Climate change impacts on future driving and walking conditions in Finland, Norway and Sweden

Road weather is a major concern for the public safety and health, industries and transport sectors. Half of the yearly 27,000 road and 50,000 pedestrian injuries in Finland, Norway and Sweden can be traced back to slippery road and walkway conditions. We simulated the climate change impacts on future roads and walkways for mid- and end-century in Finland, Norway and Sweden with the road weather model RoadSurf, driven by the regional climate model HCLIM38 with boundary data from two global climate models following the RCP8.5 scenario. Our simulations for mid-century suggest strong road surface temperature increases, especially in southern Finland (+ 5.1 °C) and Sweden (+ 7.1 °C). Snowy and icy road surface conditions decreased by 23 percentage points, causing 18.5 percentage points less difficult driving conditions during the cold season. Zero-degree-crossing days mostly decreased in autumn and spring by up to 7 days and increased in winter by up to 5 days. Sidewalks mostly showed a decrease in slipperiness, but a five percentage point increase of water above ice layers on the sidewalks in winter, suggesting the slip-season might become shorter, but more slippery. Our results are upper extreme estimates but can serve as a reference to help local decision-makers plan mitigation and adaptation measures ahead of time.


Introduction
Weather plays a crucial role for the transportation of passengers and goods (Norrman et al., 2000). Traffic accidents involving personal injury or death due to slippery conditions account for up to 50% of the over 27,000 injuries in recent years in Finland, Norway and Sweden (FNS). Medical costs and economic production losses arising from these accidents sum up to €16.2 billion yearly (European Commission Directorate-General for Mobility and Transport & CE Delft, 2019; International Traffic Safety Data and Analysis Group, 2020a, b, c, d;Statistics Finland, 2021;Statistics Sweden, 2019). An often-overlooked burden is posed by pedestrian slip injuries. A majority of the 50,000 slips that need medical attention yearly are caused by slippery conditions (Elvik et al., 2009;Elvik & Bjørnskau, 2019;Hippi et al., 2020;Port and Ocean Engineering under Arctic Conditions (POAC) (2009).
These accidents cannot be prevented even with full winter-time road maintenance (Norrman et al., 2000). Therefore, reducing accident numbers by informing the public about hazardous road weather is important for the public health sector. Road maintenance and transport sectors benefit from road weather forecasts as well, for planning ahead their actions and reducing costs (Juga et al., 2013;Keskinen, 1980;Nurmi et al., 2013).
Currently, public authorities are working with real-time road weather observations and short-term weather forecasts to issue warnings when the road weather is expected to worsen. Road weather models typically predict the road surface temperature and the amount of snow, ice or water on the surface, and in some cases also friction and driving condition (Juga et al., 2013;Kangas et al., 2015). Regional road weather models have been developed in several countries to improve the regional accuracy (Chapman et al., 2001;Crevier & Delage, 2001;Fujimoto et al., 2012;Jacobs & Raatz, 2007;Kangas et al., 2015;Yang et al., 2012).
As climate change has been posing far-reaching challenges on the transport sector worldwide (Forzieri et al., 2018;Hori et al., 2018;Matthews et al., 2017), road planners and maintenance operators are assessing the monetary funds needed to manage (either current or soon-arising) regional climate change impacts on the transport sector. Such challenges include shifts in more frequent extreme weather, safe transport routes, time and locality of accidents, shifts in tourism and agriculture and pavement material damage and road maintenance changes. (Andersson & Chapman, 2011a;Axelsen et al., 2016;Balston et al., 2017;Koetse & Rietveld, 2009). Particularly in high northern latitudes like Northern Europe or Canada, where winter road maintenance is both a necessity and a big cost point, climate change is estimated to have stronger impacts due to arctic amplification (Screen, 2014).
Whilst weather models produce road weather forecasts for the near future and help manage immediate risks, they are not capable of portraying long-term developments beyond a few weeks. Climate models on the other hand give estimates for many decades or even centuries ahead and can simulate different climate change scenarios. Driving road weather models with climate models therefore can help road management and authorities to better plan ahead for the arising challenges and opportunities of road management in the far future (Matthews et al., 2017;McSweeney et al., 2016). Our predecessor study by Toivonen et al. (2019) confirmed that the Finnish road weather model RoadSurf (Finnish Meteorological Institute, Kangas et al., 2015) produced accurate historical road temperature estimates for Finland when driven with the cycle 38 of the regional climate model HARMO-NIE-Climate (HCLIM38) and thereby set the grounds for this study.

