Same model, different answers? Structural uncertainty in Arctic climate change across downscaling approaches, forcings, and scenarios

High-resolution climate change projections are essential for assessing Arctic climate impacts and extreme events. Such projections are commonly obtained through statistical downscaling, dynamical downscaling, regional refinement of global climate models (GCMs), or other emerging high-resolution modelling frameworks. While dynamical methods allow physically consistent regional projections based on global forcing, they remain computationally expensive, limiting the number of GCMs and scenarios that can be explored. As a result, structural uncertainty associated with modelling strategy, choice of driving GCM, and emission scenario is often insufficiently quantified, particularly for Arctic climate change signals. A key challenge is that systematic comparisons of climate change signal patterns across different high-resolution modelling approaches are rare. This is partly because few modelling systems allow consistent simulations in global, regionally refined, and limited-area configurations using the same model physics. Consequently, it remains unclear to what extent differences in projected Arctic climate change arise from large-scale forcing versus the downscaling framework itself.

Here, we address this gap using the ICON (ICOsahedral Nonhydrostatic) modelling system in three complementary configurations: a global setup with uniform coarse resolution (~53 km), a globally variable-resolution configuration with enhanced Arctic resolution (~13 km) and coarse resolution elsewhere (~53 km), and an Arctic high-resolution limited-area mode (~11 km). Control and scenario simulations are available for multiple driving GCMs and emission pathways, enabling a targeted, storyline-based assessment of structural uncertainty, focusing on physically consistent responses to prescribed large-scale forcings. This experimental design allows us to disentangle uncertainties related to (i) the downscaling approach, (ii) the choice of large-scale forcing, and (iii) the emission scenario, while keeping model formulation consistent. We analyse Arctic climate change signals for near-surface air temperature and precipitation, focusing on seasonal mean responses. Pattern-based, multi-simulation comparisons are used to assess agreement and differences across simulations at both pan-Arctic and regional scales. This allows us to identify aspects of Arctic climate change that are robust across downscaling strategies, as well as regions where projected responses are particularly sensitive to model configuration or large-scale forcing, highlighting areas of enhanced structural uncertainty. Our results provide guidance for interpreting high-resolution Arctic climate projections and support targeted model selection for impact-oriented studies and regional climate assessments.