An important goal of WP2 was to develop mechanistic and predictive models for the distribution and metabolic activity of cold-water corals (CWCs) and deep-water sponges (DWS) and use these models to understand how their distribution is affected by the Atlantic Meridional Overturning Circulation (AMOC).
 Output from hydrodynamic models (VIKING20 or ROMS-Agrif) was used to simulate transport of reactive organic matter in the water column around CWC reefs of DWS grounds. The approach is inspired by Soetaert et al. (2016), in which su (...)
 An important goal of WP2 was to develop mechanistic and predictive models for the distribution and metabolic activity of cold-water corals (CWCs) and deep-water sponges (DWS) and use these models to understand how their distribution is affected by the Atlantic Meridional Overturning Circulation (AMOC).
 Output from hydrodynamic models (VIKING20 or ROMS-Agrif) was used to simulate transport of reactive organic matter in the water column around CWC reefs of DWS grounds. The approach is inspired by Soetaert et al. (2016), in which suspended organic matter dynamics above coral mounds was simulated. Here, we extend this methodology by having CWCs and DWS feeding on the suspended organic matter in the bottom layer using simple formulations for passive (CWC) and active (DWS) suspension feeding and metabolic activity. Physiological model formulation was based on data collected within ATLAS (Deliverables 2.1 and 2.2).
 We focus on three ATLAS Study regions: 1) large CWC mounds, dominated and formed by the scleractinians Lophelia pertusa and Madrepora oculata, in the Logachev mound province in the south-eastern section of Rockall Bank, 2) coral gardens, dominated by the soft-coral Viminella flagellum, on Condor seamount, and 3) extensive sponge grounds, dominated by Geodia barretti along the east Canadian shelf break in Davis Strait.
 We faced considerable computational challenges when developing the coupled models. CWC and DWS growth is slow, which implies that long simulation periods are needed to reach a (dynamic) steady state. Long simulation periods are not feasible given the high spatial and temporal (i.e. with tidal dynamics) resolution of the models. A 3-step solution procedure is proposed to tackle this issue, in which in step 1 initial suspended organic matter (OM) concentrations in the water column are calculated. In step 2, the bottom layer concentrations from step 1 are used to calculate initial concentrations for CWCs or DWS. In step 3, the coupled model is run with suspended organic matter (step 1) and CWC or DWS (step 2) as starting conditions. This approach sufficed for most of the model applications, but we acknowledge that some regions in the different model domains have not yet reached a (dynamic) steady state.
 The coupled models, based on hydrodynamics, organic matter biogeochemistry and physiology of reef-forming organisms, successfully predicted the coral and sponge distribution and biomass in the three case study areas and thereby provide a new mechanistic tool to understand the distribution (see figure below) and metabolic (not shown) activity of hotspot ecosystems.
 A striking result for Rockall Bank and Condor Seamount was that the suspended organic matter concentration in the bottom layer of the model domain was heavily modified by the passive suspension feeding CWCs. The initial PSF biomass (step 2) immediately depleted the organic matter concentration in the bottom layer to near zero across the whole model domain (see figure of Condor seamount below). As a result, the remaining organic matter concentration was insufficient to meet demands, which invoked a slow but steady reduction in PSF biomass over time. We conclude that the impact of PSF on bottom layer suspended OM concentration extends over large areas of the seafloor, including regions where the natural biomass is low.
 The distribution of CWC at Rockall Bank and Condor seamount could be accurately modelled with suspended organic matter being parameterized as labile, fast-sinking organic matter, e.g. labile marine snow and zooplankton faecal pellets. The relatively fast sinking rate of this organic matter, gives a relatively low concentration in the water column, but the high current velocities around coral mounds ensure sufficient interception by the passive suspension feeding CWCs.
 In contrast, the concentration of labile, fast-sinking organic matter OM proved grossly insufficient to meet the carbon demands of the active suspension feeding DWS. Only when we parameterized the suspended organic matter as slow sinking, relatively refractory organic matter the ambient concentration was sufficient to allow growth of DWS. This organic matter is likely characterised by smaller particles (microbial and [colloidal] DOM). The modelled DWS distribution matched field observations substantially better with slow-sinking organic matter as opposed to predictions based on fast-sinking organic matter. Experimental work (Deliverable 2.2) already hinted at these different feeding preferences between active and passive suspension feeders. We hypothesize that CWC (i.e. passive suspension feeders) and DWS (i.e. active suspension feeders) distribution on shelf breaks and slopes can be explained by a niche separation based on organic-matter type.
 Model simulations for different AMOC states were run for each of the three case study areas. We cannot conclude from the results to what extent AMOC influences the biomass of CWC and DWS. As mentioned in , it proved challenging to reach a (dynamic) steady state for the models. As a result, it remained unclear whether the small differences in hydrodynamics between AMOC states truly governed differences in biomass development. In addition, tidal dynamics proved important for the transport of organic matter to the CWCs and the tidal forcing is not influenced by AMOC. We do however believe that the models are well suited for the exploration of mechanistic relations between distributions of CWCs and DWS and for example reductions in export of organic matter or changes in the type of exported organic matter.