Data sources and models
Our main model was the road weather model RoadSurf (Kangas et al., 2015) driven by regional climate model simulations from the cycle 38 of HARMONIE-Climate (HCLIM38; Belušić et al., 2020). Since RoadSurf cannot produce its own climate scenario, it needs external data to provide the meteorological boundary data. To provide these data, we used three different sources: one reanalysis dataset and two CMIP5 global climate models, EC-Earth (ECE) and GFDL that considerably differed in their climate change predictions for the Nordics. See below for details about the models and data.

RoadSurf model and data
RoadSurf is an energy balance model that predicts driving conditions for the next 24 h. It estimates (1) the road surface temperature, (2) friction and (3) storages on the road (water, snow, ice, frost in water mm-equivalent), and from these (4) road surface condition and (5) driving conditions (driving difficulty derived from a combination of weather and road surface conditions) (Kangas et al., 2015).
The model calculates the vertical heat fluxes between the atmosphere and the road surface by means of turbulence (including traffic-induced), generalised soil material properties and meteorological input parameters. The meteorological input parameters, taken from HCLIM38 regional model output, include 2-m air temperature, precipitation, relative humidity, wind speed and downwelling shortwave and longwave radiation with a time resolution of 1 h. Hydrological processes, including freezing, melting, evaporation, sublimation, run-off from the surface and accumulation of rain and snow, are parameterized. RoadSurf assumes a flat horizontal paved surface over the whole domain with similar deep-ground properties, without elements shading the road, such as trees. Topography, water bodies or forests are implicitly accounted for through the HCLIM38 input data.
Traffic is simulated as a spatially constant atmospheric turbulence on the ground with a larger value for daytime and a smaller one for night. Traffic reduces and changes the storages, e.g. by compressing snow into ice, whilst unpacked snow is assumed to be partly blown away by the wind. The model assumes neither salting nor road maintenance and, therefore, overestimates snow and ice storages on the road (Toivonen et al., 2019). This is justified as RoadSurf is commonly used to warn about the possibility of hazardous driving conditions, as a call to action for the regional road management (Kangas et al., 2015). Despite these overestimations, however, Toivonen et al. (2019) showed that RoadSurf reproduced observed road weather conditions in Finland (2002Finland ( -2014 well when driven by HCLIM38. Furthermore, RoadSurf has been further developed to predict pedestrian sidewalk conditions (walking difficulty) (Hippi et al., 2020).
This study expands upon Toivonen et al. (2019) by exploring a future scenario of driving as well as pedestrian conditions and additional countries. Our study employed version 6.60b of RoadSurf with a few minor adjustments made in this study. We focused on four output variables: (1) road surface temperature (T RS ) (from which zero-degreecrossing days were derived); (2) road surface condition: dry, damp, wet, snow, frost, icy and partly icy; (3) driving condition: normal, difficult and very difficult; (4) pedestrian condition: normal, slippery (slipperiness may occur), very slippery due to foot-packed snow (VS I), very slippery due to water above ice layer (VS II) and very slippery due to snow above ice layer (VS III).
The boundary data for HCLIM38-ALADIN were taken from either the global ERA-Interim reanalysis dataset (ERAI, 80 km grid resolution) (Dee et al., 2011) or from two CMIP5 global climate models (GCMs) from the Coupled Model Intercomparison Project-Phase 5 (CMIP5) (Taylor et al., 2012): EC-Earth (Hazeleger et al., 2010(Hazeleger et al., , 2011 and GFDL-CM3 Griffies et al., 2011) (Table 1). We chose ECE and GFDL as driving GCMs as they cover the full range of temperature and precipitation projections for northern Europe in the RCP8.5 scenario in CMIP5 (Lind et al., 2022, in preparation). Table 1 RoadSurf simulations presented in this paper. Combinations of one global ERA-Interim reanalysis (ERAI), two different global climate models (ECE, GFDL), two different HCLIM38-Configura-tions (AROME, ALADIN), two road-user modes (car, pedestrian) and four different time frames (historic 1986-2005, present 1998-2018, mid-century 2041-2060, end-century 2081-2100)  The reanalysis-driven simulation was used for model evaluation purposes, whilst the GCM-driven runs were used to estimate the impacts of climate change.
RCP8.5 may be becoming increasingly unlikely due to the overestimation of future coal use (Ritchie & Dowlatabadi, 2017;Schwalm et al., 2020), but it remains valuable especially for mid-century projections, to assess policies or to account for higher-than-expected carbon cycle feedbacks (Schwalm et al., 2020).

