Of the 846 collected events, 50 % (421 events) are floods, including 1 % (eight events) hail flooding events (Fig. 1a). The second most frequent process was landslides (23 %; 192 events), followed by rockfalls (11 %; 96) and debris flows (8 %; 68). The remaining were avalanches (2 %; 15) and other processes (6 %; 54) including snowdrifts (4.5 %; 40) and falling trees (1.5 %; 14). Snowdrifts mainly resulted from a unique event in February 2015.

The quality of the presented database is affected by several factors. The online press articles, the main source of this database, did not report all natural hazard events affecting the Swiss transportation network. This is particularly the case for events of small intensity. Reporting of such events in articles depends on the number of casualties, the severity of the injuries, the resources available for creation of the article, the preventive or educational interest, and the presence of images. Article occurrence was theoretically higher in summer, when the news activity is lower because of quieter political activity. In some cases, the sensitivity increased, for example, after two tourists were killed on Gotthard highway in 2006 when a portion of the Eiger summit collapsed. This made journalists prone to focusing on slope mass movements (RTS, 2006a, b; Liniger and Bieri, 2006; Oppikofer et al., 2008). Conversely, when many events occur simultaneously during intense storms, only the most significant disasters are reported in the press. Event reporting likely depends on the perception linked to the region of occurrence and the type of transportation network. For instance, a 0.5 m3 rockfall on a railway track on the Plateau has more media impact than one occurring on an alpine road, where such events are more common and the consequences on the traffic are lower.

The events collected from 2012 to 2016 ranged from 60 to 269 events per year (Fig. 6f and Table S20). This may be biased because Google Alerts were only used after May 2014. Data collection was less systematic for 2012 and 2013, with 60 and 99 events, respectively. With Google Alerts, the number increased to 245 and 269 for 2015 and 2016, respectively. With 173 events, 2014 was a transitional year, with Google Alerts used for approximately half of the year. An advantage of Google Alerts is the variety of the online sources from almost all the available online newspapers, which is better than a single source (Badoux et al., 2014). Google Alerts allows us to improve the event collection for floods. Moreover, the total number of events increased yearly, even after the use of Google Alerts, due to the increase in flood disruptions (Fig. 6f). This shows that the use of Google Alerts is not fully responsible for the yearly increase in the number of events. These numbers depend strongly on the weather conditions that vary yearly. This demonstrates that the event distribution is strongly dependent on a limited number of meteorological events such as long rainfalls or severe storms.

Statistical predictions regarding a small sample of events are intrinsically imprecise (Davies, 2013). The annual cost of damages from natural hazards in Switzerland (Hilker et al., 2009) from 1972 to 2007 shows great damage disparities over the years because extreme rainfall events or successive storms greatly increase the number of events in 1 year.

From a geographic point of view, the collected data should be considered a snapshot of a short time period capturing the background of small-intensity events, representing 96 % of the total events and 76 % of the total direct damage costs.

Notably, a number of natural hazard events induce expensive maintenance operations without affecting the traffic, for example, by damaging protective infrastructure. Those events are not considered in this study because they do not generate traffic perturbation but they should be considered in risk management.

The terminology of natural hazard events on roads and railways is partially inappropriate because, although the origin of the direct event is typically natural (e.g., rainfall), the indirect origin is often anthropic. The construction of a transportation network, its use, and maintenance induce severe changes or actions that potentially affect slope stability, according to the Terzaghi (1950) classification of the mechanism of landslides (Jaboyedoff et al., 2016a). These causes of destabilizations, such as slope re-profiling, groundwater flow perturbation, surface water overland flow modifications, land degradation, inappropriate artificial structures, traffic vibration, and ageing of infrastructure affect landslide occurrence (Larsen and Parks, 1997; Jaboyedoff et al., 2016b). Furthermore, new infrastructure around tracks often induces an under-sizing of the existing drainage systems, which can induce the concentration of the surface or groundwater flow and destabilize slopes. People are thereby very often responsible for aggravation of the hazard consequences for built-up areas without having sufficient knowledge of the natural hazards and associated risk. Laimer (2017b) indicated that, along Austrian railways, 72 % of events are human-induced.

Minor and medium-sized natural hazard events are not well documented because their direct consequences are often rapidly fixed, i.e. when the road can be re-opened within a few hours of the event or is only partially closed.

The slope angle values are lower than common values for natural hazard slopes because they are not the slope angles at the event origin but at the end of the propagation, as tracks are generally located much lower than the sources of propagation.

Several factors must be considered in the slope distribution. One explanation for the lower number of events on north-facing slopes is that there are fewer tracks due to the lower number of buildings on these slopes. Furthermore, north-oriented slopes receive less solar heat than south-oriented slopes and thus have fewer freeze–thaw cycles. This can partially explain the high number of rockfalls on west-, south-, and east-oriented slopes.

