Markus Schartau

Assessment of skill and portability in regional marine biogeochemical models: Role of multiple planktonic groups

[1] Application of biogeochemical models to the study of marine ecosystems is pervasive, yet objective quantification of these models’ performance is rare. Here, 12 lower trophic level models of varying complexity are objectively assessed in two distinct regions (equatorial Pacific and Arabian Sea). Each model was run within an identical onedimensional physical framework. A consistent variational adjoint implementation assimilating chlorophyll-a, nitrate, export, and primary productivity was applied and the same metrics were used to assess model skill. Experiments were performed in which data were assimilated from each site individually and from both sites simultaneously. A cross-validation experiment was also conducted whereby data were assimilated from one site and the resulting optimal parameters were used to generate a simulation for the second site. When a single pelagic regime is considered, the simplest models fit the data as well as those with multiple phytoplankton functional groups. However, those with multiple phytoplankton functional groups produced lower misfits when the models are required to simulate both regimes using identical parameter values. The cross-validation experiments revealed that as long as only a few key biogeochemical parameters were optimized, the models with greater phytoplankton complexity were generally more portable. Furthermore, models with multiple zooplankton compartments did not necessarily outperform models with single zooplankton compartments, even when zooplankton biomass data are assimilated. Finally, even when different models produced similar least squares model-data misfits, they often did so via very different element flow pathways, highlighting the need for more comprehensive data sets that uniquely constrain these pathways.

Simultaneous data-based optimization of a 1D-ecosystem model at three locations in the North Atlantic: Part I— Method and parameter estimates

An optimization experiment is performed with a vertically resolved, nitrogen-based ecosystem model, composed of four state variables (NPZD-model): dissolved inorganic nitrogen (N), phytoplankton (P), herbivorous zooplankton (Z) and detritus (D). Parameter values of the NPZD-model are optimized while assimilating observations at three locations in the North Atlantic simultaneously, namely at the sites of the Bermuda Atlantic Time-Series Study (BATS; 31N 64W), of the North Atlantic Bloom Experiment (NABE; 47N 20W), and of Ocean Weather Ship-India (OWS-INDIA; 59N 19W). A method is described for a simultaneous optimization which effectively merges different types of observational data at distinct sites in the ocean. A micro-genetic algorithm is applied for the minimization of a weighted least square misfit function. The optimal parameter estimates are shown to represent a compromise among local parameter estimates that would be obtained from single-site optimizations at the individual locations. The optimization yields a high estimate of the initial slope parameter of photosynthesis (alpha), which is shown to be necessary to match the initial phases of phytoplankton growth. The estimate of alpha is well constrained by chlorophyll observations at the BATS and OWS-INDIA sites and likely compensates for a deficiency in the parameterization of light-limited growth. The optimization also points toward an enhanced recycling of organic nitrogen which is perceived from a high estimate for the phytoplankton mortality/excretion rate.

Parameter estimates of a zero-dimensional ecosystem model applying the adjoint method

Assimilation experiments with data from the Bermuda Atlantic Time-series Study (BATS, 1989¯1993) were performed with a simple mixed-layer ecosystem model of dissolvedinorganic nitrogen (N), phytoplankton (P) and herbivorous zooplankton (H). Our aim is to optimize the biological model parameters, such that the misfits between model results andobservations are minimized. The utilized assimilation method is the variational adjoint technique, starting from a wide range of first-parameter guesses. A twin experiment displayedtwo kinds of solutions, when Gaussian noise was added to the model-generated data. The expected solution refers to the global minimum of the misfit model-data function, whereasthe other solution is biologically implausible and is associated with a local minimum. Experiments with real data showed either bottom-up or top-down controlled ecosystemdynamics, depending on the deep nutrient availability. To confine the solutions, an additional constraint on zooplankton biomass was added to the optimization procedure. Thisinclusion did not produce optimal model results that were consistent with observations. The modelled zooplankton biomass still exceeded the observations. From the model-datadiscrepancies systematic model errors could be determined, in particular when the chlorophyll concentration started to decline before primary production reached its maximum. Adirect comparision of measured 14C-production data with modelled phytoplankton production rates is inadequate at BATS, at least when a constant carbon to nitrogen C : N ratio isassumed for data assimilation.