Road weather observations
We collected observations of road surface temperature and road surface conditions from representative road weather stations in Finland (25 stations, 2002Finland (25 stations, -2005, Sweden (5 stations, 2015-2018) and Norway (5 stations, 2015-2018) in hourly resolution (see details in Online Resource T1). Three of the Norwegian stations are located in mountainous regions near or above 1000-m elevation, which brings challenges to model evaluation as the terrain differences between the model and observations can be large. Swedish observations were from around the country, with one on a bridge. The Finnish observations were more numerous, because there were more reliable observation data available, and the same list of stations was used in the previous study by Toivonen et al. (2019).
Finnish road weather stations use the Vaisala ROSA road weather package equipped with asphalt-embedded DRS511 sensors. Swedish observation data was provided by the PT-100 resistance thermometer probe (Platina 100) embedded in the road and a DSC111 optical sensor (Vaisala). The Norwegian observations were recorded with weather stations from Scanmatic (sensors not communicated).
For the evaluation, we selected the modelled grid cell closest to the stations' coordinates. Thus, the observations were point measurements, whilst the modelled data were an average over one grid cell.

Methods
Henceforth, RoadSurf runs driven with ECE-HCLIM38-ALADIN and GFDL-HCLIM38-ALADIN will be referred to as ECE-driven and GFDL-driven runs, respectively.

Evaluation against observations
We compared historical RoadSurf simulations to the road weather observations in FNS for the years 2002-2005 to assess RoadSurf's predictive accuracy when driven with different combinations of HCLIM38 configurations (-ALADIN or -AROME) and global climate models (ECE or GFDL) as well as with ERAI reanalysis data.
Probability density functions (PDFs) of the regionally observed road surface temperatures (T RS ) were compared to HCLIM38-ALADIN simulations (ERAI, ECE-historical, GFDL-historical) over FNS. Norwegian and Swedish observations did not overlap with the modelled years, so we compared 2015-2018 with the years 2002-2005 from the model under the assumption that the climate did not considerably change during the gap. Thereafter, we calculated the monthly mean bias (MMB, model minus observations) and mean absolute error (MAE) of the modelled T RS .
Road surface condition observations were only available for Finland and Sweden. We compared the observed time fractions of the occurrence of each road surface condition to HCLIM38-ALADIN runs (ERAI, ECE-historical, GFDLhistorical). Road surface conditions in historical simulations in Norway were only compared to their corresponding ERAI run.
Modelled traffic and pedestrian indices could not be evaluated against observations as no such observation data existed to our best knowledge. However, we compared the historical simulations with the ERAI-HCLIM38-ALADIN runs.
Furthermore, we assessed whether there was an additional value of using the computationally heavier high-resolution HCLIM38-AROME configuration over HCLIM38-ALA-DIN in terms of road weather predictions with RoadSurf (three-fold longer run time with four-fold storage space needed for RoadSurf simulation with HCLIM38-AROME data) by repeating all abovementioned evaluation steps for HCLIM38-AROME and comparing them to HCLIM38-ALADIN and observations.
To analyse the changes in occurrence of road surface conditions, we counted the average seasonal time fraction for each condition (dry, damp, wet, snow, frost, icy, partly icy) and compared mid-century and end-century results to the historical period, as well as ECE-RoadSurf and GFDL-RoadSurf results against each other. For further analysis, we also assessed the spatial distributions of the changes.
We repeated the same analysis for the driving conditions (normal, difficult, very difficult) and pedestrian conditions (normal, slippery, very slippery I, II and III). The driving condition "very difficult" occurs rarely and only for short periods during extreme weather events, such as heavy snowfall. For this reason, the "difficult" and "very difficult" driving categories were grouped into one.
ZDC-days give valuable information as most traffic accidents happen when T RS is close to 0 °C (Andersson & Chapman, 2011a). We defined a ZDC-day as a day on which the T RS (at least once) drops from above + 0.5 °C to below − 0.5 °C (or vice versa) causing freezing of water (or melting of ice and snow) on the road surface. We calculated monthly averages of ZDC-days. Subsequently, we calculated the change in average seasonal ZDC-days over the century compared to the historical period and performed a Student's t-test. We also compared the PDFs of historical and future ZDC-days estimates between GFDL-RoadSurf and ECE-RoadSurf.

