We asked the participants for the most problematic ocean model bias allowing a single choice only aiming for a clear emergence of the most pressing issue from the survey. About a quarter identified processes of the Antarctic continental margin (shelf seas, slope current, ice shelf cavities) as requiring most attention (Fig. 1). Further, open ocean deep convection and water mass transformation [both 14.3 %] and mixing were highlighted. Individual free text answers mentioned deep ocean circulation, modeling of biogeochemical cycles and the carbon pump, planetary boundary layer of ice-covered seas, and impacts on benthic ecosystems as other major model biases, which are not displayed in Fig. 1. The respondents related such biases in particular to global coupled climate [38 %] and ocean models [28 %] in general such as those used for the Climate Modelling Intercomparison Project (CMIP); this may, however, reflect the dominant area of expertise of the participants. Moreover, the respondents identified a dozen specific ocean models and state estimates as well as specifically high-resolution model versions that include some of these major biases (see published survey data for details; Martin et al., 2025). We emphasize that moving towards finer grid resolution alone may yield individual improvements but will not solve all the biases – as is documented in Moreno-Chamarro et al. (2022).

It is important to understand that open ocean deep convection – while physically not unrealistic (Gordon, 1978) – is dramatically overestimated in many climate models, specifically those of coarse resolution, with consequences for ocean to atmosphere heat redistribution, sea ice coverage, bottom water characteristics, and eventually also internal climate variability (Reintges et al., 2017; Heuzé, 2021). In reality, most Antarctic Bottom Water (AABW) is formed on the continental shelf where sea ice formation and ice shelf melt play key roles in the transformation of upwelled deep water (Silvano et al., 2023) and so does mixing for its transformation into Antarctic Intermediate Water (AAIW) further equatorward (Li et al., 2022). To this end, nearly three-quarters of all participating experts pointed out biases that are inter-connected and play an imminent role in the formation of water masses, such as AABW and AAIW, crucial for the global overturning circulation and for the natural sequestration of heat and anthropogenic carbon.

Hence, it is no surprise that implementation of ice shelf cavities [20 %], convection parameterization [18 %], scale aware (mixing) parameterizations [15 %] and overflow parameterization [13 %] were listed as most urgent model development targets. There has been remarkable advancement in these directions over the past two decades (e.g., De Rydt et al., 2024; Legg et al., 2009; Bruciaferri et al., 2024). It has been demonstrated that these new developments can mitigate model biases in the Southern Ocean even at relatively coarse resolution despite remaining issues (e.g. for ice shelf cavities see Hutchinson et al., 2023). However, we believe there is often significant delay or inaction in implementing such advancements, as model development is rarely funded directly by dedicated research projects. The push for rapid research outcomes tends to favor easily implemented targets – the “low-hanging fruit” – over more complex, long-term efforts.

This behavior is evident in the responses on near-term model evolution. Among those, increasing model complexity [27 %] and spatial resolution [24 %] stand out. Other goals such as improving or developing novel parameterizations and including artificial intelligence based modules are only considered by 11 %–13 % of the participants. In this case multiple answers were possible and participants ticked or listed 2–3 responses on average. It appears that preference is given to model complexity – evolving climate models into Earth system models by coupling more components, for example, ice sheets or biogeochemistry modules – over improving model physics. However, we think this could also be a sign of a more diversified, cross-disciplinary science landscape. As computing power keeps growing, resolving model issues by enhancing grid resolution appears to be a possible avenue to reduce biases (e.g., Rackow et al., 2022). But this is a costly option and impractical for applications on centennial time scales since proper representation of mesoscale dynamics in the high latitudes of the Southern Ocean requires grid spacing of 1/8–1/20° (and finer on the continental shelf), to properly resolve the Rossby radius (Hallberg, 2013, their Fig. 1). We suggest that new observations supporting model development could thus lead to improved and yet affordable simulations on a large range of spatial and temporal scales.

Reducing the major model biases and advancing ocean and climate models as laid out above will be essential to address the key research topics identified in the survey responses. Freshwater, heat and carbon budgets are high on the scientific agenda of the modelling community (Fig. 2). Questions on process understanding and future evolution of the Antarctic ice sheet, its ice shelves and their interaction with the ocean through heat and meltwater dominate the results [40 %]. While this is research at the continental margin, heat and carbon uptake where the low latitude Southern Ocean plays a major role was named second [19 %], followed by interest in the recent and future sea ice trends [14 %]. On the one hand, these results are somewhat biased by the research areas of the participants. On the other hand, scientific interest has migrated poleward in the Southern Ocean, where major challenges have been identified, such as knowledge gaps in ice–ocean interaction affecting global sea level rise projections, and where new observational techniques for under-ice sampling and mesoscale ocean simulations have become available.

