Maritime transport is a vital sector for global trade and the world economy. Particularly for islands, there is also an important social dimension of this sector, since island communities strongly rely on it for a connection with the mainland and the transportation of goods and passengers. Furthermore, islands are exceptionally vulnerable to climate change, as the rising sea level and extreme events are expected to induce severe impacts. Such hazards are anticipated to also affect the operations of the maritime transport sector by affecting either the port infrastructure or ships en route. The present study is an effort to better comprehend and assess the future risk of maritime transport disruption in six European islands and archipelagos, and it aims at supporting regional to local policy and decision-making. We employ state-of-the-art regional climate datasets and the widely used impact chain approach to identify the different components that might drive such risks. Larger islands (e.g., Corsica, Cyprus and Crete) are found to be more resilient to the impacts of climate change on maritime operations. Our findings also highlight the importance of adopting a low-emission pathway, since this will keep the risk of maritime transport disruption similar to present levels or even slightly decreased for some islands because of an enhanced adaptation capacity and advantageous demographic changes.

Seasonal forecasting systems still have difficulties predicting temperature over continental regions, while their performance is better over some maritime regions. On the other hand, the land surface is a substantial source of (sub-)seasonal predictability. A crucial land surface component in focus here is the snow cover, which stores water and modulates the surface radiation balance. This paper’s goal is to attribute snow cover seasonal forecasting biases and lack of skill to either initialization or parameterization errors. For this purpose, we compare the snow representation in five seasonal forecasting systems (from DWD, ECMWF, Météo-France, CMCC, and ECCC) and their performances in predicting snow and 2-m temperature over a Siberian region against ERA5 reanalysis and station data. Although all systems use similar atmospheric and land initialization approaches and data, their snow and temperature biases differ in sign and amplitude. Too-large initial snow biases persist over the forecast period, delaying and prolonging the melting phase. The simplest snow scheme (used in DWD’s system) shows too-early and fast melting in spring. However, systems including multi-layer snow schemes (Météo-France and CMCC) do not necessarily perform better. Both initialization and parameterization are causes of snow biases, but, depending on the system, one can be more dominant.

Climate models and their land components still exhibit notable discrepancies in frozen soil simulations. Contrasting the historical runs of seven land-only models of the Land Surface, Snow, and Soil Moisture Model Intercomparison Project (LS3MIP) with their Coupled Model Intercomparison Project Phase 6 (CMIP6) counterparts allowed quantifying the contributions of the land surface parameterization scheme and the atmospheric forcing to the discrepancies. The simulation capabilities were assessed using observational data from 152 sites in Siberia and reanalysis data. In the winter months (December, January, and February), the LS3MIP ensemble bias in 0.2 m soil temperature was larger than the CMIP6 bias (−3.6 vs. −2.7 °C). The spread of winter 0.2 m soil temperatures was also larger in the LS3MIP ensemble (4.6 °C) than in the CMIP6 ensemble (3.0 °C). For permafrost sites, for all CMIP6 simulations, the correlations between winter soil temperatures with observations were below 0.6, and the correlations for spring/autumn correlations of snow depth were below 0.8. In the CMIP6 simulations, the median 0.2 m soil temperature was 0.3 °C warmer than in the observations when the simulated soil temperature dropped below −5 °C. However, the LS3MIP simulations were colder, with a cold bias in the median of 0.7 °C. The biases of 2 m temperature in coupled simulations had an opposite sign and were amplified in magnitude compared to the biases of their soil temperatures, especially below 0 °C. Our results indicate that land-only models have limited capability in reproducing soil temperatures and snow depth under severe cold conditions (surface air temperature below −15 °C). Furthermore, four climate models and their land components underestimated the insulating role of snow. In cases with shallow snow depth (0–0.2 m), the models simulated air-soil temperature differences of up to 10 °C, whereas in situ measurements indicated even larger differences. The CMIP6 models tended to compensate for errors in their land component with errors in the atmospheric model component. Therefore, to improve frozen soil modeling in climate projections, a more accurate representation of the surface-soil insulation is essential.

Snow cover is a critical element of the Earth’s climate system, influencing energy exchange and water storage. This study examines the sensitivity of seasonal snow simulations to initialization, parameterization, and atmospheric forcing to assess the influence on snow representation in seasonal simulations. Using the TERRA Standalone (TSA) land surface model with ERA5 reanalysis forcing, we investigate the effects of perturbations in initial snow water equivalent (SWE), albedo (parameterization), and precipitation (forcing) affect snow dynamics across Eurasia over ten winters. A +10% perturbation in the snow albedo during 3-month spring simulations leads to a median increase in the snow cover area of 24%, which closely matches the interannual variability. Similarly, a +10% perturbation in precipitation during early winter simulations leads to a median of 28% increase in total snow mass, more than three times the interannual variability. Sensitivity analysis shows that seasonal snow simulations are most sensitive to precipitation perturbations, especially during the melt season, while initialization effects persist throughout the winter. The results show that snow cover during the melt season is strongly influenced by albedo and precipitation perturbations, with cumulative positive perturbations having a larger impact than negative perturbations. Mid- and late-winter initializations effectively reduce snow bias from precipitation and albedo perturbations. In addition, improved albedo parameterization can improve snow simulation accuracy during the melt season. Spatial analysis identifies snow margin zones as particularly sensitive to albedo and precipitation perturbations. These findings underscore the importance of accurate precipitation forcing and snow parameterization to ensure reliable seasonal snow simulations for climate predictions and projections. Graphical Abstract The graphical abstract provides a visual summary of the study setup and results generation, providing the reader with an efficient way to grasp the essence of the research. The first section highlights the input data, where ERA5 reanalysis forcing and initialization are used for the TERRA Standalone (TSA) land surface model to simulate snow cover dynamics over Eurasia for 10 winters and with perturbations. The second section illustrates the simulation timeline, showing how the experiments are conducted with different starting months ranging from October to May. The final section presents the sensitivity analysis of the perturbing factors: initialization (SWE), parameterization (albedo), and atmospheric forcing (precipitation). The figures show the effects of the perturbations, such as how changes in precipitation and albedo affect snow mass and snow cover area over winter and spring. The maps also highlight snow margin zones as areas of increased sensitivity to these factors. This graphical summary highlights the study’s contribution to understanding snow simulation sensitivity and the importance of accurate initialization, parameterization, and forcing in seasonal simulations.

