We have

estimated our rate kphot from Powell and Wilson-Finelli (2003), but study the sensitivity to this parameter further in Section 5. This holds also for the fraction of ligands that undergoes aggregation pcol, which we assume to be 0.5 in the reference experiment. We assume that uptake always destroys ligands, Fluorouracil manufacturer i.e. that the fraction of ligands that is on average destroyed when phytoplankton cells take up iron pupt is one. The way that iron is modeled in different ocean biogeochemical models differs considerably, and it is conceivable that this has as large an effect on the modeled ligand distributions as varying the ligand model parameters. To obtain an idea on the sensitivity of our model results to the underlying biogeochemical model, we therefore present here results obtained with two different global biogeochemical PFT�� mouse models, with the same formulation for ligand dynamics. The two models are

PISCES (Aumont and Bopp, 2006 and Tagliabue et al., 2014), and REcoM (Hauck et al., 2013); both have been described elsewhere, but without a prognostic ligand. Both represent phytoplankton by two functional groups, diatoms and nondiatoms, and also have compartments for zooplankton and dead organic matter (detritus). REcoM is slightly simpler in that it resolves only one zooplankton and detritus class, while PISCES has two. On the other hand, REcoM allows for decoupling of the carbon and nitrogen cycling by allowing deviations of the cellular stoichiometry from the classical Redfield ratio. For a full description of the model we refer the reader to Tagliabue et al. (2014) and Hauck et al. HSP90 (2013); instead we focus here on the added component organic ligand, whose dynamics are described equally in both models. With each of the models we performed one standard model run, which was the outcome of a number of previous sensitivity studies. The ligand model parameters belonging to these standard runs (Table 1) are identical, except that REcoM uses half the ligand to carbon ratio that PISCES does; this was deemed necessary to produce realistic surface ligand concentrations, and — as we will discuss in the next section — can be traced back to a different

emphasis placed on POC remineralization and DOC excretion in the two models. To elucidate some of the dependencies of model outcomes to some uncertain parameter values, we also did a series of sensitivity experiments with one of the models, REcoM, only. In these experiments the parameters for ligand:carbon ratio rL:C (runs L2C1 and L2C2), for photochemistry kphot (runs PHOT1 and PHOT2), and for the fraction of ligands undergoing aggregation pcol (runs COL1 and COL2) are varied. Parameter values for these runs are also documented in Table 1. Both models were integrated for 2000 years with annually repeating atmospheric forcing, starting from a uniform ligand concentration (0.6 nmol L− 1 in the case of PISCES, 1.0 nmol L− 1 for REcoM).