° 10°E

50° 0°

5°

50° 10°E

Figure 1. The North Sea box structure as used in ERSEMND15 with 10 surface boxes (1–10) and 5 deep boxes (11–15).

Figure 2. The North Sea box structure as used in ERSEMND130 with 85 surface boxes. Stratified boxes are indicated by –. Numbered boxes refer to Figure 10.

is time-invariant. The coverage in time and space of most of the North Sea by observations on chemical and biological variables is so sparse that this assumption cannot be avoided.

description of the biological modules has been given in Radford (1996) and in papers referred to in Baretta et al. (1995). Version 11 of the ERSEM model has been implemented in different regional seas and in different spatial set-ups, e.g. in the North Sea with 15 boxes (ND15; Fig. 1) corresponding to the areas distinguished by ICES (1983), or with 130 11 boxes (ND130; Fig. 2). The transport model consists here of daily exchange coefficients between the boxes calculated from a fine-scale hydrodynamical model (Lenhart et al., 1995). ERSEM has also been used as a 1D vertically resolved column model of a mooring site (Ruardij et al., 1997), coupled to an entrainment/detrainment model (Van Aken, 1984; Ridderinkhof, 1992). In this case, the physical model providing the forcing for the standard ERSEM biogeochemical sub-model has been replaced by a mixed layer model (Ridderinkhof, 1992). The implementation is characterized by a high temporal resolution of the forcing functions and a high spatial resolution of the water column. Other model requirements are initial values for all state variables, forcing functions, and boundary conditions. The forcing functions, such as water temperature and irradiance, and the boundary conditions for all

Modelling ERSEM is a comprehensive ecosystem model which dynamically simulates large-scale cycling of organic carbon, oxygen, and the macro-nutrients N, P, and Si over the seasons. The model consists of an interlinked set of modules describing the biological and chemical processes in the stratified or non-stratified water column and in the benthic system, as forced by light and temperature. The North Sea area modelled has been subdivided into smaller units (called boxes or grid cells, depending on their size). Physical transport between units is included by driving the model with the aggregated output of physical circulation and dispersion models in the form of time series of daily advective and diffusive exchange coefficients for all box boundaries or by directly forcing the biological modules with a hydrodynamical sub-model. In all different spatial set-ups, the biology in each box or grid cell is exactly the same. A full

758

J. W. Baretta et al. Box 1

Box 2

2.0

2.0

Box 3

1.0 1.0

Phosphate (mmol P m–3)

0.0

J FMAMJ J A S OND Box 4

2.0

0.0

1.0

J F M A M J J A S O N D Box 5

2.0

0.0 2.0

J FMAM J J A S OND Box 7

ND15 ND130 to ND15 Data 1.0

0.0 2.0

1.0

J FMAMJ J A S OND Box 8

1.0

0.0

0.0 2.0

1.0

J F M A M J J A S O N D Box 9

1.0

J FMAMJ J A S OND

0.0

0.0 2.0

J FMAM J J A S OND Box 10

1.0

J F M A M J J A S O N D

0.0

J FMAM J J A S OND


3

Figure 3. Phosphate (mmol P m ): simulated concentrations by ERSEM in the 15-box set-up (ND15) and in the 130-box set-up aggregated according to the 15-box set-up (ND130 to ND15) compared with the median (+) and the upper (83%) and lower (17%) quartiles (bars) of the data values derived from the ECOMOD data set. Only boxes 1–5 and 7–10 are shown.

transported state variables have been extracted from available measurements. The initial values of state variables are found by running the model for a number of years with fixed seasonal forcing functions and boundary conditions until the seasonal cycle becomes invariable. For calibration and validation, independent data have been used.