Snow cover is a critical component of the Earth's climate system, covering up to 44 % of the Northern Hemisphere's land during winter and influencing energy exchange, water storage, and atmospheric circulation patterns. However, operational seasonal forecasting systems and climate models still encounter challenges in accurately representing snow processes. This leads to biases in temperature predictions and frozen soil simulations. This cumulative thesis investigates snow representation across different modeling frameworks to identify sources of error and quantify sensitivities. The first study evaluated five operational seasonal forecasting systems over Siberia, distinguishing between initialization and parameterization errors. The results showed that initial snow biases persisted throughout the forecast period, influencing snow melt timing and duration. Although multi-layer snow schemes offer theoretical advantages, they did not outperform single-layer schemes among the five systems due to initialization challenges. Both initialization and parameterization contributed to snow biases, with their relative importance varying by system. The second study used the TERRA Standalone land surface model to quantify the sensitivity of snow simulations to initialization, parameterization, and atmospheric forcing. A +10 % precipitation perturbation during early winter simulations increased total snow mass by 28 % (more than three times the interannual variability), while a +10 % albedo perturbation during spring simulations increased the area of snow cover by 24 %. Snow margin zones were particularly sensitive to albedo and precipitation perturbations, with positive perturbations having larger effects than negative ones due to the snow/no-snow threshold. The third study compared the Coupled Model Intercomparison Project Phase 6 (CMIP6) and Land Surface, Snow, and Soil Moisture Model Intercomparison Project (LS3MIP) models to evaluate frozen soil simulations in Siberia. LS3MIP showed a higher winter soil temperature bias (-3.6 °C) than CMIP6 (-2.7 °C), indicating that coupled models partially compensate for errors in the land surface component. All models underestimated the effects of snow insulation, with temperature differences of up to 10 °C less than observed at shallow snow depths (below 0.2 m). Models with updated snow thermal conductivity formulations demonstrated improved insulation representation. Together, these studies highlight the importance of accurately representing snow in climate and seasonal forecasting systems. The results demonstrate that accurate snow initialization is essential for reliable seasonal snow forecasts. Precipitation forcing and albedo parameterization dominate uncertainties in snow mass and snow cover area, and improving parameterization, especially for snow thermal conductivity and snow cover fraction is critical for realistically simulating frozen soil conditions in climate models.

Seasonally frozen ground (SFG), a critical component of the cryosphere, covers approximately 50% of the Northern Hemisphere’s land area and significantly influences land surface and overlying atmospheric processes. The coexistence of ice and liquid water in soil alters its hydraulic and thermal properties, thereby influencing the distribution of water and energy fluxes. Accurately modeling parameters such as soil moisture, soil temperature, snow depth, etc., is particularly challenging due to processes like soil freezing and subsequent thawing. Variability in model performance remains a key challenge, highlighting the need for improved representation of SFG in cold regions.The study compares standalone global simulations from two land surface models (LSMs) - JSBACH (Jena Scheme for Biosphere-Atmosphere Coupling in Hamburg) and CLM (Community Land Model) driven by 3-hourly ERA5 reanalysis data from 1941 to 2022, with a focus on the representation of SFG across the mid- to high latitudes of the Northern Hemisphere. CLM produces a more realistic distribution of frozen ground, whereas JSBACH significantly underestimates the extent of SFG, and ERA5-Land underrepresents permafrost extent. All models tend to underestimate SFG compared to the IPA (International Permafrost Association) map, with JSBACH showing the largest deviation. Global gridded outputs from JSBACH and CLM, along with ERA5-Land data, were intercompared to quantify differences between models. Results indicate that JSBACH performs poorly in cold regions, possibly due to issues in its multilayer snow scheme and the representation of snow thermal properties, which may contribute to underestimated snow depth and lower soil temperatures at 20 cm depth. In contrast, JSBACH performs better than CLM in mid-latitude snow-free regions, particularly over the Tibetan Plateau. Additionally, site-level analyses focusing on the onset and duration of SFG and snow cover revealed notable biases in simulating SFG onset and duration across all models due to snow insulation effects. The modal value of the soil-air temperature difference PDF (Probability Density Function) increases by ~5°C from shallow to thick snow, reflecting the stronger insulating effect of thicker snow. CLM tends to overestimate the insulating effect of snow, while JSBACH underestimates it. The study highlights the challenges of simulating SFG processes in cold regions, emphasizing the need to improve the representation of freeze-thaw dynamics through the thermal and hydraulic properties of soil and snow in land surface models (LSMs) to enhance model performance.