Description: The trace metal iron is considered to be the nutrient that limits marine primary production in one third of the global surface ocean (Martin, 1990; Boyd et al., 2007; Moore et al., 2013). It is also the nutrient that maintains future ocean fertility due to its irreplaceable role in the process of nitrogen fixation, which adds “new” nitrogen (another nutrient for phytoplankton) to the surface ocean (Raven, 1988; Kustka et al., 2003b; Zehr and Capone, 2020). Due to iron’s importance, it is not surprising that the demand for incorporating iron into global biogeochemical models is high. However, including iron in an earth system model has been shown to have no clear benefits with respect to model misfit against observational data (Nickelsen et al., 2015) . How smart is it then to introduce iron models into global biogeochemical models, when the benefits are not clearly identifiable? Especially, when the iron models perform poorly at reproducing observed iron patterns in the ocean (Tagliabue et al., 2016). The poor performance of iron models, coupled with their failure to improve biogeochemical tracer representation of the ocean, inspired this additional effort to identify the advantages of including iron in a global biogeochemical model, both for the preindustrial state and under conditions of a changing climate. The working hypothesis was that the relatively poor performance of iron models might come from inadequate model calibration. A first sensitivity study on biogeochemical model parameter values was conducted in order to identify key parameters for model calibration. It was found that while some of the parameters influence simulated nitrogen, phosphorus, and oxygen concentrations, few parameters influence simulated iron concentrations. This suggests that our modelling skill of the iron cycle is still limited and/or that the observational data base is insufficient for comprehensive model calibration so far. Thus it was decided not to include iron data in further model calibration. A model calibration framework (Kriest et al., 2017) was next applied to a hierarchy of global models with different implementations of iron; one without iron, one with prescribed iron concentrations, and another one with a dynamic iron cycle. Using calibration against global data sets of nitrogen, phosphorus, and oxygen, the misfit of each model was pushed to its minimum. It was found that under an assumed preindustrial steady state, the calibrated model with a full dynamic iron cycle has the lowest model misfit against observations (thus confirming the working hypothesis). It was also found that the calibrated model with a fully dynamic iron cycle has 50% less net primary production (which is closer to empirical estimations) compared to the calibrated model without iron. Finally, transient simulations for all calibrated models were integrated from their pre- industrial state until the end of the 21st century using an atmospheric CO2 concentration pathway consistent with a ’business-as-usual’ CO2 emission scenario. It was found that nitrogen fixation trends diverge among models. This divergence is caused by whether iron limits the productivity of the upwelling regions, e.g. in the eastern tropical Pacific. The export production in the eastern tropical Pacific (and other tropical upwelling regions) reacts differently to warming, depending on whether iron is a limiting nutrient. These different responses trigger a divergent chain of downstream responses that affect nitrogen fixation across the tropical oligotrophic regions in the model. Through the comparison between calibrated models, this thesis quantifies the advantages of including iron in a global biogeochemistry model and reveals how important iron is for future nitrogen fixation trends. It furthermore illustrates the interconnection between tropical upwelling and oligotrophic regions.