Materials Four main data sets allowing different types of application are available for the North Sea: (1) long-term nutrient and chlorophyll data from many different sources and incorporated in various national and international databases. (2) The National Environment Research Council (NERC) North Sea Survey data set for the

Southern Bight (August 1988 to October 1989) for macronutrients, oxygen, chlorophyll a, ciliates, heterotrophic nanoflagellates, and mesozooplankton. This data set is available on CD-ROM (Lowry et al., 1992) and also incorporates information on pelagic fluxes (primary production) as well as benthic nutrient fluxes. (3) Long-term data sets of nutrient loadings by the continental rivers from different Dutch and German authorities. Corresponding long-term data from UK rivers could not be obtained (cf. Lenhart et al., 1997). (4) A one-year data set of daily Chl a concentrations in the surface mixed layer and the bottom mixed layer at a mooring site in the Oystergrounds (5425 N, 0402 E; pers. comm. Van Haren). The first two data sets have been merged into the ECOMOD database (cf. Radach and Pa¨tsch, 1997). The data have been aggregated into climatological monthly

Data needs for ecosystem modelling Box 3

Box 2

Box 1 20

20

20

10

10

10

0

J FMAMJ J A S OND

0

J F M A M J J A S O N D

20

20

10

10

0

0

Box 7

J FMAMJ J A S OND Box 8

0

ND15 ND130 to ND15 Data

20

10

J F M A M J J A S O N D Box 9

0

20

20

20

10

10

10

0

J FMAM J J A S OND

Box 5

Box 4 Nitrate (mmol N m–3)

759

J FMAMJ J A S OND

0

J F M A M J J A S O N D

Figure 4. Nitrate (mmol N m

mean concentrations of N, P, Si, and Chl a for each of the spatial compartments in both the 15-box and the 130-box set-up. Data from the boundary regions were used to define the boundary conditions. The model was calibrated in the 15-box set-up by means of the corresponding data, while the 130-box data set was used for validation. The 15-month data set of the NERC North Sea Survey has been used in both the 15-box and the 130-box set-up to test whether the ERSEM model, after calibration, is capable of reproducing a specific annual cycle using the appropriate hydrodynamical forcing as well as the river nutrient loading for this particular period. The third data set allowed determination of the boundary conditions for river nutrient input. Finally, the mooring data set has been used in combination with the mixed layer model (Ridderinkhof, 1992) to test the capability of ERSEM to resolve the processes along the vertical axis, particularly in relation to the onset, variability, and decay of seasonal thermal stratification.

0

J FMAM J J A S OND Box 10

J FMAM J J A S OND


3

): see legend to Figure 3.

Results The results from ND130 have been aggregated into the coarse 15-box set-up for direct comparison with the ND15 results and the observations during the NERC North Sea survey (Figs 3–7). For phosphate, the simulated seasonal differences by region are largely in agreement with the observations, with the exception of Box 9 (German Bight) where observed values are higher than the modelled values (Fig. 3). It is suspected that this discrepancy is caused by a local deposition during late summer of suspended particulate matter (and its associated nutrients) originating from the Southern Bight, which fuels an enhanced benthic phosphate remineralization in the German Bight. This feature cannot be reproduced by the model, since resuspension/deposition processes, and hence bedload transport of particulates, have not been included. The observed dissolved nitrate concentrations in summer (Fig. 4) are reproduced accurately, but the start of

760

J. W. Baretta et al. 10

Box 1

Ammonium (mmol N m–3)

10

J FMAMJ J A S OND Box 4

0

Box 3

10

5

5

0

Box 2

10

5

J F M A M J J A S O N D Box 5

10

0

J FMAM J J A S OND Box 7

10

ND15 ND15 to ND130 Data 5

0 10

J FMAMJ J A S OND Box 8

5

0

5

5

0 10

J F M A M J J A S O N D Box 9

0

10

J FMAM J J A S OND Box 10

5

5

J FMAMJ J A S OND

0

J F M A M J J A S O N D

Figure 5. Ammonium (mmol N m

the decline in late winter/early spring does not coincide everywhere with the modelled decreases. Nevertheless, the rate of decrease and timing of the depletion of nitrate in the surface water is generally correct, as is the increase in autumn. The observed concentrations of ammonium are reproduced quite well in the central and western North Sea boxes, but have a tendency of being underestimated by the model in Boxes 8 and 9 along the continental coast (Fig. 5). The late-winter increase in concentrations of silicate is underestimated in the northern boxes and overestimated along the continental coast (Fig. 6). The simulated results for the central boxes tend to agree with the observations. The low summer values are even reproduced almost perfectly. The predicted spring bloom is generally not evident from the medians of the observations (Fig. 7). However, the long tail of high chlorophyll values indicates that the distribution of the observations is distinctly non-


3

0

J FMAM J J A S OND

): see legend to Figure 3.

