Landslides can be a serious natural hazard, existing worldwide and causing high numbers of fatalities and damage to property every year (Froude and Petley, 2018). Identifying areas with frequent occurrences of landslides and designating areas with high landslide probabilities is important to protect human life and economic interest (Colombo et al., 2005; Ludwig et al., 2018). Under the generic term “landslide”, a variety of types can be distinguished based on the process and the material involved (Cruden and Varnes, 1996). Several landslide classifications exist that have been refined over the years (Highland and Bobrowsky, 2008; Hungr et al., 2014). When investigating a landslide, gaining knowledge about the spatial occurrence of landslides can further improve our understanding of the underlying processes causing landslides (Malamud et al., 2004).

The study of landslides reaches from site-specific field investigations to global datasets of landslides and from event-based inspections to long-term monitoring for several years (Alberti et al., 2020; Coe, 2020; Mateos et al., 2020; Svennevig et al., 2020a). Among the different spatial and temporal approaches of landslide studies, landslide inventory mapping is a common method to investigate the spatial occurrence of landslides (Guzzetti et al., 2012; Galli et al., 2008; Hao et al., 2020). Landslide inventory mapping can be performed remotely, covering large areas and with the option to validate the dataset in the field (Zieher et al., 2016). Traditionally, landslide inventories are based on aerial imagery and optical satellite images (Brardinoni et al., 2003; Fiorucci et al., 2011). With the emergence of digital elevation data and hill shading, the quality and quantity of landslide inventories have improved substantially (Morgan et al., 2013; Kakavas and Nikolakopoulos, 2021). New areas can be investigated (e.g., forests) and volumes of displaced mass can be calculated (Cavalli and Marchi, 2008). Landslide inventories often contain information about the landslide location, geometry, date of occurrence and damage caused by the landslide (Rosi et al., 2017; Palma et al., 2020).

National elevation mapping efforts and satellite campaigns are extending the areas that are covered by elevation models (Crosby, 2012; EEA, 2016). Advances in sensor technologies and satellite orbit repeat rates are improving the spatial and temporal resolution of the available data, both for optical images and elevation data (e.g., Shugar et al., 2021). Remote sensing data provide powerful information for landslide mapping, but a combination of different datasets such as digital elevation models (DEMs) and multispectral satellite images are necessary to overcome the limitations of each individual dataset (Lissak et al., 2020). The quality of manually mapped landslide inventories strongly depends on the mapping expert's knowledge about the area of investigation (Van Den Eeckhaut et al., 2005). Evaluating the quality of landslide inventories is not straightforward and most mapping efforts do not implement quality controls in their inventory (Guzzetti et al., 2012; Pellicani and Spilotro, 2014; Hao et al., 2020).

Landslide inventories exist on regional, national, international, and global scales (Kirschbaum et al., 2009; Trigila et al., 2010; Damm and Klose, 2015; Herrera et al., 2017). Within Europe, Denmark does not have a national landslide inventory, nor a legislation framework to incorporate landslides and landslide-related damages into national law (Mateos et al., 2020). Landslides are considered a predominant natural hazard in the Nordic countries (Nadim et al., 2008), and a number of case studies investigated landslides in Denmark (Hutchinson, 2002; Prior, 1977). Pedersen et al. (1989) state that Denmark is not a country with a serious landslide problem. However, a recent paper raised concerns that the geo-hazard posed by landslides in Denmark is underestimated (Svennevig et al., 2020b).

With this paper and dataset, we present the first comprehensive landslide inventory for Denmark.

Denmark consists of the Jutland peninsula and an archipelago of 394 islands encompassing 43 938 km2 in total, with 8750 km of coastline (Fig. 1). The landscape is characterized by a low relief with the highest point 171 m above sea level in central Jutland. A long history of agricultural land use has shaped the landscape. Today, around 61 % of the area is agriculturally used, 13 % are forests, another 13 % are transport routes and built up areas, and the remaining land is covered with open habitats and water bodies (Denmark, 2019).

Today's Danish landscape was shaped by numerous glaciations, dominated almost entirely by the two latest ones, the Saalian, ending ca. 130 kyr BP and the most recent Weichselian, ending ca. 16 kyr BP, which lead into the Holocene (Houmark-Nielsen, 1999, 2011). The current landscape configuration is primarily dominated by the last glacial maximum (LGM) extent reached during the Weichselian at ca. 22 kyr BP, where a glacial advance from the northeast reached mid-Jutland, leaving two distinct surface sedimentation regimes: (1) the ice-free west was dominated by sandy glacio-fluvial outwash plains surrounding older glacial deposits from the Saalian and (2) the ice overridden eastern part of Denmark was dominated by glacial processes depositing tills with a high clay content. The landscape here was mainly shaped during the LGM advance and the numerous re-advances up until ca. 16 kyr BP (Houmark-Nielsen, 1999, 2011). Postglacial isostatic rebound has affected especially the northern part of Denmark, which has been uplifted by up to 13 m relative to the local sea level, exposing raised beaches and marine terraces.

Open waters occur in many places in Denmark (Fig. 1) and the glacial landscape is often eroded along its fringes by coastal processes. Waves induce large swash run-up on the beaches and cause erosion of the glacial landscape, forming coastal cliffs. These relatively steep cliffs are susceptible to landslides if the conditioning geology is present. The landslides in the coastal cliffs are presumably sensitive to a combination of water infiltration and specific runoff patterns over impermeable layers in the substrate and to wave erosion of the cliff toe by swash run-up during high water levels under storm conditions (Schou, 1949). The eroded sediment of the coastal cliffs and specifically the landslides in the cliffs are further transported towards deeper water in a cross-shore direction, or along the shores by wave-driven longshore currents forming accreted forms like barrier islands and spits (Kabuth et al., 2013; Kabuth and Kroon, 2014).

