Potential future exposure of European land transport infrastructure to rainfall-induced landslides throughout the 21st century

In the face of climate change, the assessment of land transport infrastructure exposure towards adverse climate events is of major importance for Europe’s economic prosperity and social wellbeing. In this study, a climate index estimating rainfall patterns which trigger landslides in central Europe is analysed until the end of this century and compared to present-day conditions. The analysis of the potential future development of landslide risk is based on an ensemble of dynamically downscaled climate projections which are driven by the SRES A1B socio-economic scenario. Resulting regional-scale climate change projections across central Europe are concatenated with Europe’s road and railway network. Results indicate overall increases of landslide occurrence. While flat terrain at low altitudes exhibits an increase of about 1 more potentially landslide-inducing rainfall period per year until the end of this century, higher elevated regions are more affected and show increases of up to 14 additional periods. This general spatial distribution emerges in the near future (2021–2050) but becomes more pronounced in the remote future (2071–2100). Since largest increases are to be found in Alsace, potential impacts of an increasing amount of landslides are discussed using the example of a case study covering the Black Forest mountain range in Baden-Württemberg by further enriching the climate information with additional geodata. The findings derived are suitable to support political decision makers and European authorities in transport, freight and logistics by offering detailed information on which parts of Europe’s ground transport network are at particularly high risk concerning landslide activity.


Introduction
This study is devoted to the assessment of climate change driven landslide hazards to European transport infrastructure (rails and roads) in the near  as well as the remote (2071-2100) future. Results are based on the so-called A1B socioeconomic scenario (IPCC, 2000) and shall provide European Transport Authorities with auxiliary information for setting up 5 cost effective and spatially targeted protection measures in order to safeguarde Europe's transport system in the future. As for the outstanding importance of land transport modes for Europe's social and economic prosperity, the free and uninterrupted movement of persons and freight is of central magnitude. For instance, the accessibility of healthcare facilities, the supply of daily goods as well as a broad range of services to communities rely on the continuous availability of roads and railway connections. Climate change and alterations in extreme weather events, which are affecting ecosystems and man-made infrastructure 10 have been investigated, published and discussed since some decades by now (e.g. IPCC, 1990IPCC, , 1995IPCC, , 2001IPCC, , 2007IPCC, , 2012IPCC, , 2014b. Even though observed changes in extreme events in terms of frequency, intensity and duration may not be directly associated with global warming, trends concerning landslide events are visible (Gariano and Guzzetti, 2016;McBean, 2011;Crozier, 2010). Thus, the assessment of land transport infrastructure exposure towards adverse climate events and related natural hazards is of significant importance for Europe's economy, for its intermodal transport, its freight and logistics networks as well 15 as for settlements in hazard-prone regions (Koetse and Rietveld, 2009;Doll et al., 2014). Therefore, information on current climate and its variability as well as on potential future climate changes is of prime importance for proactive planning and the development of adaptation strategies concerning operation, maintenance, reinforcement and construction works and for civil protection.
Extensive soil sealing across Europe (Nestroy, 2006), climate change (European Environment Agency, 2014) and extreme 20 weather impacts (Schlögl and Laaha, 2017) challenge the resilience of transport systems, which have thus grown into a matter of major concern -not only because of physical damages to assets (Kellermann et al., 2015), but also due to potential overall societal losses caused by network failures and interruptions, which often exceed infrastructure damages by far (Postance et al., 2017;Pfurtscheller and Vetter, 2015;Bíl et al., 2015;Pfurtscheller, 2014;Pfurtscheller and Thieken, 2013;Meyer et al., 2013).
The derivation of climate change induced future hazards is essentially based on two key components: (i) on sets (so-called 25 ensembles) of Global Climate Model (GCMs) runs, which are driven by potential future pathways of mankind and cascaded down to regional-scales via downscaling techniques (e.g. von Storch et al., 1993;Matulla et al., 2003) and (ii) on relationships (so-called Climate Indices -CIs) between regional-scale climate phenomena (e.g. long term rain exceeding certain thresholds) and damage events. Ensembles of climate change projections depicting corridors of future climate evolutions and CIs can be arranged in so-called Cause-Effect Tensors, which have already been successfully applied to access potential future damage 30 events to European transport infrastructure (Matulla et al., 2017).
Studies about the effects of climate change on landslide activity have gained center stage in recent years. Crozier (2010) was one of the first to systematically examine the underlying mechanisms linking climate change impacts and slope stability. It was pointed out in this work that while there is a strong theoretical basis for increased landslide activity as a result of a changing climate, a certain extent of uncertainty remains due to inherent incompleteness and inhomogeneities of historic data on climate 2 Nat. Hazards Earth Syst. Sci. Discuss., https://doi.org /10.5194/nhess-2017-393 Manuscript under review for journal Nat. Hazards Earth Syst. Sci. Discussion started: 2 November 2017 c Author(s) 2017. CC BY 4.0 License. and landslide recordings, the nature of scenario-based projections and the lack of downscaled data at an appropriate spatial resolution. However, changes in annual and seasonal precipitation patterns (in terms of both, precipitation totals and intensity) have been detected as key determinants affecting landslides Gariano and Guzzetti (2016); Sidle and Ochiai (2013).
Since the difficulties of establishing a universal relationship between climate change and landslides across the entire of 5 Europe has been pointed out by Dikau and Schrott (1999), citing the complexity of the problem as the main obstacle, several authors have undertaken efforts to establish such relationships for different parts of the continent (Gariano and Guzzetti, 2016;Crozier, 2010). Rainfall periods exceeding certain thresholds have been found to serve as a proper proxy for landslide occurrences and been applied in Central Europe (Peruccacci et al., 2017;Matulla et al., 2017;Gariano et al., 2017;Brunetti et al., 2010;Guzzetti et al., 2008Guzzetti et al., , 2007Dixon and Brook, 2007). Based on these findings, we employ a CI that inherits the 10 intensity and the totals of severe rainfall events, which have been shown to act as a primary trigger of rapid-moving landslides in Central Europe (Gariano and Guzzetti, 2016).