Gaussian and the length of the upper quartile (83%) suggests that spring blooms do occur in almost all regions. The summer concentrations in the southern and central boxes are generally within the quartiles (17% and 83%), particularly for the results aggregated from ND130. In Box 5, the spring bloom is much less pronounced in the ND130 results than in the ND15 results. This is probably due to differences in timing and strength of the spring bloom within this box (cf. Fig. 8). Figure 8 also clearly illustrates the spatially and temporally uneven distribution in the available data. There are notable differences between the aggregated results of ND130 and the results of ND15, but one set of results does not appear to be consistently better or worse than the other in explaining the observations. The only exception is perhaps chlorophyll, where ND130 fits the observations better. However, it should be noted that Chl a is a diagnostic variable in the model, which is calculated from the biomass (in mgC m
3) of the four prognostic phytoplankton variables (diatoms,

Data needs for ecosystem modelling 15

Box 2

15

10

10

5

5

5

15

J FMAMJ J A S OND Box 4

0

10

5

5

15

Box 5

15

10

0

J F M A M J J A S O N D

J FMAMJ J A S OND Box 8

0 15

ND15 ND130 to ND15 Data

0

5

J F M A M J J A S O N D Box 9

0 15

10

5

5

5

0

Box 7

10

10

J FMAMJ J A S OND

J FMAM J J A S OND

15

10

0

Box 3

15

10

0

Silicate (mmol Si m–3)

Box 1

761

J F M A M J J A S O N D

Figure 6. Silicate (mmol Si m

autotrophic flagellates, picoalgae, and inedible phytoplankton). The carbon biomass of each group is converted by means of a C:Chl a ratio of 50 for diatoms and of 25 for all other groups. Thus, calculated Chl a values do not give information as to whether the biomass distribution of the different phytoplankton groups is correct, even when they correspond exactly to the observed values. Information on variation in the composition of the phytoplankton community in space and time would be more appropriate for evaluating the simulated results. Results of the model described by Ruardij et al. (1997) indicate that ERSEM, with the additional variability in vertical structure of the water column as expressed in the thickness of the non-mixed intermediate layer, reproduced the seasonal evolution of the phytoplankton biomass, as expressed in Chl a, surprisingly well (Fig. 9). Thus, the changing stratification structure of the water column appears to be a crucial

0

J FMAM J J A S OND Box 10

J FMAM J J A S OND


3

): see legend to Figure 3.

aspect in reproducing the observed seasonal evolution of the phytoplankton. The examples given above indicate the extent to which the available data, obtained over decades and aggregated as to month and location, can be used for model testing in terms of providing an accurate representation of the seasonality in the average parameter values. The large-scale survey data allow an evaluation of how well or how badly inter-annual variability is reproduced by the model when run with the physical forcing for the particular period. The high frequency data from the mooring allows insight into whether the interaction between physical forcing and the response of the biological system is properly represented in the model as well as the degree to which the biological processes are expressed correctly. To test the sensitivity of the model to the boundary conditions, the nutrient concentrations in the crossboundary inflowing water have been halved and the

762

J. W. Baretta et al. Box 1

10

J FMAMJ J A S OND Box 4

10

0

Box 3

5

0

J F M A M J J A S O N D Box 5

10

10

J FMAM J J A S OND Box 7

ND15 ND130 to ND15 Data

–3

Chl a (mg m )

10

5

5

0

Box 2

10

5

5

0 10

J FMAMJ J A S OND Box 8

5

0

0 10

5

J F M A M J J A S O N D Box 9

0 10

5

5

J FMAMJ J A S OND

0

J FMAM J J A S OND Box 10

J F M A M J J A S O N D

Figure 7. Chl a (mg m


3

results for Chl a are compared (Fig. 10) with results from the standard run (ND15). The effect is clearly most pronounced in the boxes close to the northern boundary and decreases towards the coastal boxes. The sensitivity of the model to the boundary conditions expresses itself even more clearly in the calculated net primary production (Table 1).

