Introduction

Landslides are amongst the most destructive and frequently occurring natural disasters in mountainous terrains (Saha et al., 2002, 2022). They occur due to the movement of rocks, earth, or debris down an unstable slope in the form of slope failure, rockfall, mudflow, and debris flow (Chawla et al., 2018; Cruden, 1991). Their occurrence is controlled by geological and geomorphological processes and mainly depends on the local terrain conditions (El Jazouli et al., 2019). Landslides can occur on unstable slopes due to factors like deforestation, changing hydrological conditions, stream erosion, volcanic eruption, and anthropogenic activity (Dahal et al., 2006; Dai et al., 2002; Glade, 2002; Gorsevski et al., 2006; Keefer, 1984; Pandit et al., 2021; Raghuvanshi et al., 2014; Schuster & Highland, 2007; Tropeano & Turconi, 2002; van Beek & van Asch, 2004). High precipitation and seismicity, in particular, are the two main triggering factors associated with landslides (Chawla et al., 2017; Lin et al., 2017; Niu et al., 2014; Xu et al., 2014).

Landslides inflict extensive damage globally, affecting the natural landscape, causing economic losses, and resulting in human tragedies (Batar et al., 2017; Fell et al., 2008; Geertsema & Pojar, 2007; Kanungo et al., 2006; Meusburger & Alewell, 2008; National Disaster Management Authority, 2019; Promper et al., 2014; Schuster & Fleming, 1986; Schuster & Highland, 2007; Tan et al., 2020). Despite their impact, the development of a reliable, cost-effective, and efficient landslide early warning system is in its early stages (Naidu et al., 2018). Hence, in the case of landslides, zonation maps incorporating technologies like Remote sensing, Geographical Information Systems (GIS), and modern techniques like Machine Learning (ML) and Deep Learning are increasingly employed to prioritize resources in the event of a disaster striking (Chang et al., 2020; Huang et al., 2017, 2020; Park & Kim, 2019; Peethambaran et al., 2019; Pourghasemi et al., 2018).

Several GIS-based methods, such as heuristic, quantitative, physical-based, and process-based approaches are employed in landslide zonation mapping studies (Conforti et al., 2014; Kim et al., 2018; Pradhan & Lee, 2010; Soria et al., 2011; Zare et al., 2013). Digital Elevation Models (DEMs) play a crucial role in such studies by providing a digital representation of the physical terrain. Previous studies have explored the impact of DEMs on geomorphological and landslide-related analyses with studies exploring publicly available DEMs like Advanced Spaceborne Thermal Emission and Reflection—Global DEM (ASTER-GDEM), Shuttle Radar Topography Mission (SRTM-DEM), Advanced Land Observing Satellite (ALOS) AW3D30-DEM (ALOS-3D), ALOS Phased Array type L-band Synthetic Aperture Radar (ALOS PALSAR) DEM, and Carto-DEM Ver. 3R1 (Huang et al., 2017; Kanungo et al., 2006; Peethambaran et al., 2019; Pradhan, 2013; Roy & Saha, 2019).

It is reported that DEMs serve as the fundamental framework for modeling the causative factors that contribute to landslides (Batar & Watanabe, 2021; Naseer et al., 2021). The characteristics of these causative factors, in turn, rely on the spatial resolution of the DEMs used, as reported by researchers (Chow & Hodgson, 2009; Rabby et al., 2020). It has been observed that the different processing techniques can affect DEM quality (Mahalingam & Olsen, 2016). The impact of DEM resampling on terrain feature estimation has been reported in various studies (Chow & Hodgson, 2009; Deng et al., 2007). Studies have also reported the impact of some publicly available DEMs on landslide-specific modeling applications (Sarma et al., 2020). Moreover, the impact of DEM quality on the final accuracy of landslide zonation analyses has also been widely reported (Kamiński, 2020; Rabby et al., 2020). However, the impact of the Carto-DEM Ver. 3R1 and the ALOS PALSAR DEM on geomorphological modeling and terrain feature estimation have not been adequately reported in the literature. While several studies have focussed on the final accuracy of the landslide zonation analysis, limited studies have discussed the impact of different DEMs on terrain feature estimation at landslide locations.