Climate
Since the goal of this study is to investigate potential changes in landslide events jeopardizing Europe's transport infrastructure, 15 climate data used refer to two future periods relative to past conditions: the near future (2021-2050) and the remote future (2071-2100). Thereby an ensemble consisting of 17 dynamically downscaled and bias-corrected climate change projections (Imbery et al., 2013), which are driven by the so-called A1B SRES socio-economic scenario (IPCC, 2000), is applied to calculate future frequencies of landslide events. The A1B scenario describes a future world characterized by a dynamical economic development, decreasing social and income inequalities a rapid dissemination of new and efficient technologies as 20 well as a balanced use of energy sources. Daily precipitation totals, given on a 5 km grid across large parts of Central Europe (see Fig. 1) are analyzed in order to identify changes in the occurrence of rain-periods stretching over three days, exceeding 37.3 mm and exhibiting at least one total larger than 25.6 mm.

Infrastructure
The graph of the road infrastructure network is based on a data extract of OpenStreetMap (OSM) obtained in June 2017.

25
The high-level road network used in this study is derived from the OSM data set by applying a filter selecting only the following highway tags: motorway, motorway_link, trunk, trunk_link, primary and primary_link. The thusly selected network contains highways as well as major roads (OSM, 2017). The railway network is represented by the Natural Earth Railroads vector data set, version 3.0.0 (Natural Earth, 2017). In order to obtain results at an adequate resolution, all connections exceeding 500 m have been split into multiple segments every 500 m.

Topography
The EU-DEM v1.1 Digital Elevation Model (DEM) is used in this study to analyze topographic properties. This DEM, which is a hybrid product based on SRTM (Shuttle Radar Topography Mission) and ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) DEM data, fused by a weighted averaging approach, is provided by the Copernicus programme 5 (Copernicus, 2017). The slope and Terrain Ruggedness Index (TRI) are derived with gdaldem using Horn's formula (Horn, 1981).

Methods
Ever since the impact of mankind on the climate system has been proven (e.g. IPCC, 2000) it has become increasingly important to derive estimates of future climate states. Global Climate Models (GCMs) are presently the state-of-the-art tools to investigate 10 future climate developments, mimicking global scale consequences for the climate system of the Earth. GCM results are valid on global and continental scales and have to be 'downscaled ' (e.g. von Storch et al., 1993) to regional scales for the assessment of future hazard occurrences. This approach has been shown to yield valuable results for the evaluation of rainfall-induced landslides at regional scales (Gariano et al., 2017;Matulla et al., 2017). The assessment of associated future climate threats jeopardizing Central European transport assets relies on ensembles of daily-based, regional-scale climate change projections.

15
Such ensembles of climate change projections driven by potential pathways of mankind can be generally derived via two techniques: dynamical and statistical-empirical downscaling (Matulla et al., 2002;von Storch et al., 1993). Here we make use of an ensemble consisting of 17 projections forced by the A1B socio-economic scenario (see above).
Potentially landslide triggering events are established by using a proxy indicating precipitation periods extending over at least three days, which are generating overall totals of more than 37.3 mm, and comprising at least one day with a total exceeding 20 25.6 mm (Matulla et al., 2017;Guzzetti et al., 2008). Changes in hazard occurrences (predefined via this CI) over time can be analyzed by comparisons of past and potential future probability density functions.
In order to obtain infrastructure exposure of the road and railway assets in Central Europe at the network level, values from above described gridded data sets have been extracted at respective grid points -considering the skillful scale (Jóhannesson et al., 1995;von Storch et al., 1993) -and assigned to underlying road segments.