Discussion The data set compiled from the ECOMOD database into climatological monthly means for nutrients and Chl a clearly shows that the seasonal cycle represents the dominant signal, except for areas heavily influenced by river run-off. In particular, the monthly means for the Dutch continental coastal area (Box 8) and the German Bight (Box 9) show only a weak seasonality and the observed ranges in concentrations are mostly uniformly high throughout the year.

0

J FMAM J J A S OND

): see legend to Figure 3.

The availability of the climatological data set allowed us to test the ability of the 15-box model to reproduce the seasonal cycle in the nutrient dynamics and the regional variation therein. The coarse spatial resolution of this set-up, with its implicit assumption of large-scale homogeneity in all the observed properties, dictated aggregation of all measurements taken within one box, thereby hiding both temporal and spatial variability in the range around the mean. The model produces a spring bloom, even where the data do not indicate one. In contrast, the simulated decline in dissolved inorganic nutrients in spring matches the available nutrient data quite well. Within the model domain, advective transport cannot be held responsible for such a rapid decline in nutrient concentrations. Incorporation of nutrients in phytoplankton, and subsequently into the other components of the food web, appears to be the only likely explanation, and therefore our conclusion is that in reality there always is

Data needs for ecosystem modelling 10

Box 58

Chl a (mg m–3 )

10

10

5

5

0

Box 59

10

J FMAMJ J A S OND Box 65

0

763 Box 60

5

J F M A M J J A S O N D Box 66

10

0 10

J FMAM J J A S OND Box 67

ND130 Data 5

0 10

J FMAMJ J A S OND Box 68

0 10

J F M A M J J A S O N D Box 74

J FMAMJ J A S OND

0

0 10

J FMAM J J A S OND Box 75

5

5

5

0

5

5

J F M A M J J A S O N D

0

J FMAM J J A S OND


3

Figure 8. Chl a (mg m ): simulated concentrations in the 130-box set-up (ND130) for boxes 58–60, 65–68, 74, and 75 (comprising box 5/15 in ND15; cf. Fig. 2) compared with the median (+) and the upper (83%) and lower (17%) quartiles (bars) of the data values derived from the ECOMOD data set.

a spring bloom. The monthly time interval in this data set appears to be simply too large for the spring bloom to be sampled effectively. This view is supported by the large ranges in the Chl a measurements during spring, which suggest that the spring bloom is a local short-term event that may easily be missed at low sampling frequencies. Taylor and Howes (1994) state that the monthly measurement interval typical of large-scale ecosystem studies may result in 5- to 30-fold undersampling with respect to the high-frequency variations in productivity as well as in oxygen dynamics, and in estimates of annual production being off by 30%. The testing of the performance of the model in reproducing a specific seasonal cycle (1988–1989) with correct regional differences is possible with the NERC data set, but the repeat frequency of one month of the survey does not allow for the resolution of local phenomena such as phytoplankton blooms. Since the

physical parameters are also averaged over large areas and therefore damped, it is not possible to validate the model extensively as to its ability to properly respond to small-scale variability in physical forcing. However, the mooring data set does show the power of the model to respond to high-frequency variability in physical forcing along the vertical axis. During this project, both the major forcing functions (windspeed and heat flux) and the response of the biological system as represented by the biomass evolution of phytoplankton have been measured directly, in combination with regular nutrient concentration measurements. Such a comprehensive data set does allow testing whether the response of biological processes to physical forcing has been properly implemented and whether the model is sufficiently non-linear to reproduce the observed non-linearity of biological processes. This is only true if the mooring site is located in an area of low spatial heterogeneity and if

764

J. W. Baretta et al. 7

7 (b) 6

5

5

Chl a (mg m–3)

6

–3

Chl a (mg m )

(a)

4 3

4 3

2

2

1

1

0

J

F

M A M

J J 1994

A

S

O

N

D

0

J

F

M A M

J J 1994

A

S

O

N

D

Figure 9. Concentrations of Chl a (mg m
3) at 13 m (a) and 43 m depths (b) as simulated with the mixed layer physical model (from Ruardij et al., 1997). Average daily concentration as measured with a recording fluorimeter on the mooring. Grey area: modelled concentration.

Figure 10. Simulated concentrations of Chl a (mg m
3) in the standard 15-box set-up (ND15) and in the run with 50% reduced nutrient concentrations in the boundary conditions (50% bound).