Evaluation against observations
We compared T RS , PDFs, MMB and MAE of HCLIM38-ALADIN-RoadSurf and HCLIM38-AROME-RoadSurf driven by the ERAI reanalysis data to observations in FNS to quantify if the higher horizontal resolution of HCLIM38-AROME configuration brought benefits over HCLIM38-ALADIN when used as an input in RoadSurf. Means, temporal variation and seasonal patterns of modelled road surface temperature (T RS ) and road surface conditions agreed well with observations in Finland, Norway and Sweden (FNS).
The PDFs showed that both HCLIM38 configurations reproduced observations almost interchangeably well, with a small exception for spring in Finnish Lapland, where HCLIM38-AROME estimated stronger variability around 0 °C than observed (not shown).
The MAE showed that both HCLIM38 configurations reproduced observed T RS accurately over flat inland regions.
For Norway however, both HCLIM38 configurations became increasingly inaccurate. In the Norwegian mountains and northern Norwegian Fjords, HCLIM38-AROME showed better results. This may be attributed to the fact that HCLIM38-AROME can resolve topography-related features better than HCLIM38-ALADIN. However, considering all the above results, the improvement was not deemed sufficient to warrant the use of computationally more expensive HCLIM38-AROME simulations for the climate change analysis.
HCLIM38-ALADIN-RoadSurf driven by three different boundary data, the ERAI, ECE-historical and GFDLhistorical, reproduced T RS observations satisfyingly over Finland, but produced slightly lower autumn-and wintertime T RS compared to observations in Sweden and Norway (Fig. 1). The inaccuracy in the simulated wintertime T RS over Norway might stem from topography inaccuracies on the one hand, and sensors submerged in snow on the other. However, due to the lack of information which sensors were used (optical or asphalt-embedded, etc.), it is difficult to estimate the root cause for the negative bias.
ECE-and GFDL-RoadSurf both overestimated the occurrence of icy road surfaces whilst grossly underestimating frost (Online Resource F1a). Overestimation of icy surfaces was acceptable for the original near future (0-24 h) warning purposes of the model, but must be kept in mind when interpreting the future results in this study. GFDL-Road-Surf yielded more extreme under-and overestimations than ECE-RoadSurf.
The ECE-and GFDL-RoadSurf driving conditions agreed very well with the ERAI reanalysis (Online Resource F1b). GFDL-RoadSurf produced overall more frequent occurrence of difficult driving conditions, with the largest model differences in Finnish Lapland.
Despite the small discrepancies between the modelled and observed road weather parameters, we concluded that the overall performance of RoadSurf was satisfying for this study.
During autumn (September to November, SON), T RS increased 2.5 °C/3.7 °C on average in the mid-century period and 4.9 °C/6.8 °C in the end-century, compared to historical values ( Fig. 3 and Online Resource F3). A slight north-south gradient in T RS increase was simulated to intensify over the course of the century for autumn as well.
The strongest T RS warming was simulated for meteorological winters (December to February, DJF), with an average increase of 2.7 °C/3.9 °C in the mid-century and a 6.2 °C/7.2 °C increase in the end-century.
GFDL-RoadSurf simulations showed an extreme T RS increase of > 4.5 °C regionally in Finland (especially southeastern Finland) during the mid-century, which intensified and extended into the Norwegian mountain range and its entire piedmont with > 8 °C increase in the end-century period.  The whiskers represent the minimum and maximum simulated temperatures simulated by ECE-HCLIM38-ALADIN-RoadSurf (ECE, dark shades) and GFDL-HCLIM38-ALADIN-RoadSurf (GFDL, light shades) following the RCP8.5 scenario ECE-RoadSurf, on the other hand, simulated an almost uniform T RS increase across the three countries in mid-century winter, except for weaker T RS increase in the southern tip of Sweden. A strong acceleration in T RS warming occurred in end-century with > 7.5 °C regional T RS increase in the northern parts of FNS. This roughly agreed with the estimated warming from the GFDL-RoadSurf simulations.
Springtime T RS values (March to May, MAM) warmed 2.2 °C/2.8 °C in the mid-century and 4.6 °C/5.0 °C in the end-century. Whilst GFDL-RoadSurf showed a spatially uniform T RS increase, ECE-RoadSurf estimated a noticeably weaker T RS warming for southern Norway and Sweden, with partially < 1.5 °C T RS increase (compared to > 2.5 °C in more northern latitudes) in the mid-century and < 4 °C (compared to > 5 °C in more northern latitudes) in the endcentury period during spring.