The monthly distribution indicates that floods mostly depend on two meteorological conditions: thunderstorms and long-lasting rainfalls, which mainly occur in spring and summer, particularly in combination with snowmelt in summer. The near absence of floods in winter is the result of the Swiss winter climate, with a lack of long or brief but intense precipitations and precipitation in mountains falling as snow. However, exceptions are possible, such as floods caused by winter storms in January 2018 (RTS, 2018). Debris flows mostly occurred in summer as the result of powerful and stationary thunderstorms. Landslides mainly occurred in spring due to long-lasting rainfalls with melting snow, generating water, saturated soils, and low evaporation. Snowmelt is the second trigger of landslides after intense rainfalls on Austrian railway tracks for 2005–2015 (Laimer, 2017b). Laimer (2017b) has shown that intense precipitation is a trigger for 78 % of landslides on railway tracks in Austria from 2005 to 2015. Freeze–thaw cycles during the winter are also a strong trigger of rockfalls.

Rockfalls do not follow the trend of occurring mainly in spring and summer. They occur in every season, mainly in autumn, winter, and spring due to numerous freeze–thaw cycles during these seasons which weaken the cohesion of rocks. Unsurprisingly, avalanches occurred mostly in winter. They also occurred in autumn as the result of fresh avalanches on soils that are not yet covered with snow and non-effective winter track closures of roads in the Alps. The absence of avalanches in the spring is likely due to the presence of road winter closures.

Floods mostly occurred in the afternoon, probably after strong thunderstorms. Debris flows mostly occurred in the evening, probably after strong thunderstorms in the late afternoon or in the early evening. Landslide event triggers were not time dependent as the other event processes were. Rockfalls appear to be triggered during thawing, which occurs mostly in the morning. Snowdrifts from the other category began in the afternoon, after a few hours of strong wind. This is why the other category events are concentrated in the afternoon. Notably, the time of the event does not always match the actual event time, especially for events occurring during the night or on tracks with little traffic such as country roads.

The high proportion of landslides on train tracks can be explained by the presence of soil embankments or unsuitable filled material along railway tracks and due to their inclination limitations. In addition, despite more protections than average, highways are proportionally more vulnerable than other roads because of the alignment with many imposing cuts and fills. Similar to motorways, railway tracks require a balanced gradient ratio and thus must run along valley sides over far distances. This requires long and steep cut slopes (Laimer, 2017b).

Regional railway tracks may have suffered from a lack of maintenance on track embankments during recent decades, which caused landslides and rockfalls on old infrastructures that were built long before the basics of soil mechanics were understood (Terzaghi, 1925; Michoud et al., 2011; Laimer, 2017a, b).

The higher number of direct impacts (24) than indirect (11) impacts on roads shows that drivers can generally stop their vehicles before being affected by a fallen event, unlike trains, which cannot be stopped within a short distance and reach the fallen mass (seven indirect impacts and one direct impact). In addition, there is a much higher probability that a vehicle on a road would be directly impacted by an event than a train on a track because the road traffic is excessively denser than the railway traffic.

Deviation lengths for railways are difficult to evaluate. In the case of replacement buses, the distance of deviation is calculated using the distance of the replacement buses on the road. For 72 events on railways (75 % of all events on train tracks), there were no possibilities of deviations using other train tracks. In cases of no replacement service, the deviation length for the railway was the distance of train track between the two stations on both sides of the track closure. The average distance of deviation for this configuration was 65 km.

An example of an event from our database can be summarized as follows: a flood event occurred in June during the afternoon in the Plateau region on a small south-oriented slope with a minor road. It generated a road closure of several hours with a deviation distance of less than 1 km and caused no injuries or deaths. The possibility of road deviation is large. On the day of the event, the sun shined for half of the day, 10 mm of rain fell (20 mm during the previous 5 days and 35 mm during the last 10 days), and the average temperature during the event was 20 ∘C. There were approximately 1000 lightings around the event location on the event day and the wind speed was 7 km h-1 in a northeast direction.

Direct damage costs include all costs directly related to the rehabilitation of the track to ensure traffic service. All repair costs of the tracks are included. The estimated direct costs did not consider indirect costs such as vehicle repairs (the repair of a train costs a lot), implementation of deviations, replacement buses in case of railway closure, costs generated due to the traffic restriction for road and railway users or mitigation work, and protective measures.