Data needs for ecosystem modelling Table 1. Net annual primary production (gC m
2) in the standard model run (ND15) and in a run with nutrient concentrations in the boundary conditions halved (50% bound). The calculated reduction is given as a percentage of the net primary production in the standard run. Box

ND15

50% bound

Reduction in %

1 2 3 4 5 6 7 8 9 10

106.07 102.70 85.00 111.84 197.53 70.15 178.82 299.12 249.06 177.95

18.75 19.86 18.81 38.59 115.39 22.27 104.77 224.91 178.92 87.56

82.32 80.66 77.88 65.49 41.59 68.25 41.41 24.81 28.16 50.79



horizontal advection does not change the concentrations of the constituents appreciably. This paper has focused on nutrient and chlorophyll data, because these are the only biologically relevant data available on a system-wide basis. Corresponding time-series data on functional groups such as diatoms, autotrophic and heterotrophic flagellates, bacteria, microzooplankton, mesozooplankton, and benthic state variables for use in ecosystem models are rare. Some biomass data on these groups are available for the Southern Bight in the NERC survey data set and for mesozooplankton from the long-term CPR (Broekhuizen et al., 1995; Broekhuizen and McKenzie, 1995). However, a proper test of the biological concepts not only requires information on biomasses of functional groups but also on the fluxes within and between these groups, because such information would reveal the extent to which our definition of the interactions within the system is correct. The difficulty of measuring in situ turnover times of nutrients and biological state variables, in combination with the uncertainty in extrapolating spot measurements of primary productivity to seasonal net primary production, may still result in systematic over- or underestimates of the fluxes predicted by the model, even when standing stocks are predicted correctly. To put it quite bluntly, proper verification of model results requires more direct information on in situ biological rates in the system. Measurements from mesocosm experiments can be used to calibrate and validate biological process descriptions (Baretta-Bekker et al., 1994, 1998). Although boundary conditions are irrelevant in such closed systems, the initial concentrations are of crucial importance, as they determine the total pools of matter available.

Conclusions

A coarse vertical resolution into surface and bottom boxes suffices to reproduce the climatological sea-







765

sonal cycle in the North Sea as expressed in the distribution of inorganic nutrients and Chl a over the water column. To correctly reproduce short-term events such as spring blooms, a higher vertical resolution, combined with realistic physical forcing, is required. The frequent absence of a spring bloom in the Chl a data indicates that the sampling frequency in monitoring programmes is too low to reliably capture the temporal variability of phytoplankton biomass. The scarcity of data on the dynamics of bacterioplankton abundance, phytoplankton community composition, detritus, total particulate N, P, and Si, and oxygen saturation inhibits a more comprehensive testing of the model. The only way forward is to collect more information during mooring programmes and/or from mesocosm experiments. The further development of coupled hydrodynamic/ ecological models requires simultaneous measurements of physical, chemical, and biological variables at sampling frequencies that guarantee coverage of the entire short-term range of variation without sacrificing the duration of the sampling period required for quantifying seasonal changes. In future observational programmes, measurement of fluxes and rates should be emphasized.

Acknowledgements The ERSEM model development was funded by the EU-MAST programme under contract numbers MASTCT90-0021 and MAS2-CT92-0032. The preparation of this paper was partially funded by the EU-MAST programme under contract number MAS3-CT96-0058. The authors gratefully acknowledge the ERSEM partners for their contributions to the model.

References Baretta, J. W., Ebenho¨h, W., and Ruardij, P. 1995. The European Regional Seas Ecosystem Model: a complex marine ecosystem model. Netherlands Journal of Sea Research, 33: 233–246. Baretta, J. W., and Ruardij, P. 1988. Tidal flat estuaries. Ecological Studies, 71. Springer Verlag, Heidelberg. 353 pp. Baretta-Bekker, J. G., Baretta, J. W., Hansen, A. S., and Riemann, B. 1998. An improved model of carbon and nutrient dynamics in the microbial food web in marine enclosures. Aquatic Microbial Ecology 14: 91–108. Baretta-Bekker, J. G., Riemann, B., Baretta, J. W., and Rasmussen, E. K. 1994. Testing the microbial loop concept by comparing mesocosm data with results from a dynamical simulation model. Marine Ecology Progress Series 106: 187– 198. Billen, G. 1978. A budget of nitrogen recycling in North Sea sediments off the Belgian coast. Estuarine coastal marine Science 7: 127–146.