Modeling the speciation and biogeochemistry of iron at the Bermuda Atlantic Time‐series Study site

By means of numerical modeling, we analyze the cycling of iron between its various physical (dissolved, colloidal, particulate) and chemical (redox state and organic complexation) forms in the upper mixed layer. With our proposed model it is possible to obtain a first quantitative assessment of how this cycling influences iron uptake by phytoplankton and its loss via particle export. The model is forced with observed dust deposition rates, mixed layer depths, and solar radiation at the site of the Bermuda Atlantic Time-series Study (BATS). It contains an objectively optimized ecosystem model which yields results close to the observational data from BATS that has been used for the data-assimilation procedure. It is shown that the mixed layer cycle strongly influences the cycling of iron between its various forms. This is mainly due to the light dependency of photoreductive processes, and to the seasonality of primary production. The daily photochemical cycle is driven mainly by the production of superoxide, and its amplitude depends on the concentration and speciation of dissolved copper. Model results are almost insensitive to the dominant form of dissolved iron within dust deposition, and also to the form of iron that is taken up directly during algal growth. In our model solutions, the role of the colloidal pumping mechanism depends strongly on assumptions on the colloid aggregation and photoreduction rate.

Basin-scale performance of a locally optimized marine ecosystem model

A marine ecosystem model, that had previously been calibrated in a one-dimensional (1D) mode against observations at three time-series and process-study sites simultaneously, is coupled to a three-dimensional (3D) circulation model of the North and Equatorial Atlantic. Compared to an experiment with a previously employed subjectively tuned ecosystem model, the new 3D-model does not only reduce the model-data misfit at those locations at which observations entered the 1D optimization procedure, but also at an oligotrophic site in the subtropics that had not been considered in the 1D calibration. Basin-scale gridded climatological data sets of nitrate, surface chlorophyll, and satellite-derived primary production also reveal a generally lower model-data misfit for the optimized model. The most significant improvement is found in terms of simulated primary production: on average, primary production is about 2.5 times higher in the optimized model which primarily results from the inclusion of a phytoplankton recycling pathway back to dissolved inorganic nitrogen. This recycling pathway also allows for a successful reproduction of nonvanishing surface nitrate concentrations over large parts of the subpolar North Atlantic. Apart from primary production, the parameter optimization reduces root-mean-square misfits by merely 10–25% and remaining misfits are still much larger than observational error estimates. These residual misfits can be attributed both to errors in the physical model component and to errors in the structure of the ecosystem model, which an objective estimation of ecosystem model parameters by data assimilation alone cannot resolve.

The Ocean's Vital Skin: Toward an Integrated Understanding of the Sea Surface Microlayer

Despite the huge extent of the oceans surface, until now relatively little attention has been paid to the sea surface microlayer (SML) as the ultimate interface where heat, momentum and mass exchange between the ocean and the atmosphere takes place. Via the SML, large-scale environmental changes in the ocean such as warming, acidification, deoxygenation, and eutrophication potentially influence cloud formation, precipitation, and the global radiation balance. Due to the deep connectivity between biological, chemical, and physical processes, studies of the SML may reveal multiple sensitivities to global and regional changes. Understanding the processes at the oceans surface, in particular involving the SML as an important and determinant interface, could therefore provide an essential contribution to the reduction of uncertainties regarding ocean-climate feedbacks. This review identifies gaps in our current knowledge of the SML and highlights a need to develop a holistic and mechanistic understanding of the diverse biological, chemical, and physical processes occurring at the ocean-atmosphere interface. We advocate the development of strong interdisciplinary expertise and collaboration in order to bridge between ocean and atmospheric sciences. Although this will pose significant methodological challenges, such an initiative would represent a new role model for interdisciplinary research in Earth System sciences.