Road surface conditions
We calculated the change in occurrence of road surface conditions (dry, damp, wet, snow, frost, icy, partly icy) in the mid-century and end-century compared to the historical period.
The simulations for the cold season from September to May showed an overall decline in snowy and icy road surfaces with up to 17.3/23.0 percentage points (PPs) less in the mid-century and up to 39.6/47.0 PPs less in the end-century (Fig. 4). Autumn and spring showed large changes during the mid-century and a decelerated change until the endcentury. The strongest changes were indicated for autumn with a combined average decrease of 9.5/14.7 PPs in snow and ice storages on roads in the mid-century and a decrease of 17.4/22.5 PPs in the end-century. Hence, dry roads were simulated to be 7.7/10.1 PPs more frequent in mid-century autumn and 13.4/15.6 PPs more frequent in end-century autumn, compared to historical ones. Spring showed 3.0/4.4 PPs more frequent dry road surface conditions for mid-century and 7.5/7.0 PPs more frequent for end-century.
The net change in the occurrence of road surface conditions in DJF in Fig. 4 seems to be small compared to the change seen in SON, especially in the mid-century, which might leave the illusion that mid-century winters were simulated to expect as much snow and ice on roads as in the historical period. Further investigation however showed that the small net change was attributed to strongly contrasting, region-dependent developments.
Whilst northern parts of FNS showed a 1-9/9-15 PPs shift towards icy road surface conditions in winter, southern Finland and the southern tip of Sweden showed a 3-11/0-21 PPs increase in dry winter road surface conditions. Snowy road surfaces in the Norwegian mountain range, central Sweden and elevated parts of Lapland increased by up to 6.4/6.4 PPs in the mid-century estimates, but decreased by the same amount for southern Sweden, southern Finland and the Norwegian fjords. In autumn and spring, almost all regions showed drier road surface conditions, possibly due to earlier snow melt, except for the glacial regions of Norway, where the earlier onset of spring caused an increase of 0-5/0-5 PPs in wet road surface conditions as well as slight increases  (MIDC, 2041(MIDC, -2060 and end-century (ENDC, 2081-2100, light bars) compared to the historical period (1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005) in Finland, Norway and Sweden as estimated by ECE-HCLIM38-ALADIN-RoadSurf (ECE, blue bars) and GFDL-HCLIM38-ALADIN-RoadSurf (GFDL, grey bars) under the RCP8.5 scenario. The bar labels show the resulting change by the end of the century compared to the historical period in gained or lost percentage points Table 2 Average historic ratio of road surface conditions (HIST) in percent and change in road surface condition occurrences for the mid-century (MIDC) and end-century (ENDC) in percentage points compared to the historic period in the northern and southern halves of Finland, Norway and Sweden. Results are shown in the format "ECE-RoadSurf estimates/GFDL-RoadSurf estimates" with increases in green and decreases in red