4 Results and Discussion
In terms of the time horizon under consideration, results of this study refer to two different time periods throughout the twentyfirst century. The first period (near future) covers the years 2021 to 2050, while the second period (remote future) ranges from 2071 to 2100. In this context it has to be noted that projected values of KLIWAS17 are relative to current climate conditions. This entails that all results have to be interpreted as variations that are averaged over the whole future period with respect to 30 present day conditions as a baseline reference. Results are visualized as exposure maps for both the high-level road network (Fig. 1) as well as the railway network (Fig. 2).
In order to provide information about future probability density functions of the selected CI (and not only about the best estimate), their quartiles (i.e. 25 th percentile, median and 75 th percentile) are reported for each of the two time periods under consideration, resulting in six different facets. While the first row of each figure refers to the near future, the second row 5 displays projection results for the remote future. The three columns represent the quartiles in increasing order respectively. Generally speaking, results show that the most risk-prone areas are located in regions that are characterized by structured topography, e.g. in uplands or in the Alpine forelands. Concerning near future changes, at least a slight increase of rainfallinduced landslides has to be expected all over the entire region. As expected, only the near future 25 th percentile does not show any changes in potential landslide occurrences in certain lowland areas, which are mainly located in north-eastern and 10 middle Germany. This is consistent with the physics of maritime influenced regions since large water bodies tend to damp rapid changes in surrounding areas. In contrast, there are several regions that are expected to face up to seven additional landslide-inducing periods, namely the Vosges, the Black Forest ("Schwarzwald"), the Swabian Jura ("Schäbische Alb"), the  Bergisches Land, the Jura Mountains, the foothills of the Northern Limestone Alps, the Alpine foreland in Austria and Bavaria as well as the Bohemian Forest ("Šumava" or "Böhmerwald"). Our findings show that changes in occurrence frequencies of landslide-triggering extreme climate events slightly increase over time, as is clearly visible when comparing the second row of Figs. 1 and 2 to the first row of each Figure respectively. Basically, the same patterns apply, but the magnitude of the changes 5 is increased in the remote future period. This is largely consistent with findings from the IPCC's fifth assessment report (IPCC, 2014a). It has to be noted, though, that the overall occurrence of landslide-inducing rainfall events appears to increase only slightly in pace throughout the twenty-first century. Nevertheless, the aforementioned areas along the north side of the Alps are likely to experience substantially increased landslide activity in the remote future compared to current climate conditions,  With respect to infrastructure exposure towards potential landslide susceptibility the following regions can be discriminated: 1. regions with a substantial increase in rainfall-induced lanslide activity are the Jura Mountains, the Vosges, the Black 3. the rest of Central Europe, where changes in occurrence frequencies of rainfall-induced landslides have to be expected only to a minor extent. 10 As far as the backbones of the European road and railway infrastructure -the so-called Trans-European Transport Networks (TEN-T) -are concerned, most TEN-T core-network corridors are likely to be affected by increased landslide activity. In Meteorological impacts on geomorphological events and driving landscape-change processes over short time scales have been found to have serious consequences, particularly in climatically sensitive regions such as the European Alps Keiler et al. (2010). Yet it has to be noted that while this analysis refers to potentially hazardous rainfall events that may trigger landslides, other environmental information have not been taken into account so far. Therefore, the consideration of reduced 20 static information commonly used for landslide susceptibility evaluations (Günther et al., 2014(Günther et al., , 2013Hervás et al., 2007) is illustrated and discussed at the example of a selected, risk-prone area, henceforth called "target region". The selected region is centered around the lowlands of the Alsace and the Black Forest mountain range, covering parts of France, Germany, Switzerland, Luxembourg and Austria (Fig. 3). This area has been selected for two reasons. First, it is one of the regions showing the largest increase in the selected CI. Second, consequences of interruptions are quite severe in this area, as the recent mass movements in the Rhine Valley between Baden-Baden and Rastatt have shown (Ackeret, 2017). As far as topography is concerned, the slope angle and the Terrain Ruggedness Index (TRI) are considered. Slope angles are known to be a key parameter in estimating susceptibility to developing earth flows (Donnarumma et al., 2013). The TRI is defined as the mean difference between a central pixel and its surrounding cells (Wilson et al., 2007) and can be used to quantify landscape heterogeneities, which could exert influnce on the localization of the triggering area of shallow landslides (Persichillo et al., 2016). Results show that the target region is not only prone to an increased amount of rainfall that induces 10 landslides, but also susceptible to mass wasting due to its topographic properties (Fig. 4). Both road and rail infrastructure in this area are frequently located in rugged terrain, in valleys, on hillsides or at foothills of mountains. This is particularly the case for the Rhine-Alpine Core-Network Corridor, parts of which are located along the steep western slopes of the Black Forest.
In order to account for the nature and the properties of the ground, which are important aspects affecting landslide sus-15 ceptibility too, rock and soil type have been mapped for the target region (Fig. 5). Lithological types are obtained from the International Geological Map of Europe and Adjacent Areas (Asch, 2005), and the dominant Soil Typology Unit is mapped according to the World Reference Base for Soil Resources as available in the Euopean Soil Database (Panagos et al., 2012;Panagos, 2006). Sedimentary rock types (sandstone, mudstone and limestone) are prevalent in the area. Igneous (granite and basalt) and metamorphic (gneiss and schists) rocks can be found in the Vosges and Schwarzwald mountain ranges as well as the