The landslide inventory consists of 3202 unique polygons of mapped landslides. The count of types of movement and the number of coastal and inland landslides are shown in Table 2. Alongside the polygonal shape, every landslide is associated with a unique identifier. The planar area (m2) and perimeter length (m) of every landslide are provided as are the x and y coordinates of the center point. By planar area, the largest slides comprise 327 000 m2. Landslides were mapped to a minimum area of 25 m2. An analysis of the mapped landslides shows that most landslides in Denmark are shallow rotational slides. The database underrepresents processes with indistinguishable morphologies and expressions in the DEM, such as rockfalls and mudflows. However, there are only a few areas in Denmark with the geological preconditions facilitating rockfalls. The vast majority of landslides recorded are located in landscapes covered in glacial till. Although all mapped landslides must have occurred after the last glaciation, as their morphological expression would have been erased by the activity of the ice sheet, there are no data available in the landslide database when individual landslides emerged. Landscapes that were not covered by ice during the LGM are almost entirely absent of landslides today (Fig. 1).

We interpret most of the mapped landslides as single events with process durations that span from an instantaneous event to several decades or even centuries and, thus, some are still active while others are inactive landforms today. Landslides that are clearly not a single event are mapped as separate polygons. The present landslide inventory only represents a snapshot of the landslide activity in Denmark at the time of recording from the 2015 DEM. However, the landslide inventory does not contain any information about current or past activity or inactivity. In some cases, landslide areas overlap each other, making it more difficult to distinguish individual landslide morphologies. Without dating every single landslide, a further distinction is not possible in these cases. The datasets and morphological criteria used for the mapping are not suitable for mapping slides with small volumes or faint morphologies, such as rockfalls and mudslides. Thus, these types are expected to be underrepresented in the database. Land use such as farming and infrastructure development may have led to an underrepresentation of landslides in these areas due to intensive cultivation and site development, especially in the inland areas. Nevertheless, around 85 % of the mapped landslides are in coastal environments, often on a cliff at the edge of agriculturally used land. Farmers usually avoid those steep slopes with their heavy and expensive equipment. In some areas along the coastal cliffs, abandoned quarries show morphological expressions similar to landslides in the DEM. The absence of landslide deposits can be the only distinction between the morphological expression of a coastal quarry and a landslide in the DEM. Occasionally, quarries may have been mistakenly mapped as landslides during the mapping. In some cases, landslides evolved on the steep slopes of a quarry, sliding into the former pit and, in other cases, quarry activity may have overprinted landslides.

Within the area of the subsample plots for quality control, the two experts had initially mapped 1029 landslides and the quality control mapped 1057 landslides, a difference of 2.7 %. However, 899 of those landslides were identical, 130 landslides were only mapped during the initial mapping and 158 only during the quality control (Fig. 3). Provided that the combined landslide mapping effort of the initial investigation and the quality control detected the true number of landslides (1187), the initial effort discovered 87 % and the quality control 89 % of all landslides. Furthermore, 151 (4.7 %) landslides in the entire study area were validated by visiting the landslides in the field or by mentions in other resources such as previous publications or newspaper articles.

Based on the careful observation of the entire study area and the implemented quality control, the landslide inventory can be considered 87 % complete with a confidence level of 90 % and an error of 5 % for the 2015 DEM. However, a few landslides always remain undetected and new landslides will have emerged since the DEM was recorded. According to the landslide inventory protocol from Burns and Madin (2009), we only mapped landslides with a moderate to high confidence. The high confidence level, in combination with the high quality of the input datasets, lets us conclude that all landslides included in the database are actually landslides.

The landslide dataset and a document with metadata are freely available from 10.6084/m9.figshare.16965439.v2 (Svennevig and Luetzenburg, 2021) and can also be viewed through a web map environment (https://data.geus.dk/landskred/, last access: 6 June 2022) where layers such as the hillshade model, soil map, pre-Quaternary geology, etc. can be displayed for context. The landslide dataset is provided in the form of an Environmental Systems Research Institute (ESRI) shapefile including the following attributes: landslide ID, area, perimeter length, center point coordinates, coastal or inland, movement type, field validation, quality control confirmation, original mapper and modifying mapper. The definitions of each attribute are provided in an additional metadata text document. The DEM is available for download in 10 km tiles (https://datafordeler.dk/dataoversigt/?emne=landkort%20og%20geografi, last access: 6 June 2022; Geodatastyrelsen, 2015b).

The motivation for creating and freely providing this landslide inventory is twofold:

The first national landslide inventory for Denmark is an important step towards a more comprehensive hazard and risk framework for Denmark. Making the inventory available enables local, regional and national stakeholders to implement landslides into their risk reduction strategies. Furthermore, a legislative framework implementing landslide risk and damage may build upon this dataset. With the expected increase in global landslide activities due to climate change, a landslide risk reduction strategy is now more important than ever before (Gariano and Guzzetti, 2016). In Denmark, a combination of increases in frequency and magnitude of heavy precipitation events, ground water level rises, storm surges and a general increase in relative sea level make higher landslide activity in the future very likely. Therefore, it is crucial to better understand the underlying processes causing landslides and develop effective risk reduction strategies to protect human lives and property.