766

J. W. Baretta et al.

Billen, G., and Lancelot, C. 1988. Modelling benthic nitrogen cycling in temperate coastal ecosystems. In Nitrogen cycling in coastal marine environments, pp. 343–378. Ed. by T. H. Blackburn and J. So¨rensen, SCOPE. John Wiley, Chichester. Broekhuizen, N., Heath, M. R., Hay, S. J., and Gurney, W. S. C. 1995. Modelling the dynamics of the North Sea’s mesozooplankton. Netherlands Journal of Sea Research 33: 381–406. Broekhuizen, N., and McKenzie, E. 1995. Patterns of abundance for Calanus and smaller copepods in the North Sea: time series decomposition of two CPR data sets. Marine Ecological Progress Series 118: 103–120. Damm, P. E. 1989. Klimatologischer Atlas des Salzgehalts, der Temperatur und der Dichte in der Nordsee, 1968–1985. Technical Bericht 6–89, Institut fu¨r Meereskunde, Universita¨t Hamburg. 81 pp. Fransz, H. G., and Verhagen, J. H. G. 1985. Modelling research on the production cycle of phytoplankton in the Southern Bight of the North Sea in relation to river-borne nutrient loads. Netherlands Journal of Sea Research 19: 241–250. Kremer, J. N., and Nixon, S. W. 1978. A coastal marine ecosystem: simulation and analysis. Springer Verlag, Berlin/ Heidelberg/New York. 127 pp. Lenhart, H. J., Radach, G., Backhaus, J. O., and Pohlmann, T. 1995. Simulations of the North Sea circulation, its variability and its implementations as hydrodynamic forcing in ERSEM. Netherlands Journal of Sea Research 33: 271–299. Lenhart, H. J., Radach, G., and Ruardij, P. 1997. On the effects of river input on the ecosystem dynamics in the continental coastal zone of the North Sea using a box refined ERSEM model. Journal of Sea Research 38: 249–274. Levitus, S. 1982. Climatological atlas of the world ocean. NOAA Prof. Paper 13. US Government. Printing Office, Washington, DC. 173 pp.

Lowry, R., Cranmer, K., and Rickards, L. 1992. North Sea Project CD ROM and User Guide. British Oceanographic Data Centre, Natural Environmental Research Council, Swindon, UK. Mommaerts, J. P., Pichot, G., Ozer, J., Adam, Y., and Baeyens, W. 1984. Nitrogen cycling and budget in Belgian coastal waters: North Sea areas with and without river inputs. Rapports et Proce`s-verbaux des Re´unions du Conseil International pour l’Exploration de la Mer 183: 57–69. Radach, G., and Pa¨tsch, J. 1997. Climatological annual cycles of nutrients and chlorophyll in the North Sea. Journal of Sea Research 38: 231–248. Radford, P. J. 1996. The modules of the European Regional Seas Ecosystem Model II. Internal ERSEM report. 390 pp. Radford, P. J., and Joint, I. R. 1980. The application of an ecosystem model to the Bristol Channel and Severn Estuary. Institute of water pollution control. Annual Conference. Water Pollution Control 19: 244–254. Ridderinkhof, H. 1992. On the effects of variability in meteorecological forcing on the vertical structure of a stratified water column. Continental Shelf Research 12: 25–36. Ruardij, P., Van Haren, H., and Ridderinkhof, H. 1997. The impact of timing of the thermal stratification on phytoplankton and nutrient dynamics in shelf sea: a model study. Journal of Sea Research 38: 311–331. Steele, J. H. 1974. The structure of marine ecosystems. Harvard University Press, Cambridge, Massachusetts, USA. 128 pp. Taylor, C. D., and Howes, B. L. 1994. Effect of sampling frequency on measurements of seasonal primary production and oxygen status in near-shore coastal ecosystems. Marine Ecological Progress Series 108: 193–203. Van Aken, H. 1984. A one-dimensional mixed layer model for stratified shelfseas with tide- and wind-induced mixing. Deutsches Hydrographisches Zeitschrift 37: 3–27.