Driving conditions
We calculated the change in driving condition occurrences (normal, difficult, very difficult) from historical to midand end-century. "Normal" driving conditions increased strongly in all parts of FNS. The mid-century cold season (September until May) showed a 12.2/18.5 PP increase of normal driving conditions, which increased to + 28.5/33.2 PPs for the end-century compared to the historical period (Fig. 5a). The strongest increase of easier driving conditions was seen in autumn with + 6.3/9.8 PPs in the mid-century and + 11.2/14.3 PPs by the end-century, followed by spring with + 3.7/5.0 PPs in the mid-century and + 8.6/8.6 PPs in the end-century.
Winter showed a moderate change at first with only + 2.2/3.6 PPs in mid-century, but a strong acceleration of change for end-century with an + 8.7/10.3 PPs increase in normal driving conditions. ECE-RoadSurf simulated the strongest increases in normal driving conditions along the Norwegian coasts and the southern half of Finland ( Fig. 5b and Online Resource F4). GFDL-RoadSurf estimated an overall stronger change with the largest changes in southern Norway, southern Finnish Lapland and the valleys of the Scandinavian mountains.

Pedestrian conditions
We calculated the occurrence of pedestrian conditions normal, slippery and very slippery (VS) I, II and III in the historical, mid-century and end-century periods.
The projected walking conditions showed strong seasonality as well as regional differences. Generally, in the cold season between September and May, the pedestrian condition "slippery" was estimated to decrease by 7.0/8.5 PPs in the mid-century and 18.4/18.4 PPs in the end-century, compared to the historical period. An increase was simulated for "normal" (+ 4.9/5.0 PPs mid-century and + 17.7/11.8 PPs end-century) and "VS II" (+ 3.9/5.3 PPs mid-century and + 6.9/11.8 PPs end-century) (Fig. 6).
Similarly, to the changes in driving conditions, the strongest changes in pedestrian conditions were seen for autumn, during which "normal" walking conditions increased by an average 7.8/11.5 PPs for mid-century and 14.3/17.5 PPs for end-century. The most strongly affected regions were the Norwegian fjords and the piedmont of the Scandinavian mountains (Online Resource F6). Estimates for winter showed a clear shift from "normal" conditions towards "slippery" pedestrian indices with the strongest increase in "VS II" (+ 2.9/4.1 PPs in mid-century and + 6.8/10.6 PPs in end-century).

Zero-degree-crossing days
We calculated the seasonal average number of ZDC-days, i.e. days when T RS decreased from minimum + 0.5 °C to − 0.5 °C (or vice versa). ECE-RoadSurf and GFDL-Road-Surf estimated a mostly steady increase in winter ZDC-days and decrease in autumn and spring (attributed to the later onset and earlier end of cold winter temperatures), however with strong regional differences (Fig. 7).
Regions above the Arctic Circle and around the Norwegian mountain range showed minor changes (max. ± 2 ZDC-days) in all seasons during the mid-century (Fig. 7). Finnish regions south of the Arctic Circle showed 2-3/3-6 additional ZDC-days on average in mid-century winter, with the strongest increase in south and south-west Finland. The same regions showed an estimated decrease of 2-4/4-5 ZDC-days in autumn and a decrease of 1-5/3-7 ZDC-days in spring.
Other considerable changes in mid-century ZDC-days were simulated for the Norwegian fjords, southern Sweden and Swedish regions around the Bothnian sea. Mid-century winters showed 1-4/0-1 fewer ZDC-days to regions around the Kattegat strait and 3-5/4-6 more ZDC-days to regions around the Bothnian sea. Autumns of southern Sweden and the Norwegian fjords showed 3-5/3-5 fewer ZDC-days, whilst spring showed 2-5/4-8 more ZDC-days compared to the historical period.