Thus, an attempt has been made to analyze the effects of Carto-DEM Ver. 3R1 (approx. 30 m) and ALOS PALSAR DEM (12.5 m) on the spatial characterization of landslide locations in the present study. The selection of these DEMs is motivated by the distinctions in their acquisition technologies, quality, and spatial resolutions. To achieve the objective of the study, the DEMs are used to prepare two geodatabases consisting of digital maps of six causative factors each, including Elevation, Slope, Aspect, Curvature, Terrain Ruggedness Index (TRI), and Distance to Drainage. Subsequently, terrain feature estimation is performed at the landslide locations by conducting a spatial frequency distribution analysis of the selected causative factors. The study highlights differences in the characterization of landslide locations using the selected DEMs, which are based on different processing techniques. The results of the study may guide future researchers in making informed decisions when performing, analyzing, and correlating the effects of DEMs on the estimation of terrain features at landslide locations. Furthermore, it may contribute to the realization of the Sustainable Development Goal (SDG) 15 and Sendai Disaster Risk Reduction (DRR) global targets (a), (b) and (g), and help future researchers in evaluating potential causative factors when undertaking studies related to landslide susceptibility, hazard, or risk zonation mapping.

Study Area

The majority of the study area lies in the Shimla and Solan districts of Himachal Pradesh and the Ambala district of Haryana in the northern part of India. Geographically, the study area lies between longitude 76°53′0′′E to 77°13′30′′E and latitudes 30°47′30′′N to 31°09′00′′N covering a cumulative area of approximately 609.91 km2 (Fig. 1). Kalka, Parwanoo, Solan, and Shimla are some of the major population centers. The study area is dominated by landslide-inducing factors like steep slopes, the highly dissected nature of the hills, high precipitation, and severe earthquake intensities. The annual average temperature varies between a minimum of 9.8 °C and a maximum of 25.9 °C. The study area receives an annual average rainfall of approximately 1251 mm. Geomorphologically, the area is composed of highly dissected hills with intermontane valleys. Major rock groups present in the study area are phyllite and quartzite of the Jutogh group, diamictite and shale of the Baliana group, shale of the Sirmaur/Dharamshala group, and sandstone and siltstone of the Dagshai and Kasauli group. Loamy mountainous soil is the predominant soil type in the study area. Seismologically, the study area is very active and frequented by many earthquakes. According to seismic zonation maps prepared and published by the Government of India, it lies in seismic zone IV and has an attributed peak ground acceleration value of 0.24 g (Bureau of Indian Standards New Delhi, 2002).

Fig. 1
figure 1

Location map of the study area and landslide locations

Full size image

Data and Methodology

Landslide Inventory Map

The landslide inventory is prepared in polygon form using multiple data sources, i.e., Bhukosh, the data dissemination portal of the Geological Survey of India, and interpretation of Google Earth images (Amatya et al., 2019; Batar & Watanabe, 2021; Hao et al., 2020; Roy & Saha, 2019; Shirzadi et al., 2018). The inventory is then rasterized to match the resolutions of Carto-DEM Ver. 3R1 and ALOS PALSAR DEM. The cumulative area of all the landslide polygons in the original inventory is approximately 647,081 m2. However, after rasterization, the final areas of the two inventory maps are approximately 655,997 m2 at the resolution of Carto-DEM Ver. 3R1 and approximately 667,188 m2 at the resolution of ALOS PALSAR DEM. This difference may be neglected as the total error in the cumulative landslide-affected area in both the inventory maps is only + 1.38% and + 3.10%, respectively. Hence, the prepared inventory maps are used for further analysis.

Landslide Causative Factors

Two geodatabases of DEM-derived landslide causative factors were generated from the selected DEMs, each comprising digital maps of elevation, slope, aspect, curvature, TRI, and distance to drainage, respectively (Figs. 2 and 3). The values of the causative factors were extracted at the landslide locations using raster sampling techniques. Variations were observed between these geodatabases because of the differences in their accuracies, processing technologies, and quality. On analyzing the extracted raster values using the Kruskal–Wallis test, it was seen that the distributions of elevation, slope, aspect, and curvature were statistically the same between the geodatabases while significant differences were observed in the distribution of TRI and distance to drainage. Table 1 presents the results of exploratory data analysis for the two geodatabases used in the study.

Fig. 2
figure 2

Thematic maps of the causative factors derived from Carto-DEM Ver. 3R1: a Elevation, b Slope, c Aspect, d Curvature, e TRI, and f Distance to drainage

Full size image
Fig. 3
figure 3

Thematic maps of the causative factors derived from ALOS PALSAR DEM: a Elevation, b Slope, c Aspect, d Curvature, e TRI, and f Distance to drainage

Full size image
Table 1 Spread of values of the landslide causative factors in the two generated geodatabases
Full size table