The estimation of direct damage costs depends on many factors that are difficult to estimate. The hour has an impact on the cost: repair work during the night or the weekend costs more than that during office hours. The event location also affects the costs; for example, costs in an alpine valley far from construction companies are higher than those in an agglomeration where construction machines and landfill for the excavated material are close by. The date also impacts the costs: an event occurring during a period when weather conditions are difficult will last longer. The emergency of the situation also influences the direct costs, as damage on a secondary road or a highway will be treated with a different emergency level. There were also influences from traffic, the presence of damaged retaining walls and protective measures, the slope angle, the financial situation of the administration responsible for the repair work, and the necessity of work on the slope or cliff above the track. Work on railways costs more than that on roads because the access is often more difficult and because contact lines and rail repairs can be more expensive.

An estimation of the direct costs of the small events is more credible than the costs of events of higher damages because the main work is to clear the road of fallen materials. Cost estimation for the middle and large events is more complicated because the repairs require large construction sites, which have their own characteristics that cannot be generalized.

The estimated costs must be considered as an order of magnitude of the direct costs generated by natural hazard events on the Swiss transportation network. These costs could be up to 10 times higher than the given cost estimation. However, the results are more refined than those of the previous study of Voumard et al. (2016), in which costs of events below EUR 8500 were not considered.

Compared to the annual direct damage cost estimation of EUR 3.4 million for natural hazards on the Swiss transportation network, annual damages caused by natural disasters in Switzerland for 1972–2011 are estimated at EUR 290 million per year (OFEV, 2013). Switzerland allocates EUR 2.5 billion each year for protection against natural hazards, which corresponds to 0.6 % of its GDP. A total of 21 % (EUR 0.5 billion) of this allocated amount concerns intervention and repair (OFEV/OFS, 2007, 2011).

There are several methods to quantify the costs of track closures (Nicholson and Du, 1997; Erath et al., 2009). However, they are unsatisfactory because the quantification of costs, especially the indirect costs, is difficult and the resilience must be carefully considered, as people often find solutions to bypass the track closure (deferred travel, meeting realized with digital technologies, alternative sources of supply, etc.).

The closure costs due to natural hazards, such as traffic congestion costs, are not compensated for in Switzerland. However, models must include the potential loss of income in taxes if the economy of the region is slowed. In addition, there are several ways to replace a transportation route or means. For example, trains can be replaced by buses between two stations. Using other train routes can be very complicated and long. Road deviation is usually much easier; however, in some valleys in the Alps, the deviation lengths can reach hundreds of kilometres and there may be no possibility of deviation. Notably, the increase in the travel duration in the case of railway closures is more relevant for passengers than the distance of deviation.

The spatial distribution (Fig. 2) indicates a high density of events in populated areas, principally on the Plateau. This concentration of events around populated areas can be explained by various factors. First, when a meteorological event occurs in a densely populated area, it may primarily affect tracks because the transportation networks are dense in those areas. Conversely, a meteorological event that covers a similar surface but occurs in a sparsely populated area, for example, in an Alpine lateral valley, will affect few tracks. Second, the number of people impacted, the associated economic consequences, the population sensitivity, the number of journalists available, and the number of reporter-readers impact the media coverage of the natural hazard events. This leads to better media coverage of events in densely populated areas.

Davies (2013) notes the importance of the event in the context of the affected persons. A minor landslide that affects a person is unworthy of notice to the vast majority of the population but is considered momentarily catastrophic for the person, as they must reconsider their travel, find an alternative route, or cancel their appointment.

Information acquisition is challenging in the development of such a database because it depends on several people working in the field, such as road menders, railway maintenance workers, and forestry workers, who may have little time or interest in filling in the relevant attributes of the database. Hence, improvements to the database quality are possible using new tools such as offline collaborative WebGIS (Balram and Dragicevic, 2006; Pirotti et al., 2011; Aye et al., 2016; Olyazadeh et al., 2017), which can facilitate event data collection directly in the field using smartphones.

Furthermore, data acquisition and data analysis should distinguish the specific types of transportation networks. For instance, the sensibility to a natural hazard event on a railway track, where a 1 dm3 rock can derail a train, is different from the sensitivity of an alpine road to the same volume of rock. Similarly, a landslide generating a track gauge change of 1 cm can lead to a train derailment whereas a landslide-inducing track displacement of few tens of centimetres will probably not seriously affect the traffic of a mountain road. The liabilities in case of accidents on a railway track or road also differ. The railway manager and operator are responsible for the passengers' safety whereas the road manager allocates part of the responsibility to the driver. Therefore, compared to the road network, the railway network has a much higher sensitivity. The collection of the natural hazard events affecting roads and railways can be improved using different communication channels including social media such as the Facebook page of the Colorado Department of Transport (CDT) in the United States. This diffusion channel allows for the CDT to highlight natural hazard events that affect roads in Colorado, informing drivers of their travel impacts.