Discussion and conclusions
In this study, we assessed the climate change impacts on driving and walking conditions in Finland, Norway and Sweden (FNS) based on simulations carried out with FMI's road weather model RoadSurf driven by ECE-HCLIM38-ALADIN and GFDL-HCLIM38-ALADIN (ECE-RoadSurf and GFDL-RoadSurf, respectively). Our results are based on the RCP8.5 scenario, a worst-case climate change scenario that anticipates an unprecedented increase in greenhouse gas emissions. Therefore, our simulations are upper extreme estimates, but can serve as a reference of locality and seasonality of the expected changes, and thus help local decision-makers to plan for mitigation and adaptation measures ahead of time.
We concluded that the model performance was accurate enough to carry out the climate change projections for this study, albeit RoadSurf was shown to overestimate dry and icy road surface conditions. The considerably lower computational expense of the HCLIM38-ALADIN configuration allows a wider set of modelled road weather scenarios and is generally preferable, however at the cost of losing topographical features compared to the HCLIM38-AROME configuration, which should be considered for studying regions with large elevation differences.
Our simulations suggested a strong increase in the average annual road surface temperatures (T RS ), with the strongest regional T RS increases in southern Finland and southern Sweden. Winters showed an increase in water above ice on walkways (except for the southern tip of Sweden). Autumn and spring, however, showed a decrease in slippery pedestrian conditions. Hence, the future pedestrian slip-injury season might be shorter, mostly confined to winter, but with an exacerbated risk for slipping. Zero-degree-crossing days (ZDCsdays) were estimated to decrease in autumn and spring (apart from an increase north of the Arctic Circle), but to increase in winter. An increasing number of ZDC-days is a major hazard for pedestrians, as temperatures near 0 °C pose one of the greatest risks for pedestrian slip accidents (Hippi et al., 2020).
The results also showed a decrease in snowy, frosty and icy road surface conditions and an increase in dry, damp and wet road surface conditions. As a result, there was a strong decrease in difficult driving conditions during the cold season between September and May. This is congruent with studies of countries with comparable climates like Canada, where 43% less slippery-related accidents and less road maintenance costs were estimated due to the easier driving conditions (Andersson & Chapman, 2011b). Intuitively, one could expect less traffic accidents due to slippery roads based on our results (Norrman et al., 2000). However, in different studies, the number of accidents did not change in less slippery road conditions, as they evened out with drivers paying less attention in less dangerous driving conditions (Andersson & Chapman, 2011a;Bernard et al., 2001). The combination of more frequently wet roads and more ZDCoccurrences observed in our study could still be exacerbating the risk for traffic accidents, as slippery conditions near 0 °C might be less apparent to drivers (than e.g. snow-covered roads); thereby, drivers could be at risk of underestimating the immediate slip risk and not adjust their level of attention.
As climate change progresses, it is important to understand and be ready for the changes to come. Our results motivate us to recommend road managers and authorities in Nordic countries-amongst others, Destia, Traficom and Fintraffic in Finland, Trafikverket in Sweden and Statens vegvesen in Norwayto plan actions and secure necessary budgets for managing the shorter, but more intense, slippery roads and walkways in the future successfully.
Acknowledgements This work was supported by the Academy of Finland (grant number: 329241) and the Maj and Tor Nessling foundation. The HCLIM simulations were performed by the NorCP (Nordic Convection Permitting Climate Projections) project group. The HCLIM simulations were partly funded by the European Climate Prediction System project (EU Horizon 2020; Grant no. 776613). We acknowledge the storage resource Bi provided by the Swedish National Infrastructure for Computing (SNIC) at the Swedish National Supercomputing Centre (NSC) at Linköping University. Furthermore, the lead author would like to express her thanks to Prof. Tommi Ekholm for his support in the writing process.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.