The results suggest that oceanic processes themselves, such as dynamics from mesoscale eddies to large-scale circulation, tides, waves and mixing are not “big questions” by themselves anymore despite remaining issues and their important role in current “grand challenges” like ice–ocean interaction, warm water intrusion onto the continental shelf and biogeochemical modeling. Further, we assume that the underrepresentation of biogeochemical and ecosystem research as well as atmospheric process understanding, most prominently clouds and aerosols, is likely a consequence of questionnaire design and the focus group addressed.

Before diving into a discussion on the observational needs of the modelling community, we would like to call attention to the data sources actually used. There is an unbroken preference by modellers to use gridded data products [28 %], i.e. statistically interpolated fusions of observations from various sources, and reanalysis or state estimates [21 %], which are based on a numerical model and incorporate observations through, e.g. assimilation techniques. Likely also due to their extensive spatial coverage, satellite-borne remote sensing products are favored as well [15 %]. Data from ship-borne instruments, moorings, and floats appear to be used less often directly [10 %–13 %]. We acknowledge that observational data of all kinds, in particular including in situ data, feed into the gridded products. Nevertheless, it is important to note that modelers tend to validate their simulations against these kind of “observations”, which in fact are advanced data products and rather not viewed as actual observations by the observing, sea-going science community. Further, any in situ data that are not included in such gridded products is likely less used or even overlooked by modelers and thus does not contribute as much to the improvement of models.

Based on our personal experience, modelers tend to lean on derived products, such as reanalyses and state estimates, because (1) observations are typically sparse and direct comparisons require complex subsampling of model output, (2) formats and platforms used to share observational data are still not unified and easily accessible to users despite ongoing efforts, and (3) original measurements require advanced background knowledge and interpretation for use in model validation. These issues can be mitigated by, for instance, providing open-source validation packages along with the measurements and the development of virtual instruments, i.e. tool boxes allowing for the subsampling of model output in ways emulating specific observational instruments. Further, we suggest that communication between the science communities should be strengthened. Additionally, educational programs, such as summer schools, could introduce instruments, proper data handling and aforementioned tool boxes to the next generation of modelers.

For reducing observational gaps, especially with mitigating model biases in mind, modelers should be given and should take the opportunity to provide input to evolving observational programs early in the planning phase. In the end, this would ensure multiple use of the data collected. Figure 3a shows observations desired for bias mitigation with in situ observations clearly dominating over remote sensing data. This preference can be interpreted as a need for better process understanding in which in situ data are considered actual “ground truth” and often provide higher resolution in space and time, which is useful in several aspects, such as model validation, identifying of processes resolved at a given grid spacing, improving model parameterizations, etc. Another interpretation is that modellers are well aware of in situ measurements being crucial for better quality gridded products. And while such products are preferred in the actual validation process, the dire need for more ground truth data in a changing climate is acknowledged and its collection valued.

The scientific goals and observational plans of Antarctica InSync appear to be very much in line with the needs of the modelling community wishing for year-round data, especially in ice-covered seas and combining physical and biogeochemical measurements (Fig. 3a). The strong desire for winter observations and year-round monitoring in the Southern Ocean [29 %] is linked nearly equally to all major biases identified (Fig. 3b). As is already discussed by the Antarctica InSync community, building capacity for sustained monitoring in preparation for the International Polar Year in 2032–2033 and beyond would also strongly support model improvement and advancement [25 %]. Observations of the upper ocean stratification and surface fluxes would enable a better process understanding of the mixed layer and help to constrain vertical/diapycnal mixing parameterizations in models. Interestingly, the role of other climate system components causing biases in the ocean, for example, sea ice and snow, clouds and radiative processes, was not highly considered. This is likely owed to the limited choice of pre-defined answers we provided and the behavioral bias of the respondents preferring to tick one of those rather than entering individual answers. Nevertheless, we take this as an opportunity to point out the need for improved fundamental understanding and acknowledgement of coupled mechanisms and feedbacks within and beyond the focus group. Having better records of magnitude and variability of Southern Ocean surface fluxes of both physical and chemical quantities will help significantly. Similarly, any observations in support of heat and freshwater budgets, especially with a focus on ice–ocean interaction, will be instrumental in advancing models and improving climate projections. And last but not least, high-resolution bathymetry data of the Southern Ocean from the Antarctic Circumpolar Current to the continental margin and into the ice shelf cavities are urgently needed. Topography is a key ingredient for realistic simulations of the ocean circulation, specifically the import of Warm Deep Water and the export of dense bottom waters, and therefore crucial for reliable projections of ice shelf melting.