The elevation of a place affects factors like the degree of weathering of the slope-forming materials (Pham et al., 2016), the amount of precipitation received at a location, and local temperature, which further affect the stability of the hill slopes. The slope angle of the hill faces is one of the primary factors that can induce instabilities and lead to failures. The slope maps of the study area are derived using surface analysis of the DEMs. The aspect of the hill slopes impacts the landslide occurrences as the slopes facing different directions receive different amounts of rainfall and solar radiation, which may affect the vegetation density. Hence, slopes facing a particular direction may be more prone to landslides (Sarkar et al., 1995, 2013). Curvature is a geomorphological factor that determines the flow of water, thus, affecting landslide occurrences (Ayalew et al., 2004; Tien Bui et al., 2017). In this study, an upward convex surface is depicted as having positive curvature, flat surfaces have no curvature, while an upward concave surface has negative curvature. TRI is used to assess the overall terrain heterogeneity and degree of undulations of the surface features. TRI is defined as the mean absolute difference in elevation between a central pixel and its eight surrounding pixels on a DEM (Riley et al., 1999). The presence of drainage may aggravate the erosional activity on the hill faces due to excessive surface runoff. Hence, slope failures may be associated with the drainage channels (Sarkar & Kanungo, 2004). Therefore, the drainage network of the study area is extracted using watershed analysis of the DEMs. During analysis, the accumulation area is kept constant by proportionately adjusting the number of contributing pixels for both the DEMs. Then, Euclidian distance is calculated from the drainage networks to obtain the distance to drainage maps.

Spatial Frequency Distribution Analysis

After preparing the digital maps for landslide inventory and the landslide causative factors, raster values of all the causative factors are extracted at the landslide locations using raster sampling tools in GIS software and compiled in the form of a tabular worksheet by reclassifying them into different classes. Furthermore, the number of pixels occupying a particular class is converted to a percentage of the total number of pixels representing the cumulative study area. Similarly, the landslide-affected pixels within each class are also converted to a percentage of the cumulative landslide-affected area. Then the impact of the selected DEMs on the spatial characterization of the landslide locations is analyzed in terms of variations in the Landslide Severity Index (LSI) values calculated for all the landslide causative factors (Eq. 1):

$${\text{LSI}} = \frac{{{\text{Landslide }}\,{\text{affected }}\,{\text{pixels }}\,{\text{in }}\,{\text{a }}\,{\text{class}}\,{\text{ of }}\,{\text{landslide }}\,{\text{causative}}\,{\text{ factor}} \left( \% \right)}}{{{\text{Total }}\,{\text{number}}\,{\text{ of}}\,{\text{ pixels }}\,{\text{in }}\,{\text{a }}\,{\text{class }}\,{\text{of }}\,{\text{landslide }}\,{\text{causative }}\,{\text{factor}} \left( \% \right)}}$$
(1)

Results

The following sections elaborate on the results of the spatial frequency distribution analysis carried out to determine the impact when Carto-DEM Ver. 3R1 and ALOS PALSAR DEM are used for the terrain feature estimation at the landslide locations. The observations of the varying LSI values are also presented along with these results for both DEMs.

Dataset Derived from Carto-DEM Ver. 3R1

Table 2 summarises the spatial characteristics of the landslide locations based on the analysis conducted using the Carto-DEM Ver. 3R1 geodatabase. The elevation of the study area is divided into nine classes using Jenks natural breaks classification. The results reveal that the maximum landslide-affected area (18.58%) lies between the elevation range of 1598 m and 1758 m, followed by the elevation between 973 and 1166 m (17.77%) and between 1314 and 1452 m (15.33%). However, the LSI is the highest (2.06) at relatively lower elevations between 713 and 973 m followed by the range 973–1166 m (1.65) and 1598–1758 m (1.46). The slope is divided into eight classes using equal intervals classification for ease of comprehension of the results (Zhao & Chen, 2020). It is observed that the majority of the study area (60.39%) has a slope variation between 20° and 40°. Moreover, approximately three-fourths of the landslide-affected area (75.63%) also lies in the same class. LSI is highest (176.34) for the slope class between 60° and 70°, though the fraction of the study area for this class is very small (0.002%). The second and the third highest landslide severity occurs in the slope ranges of 50° to 60° (3.88), followed by 40° to 50° (2.90). The slope aspect is classified into nine classes using equal interval classification starting from 0°, denoting the north direction, and increasing clockwise till 360° to denote the north direction again (Rawat et al., 2015). The slopes that are not oriented in any direction are denoted as flat. The highest percentage of landslides-affected areas have an aspect angle between 180° and 225° (22.49%) followed by the area having an aspect angle between 225° and 270° (19.86%). The third most landslide-affected slopes have aspect angles between 135° and 180° (12.16%). The highest LSI values occur in the aspect range of 180° to 225° (1.49), followed by 225° to 270° (1.28) and 90° to 135° (1.10). In terms of curvature, higher landslide occurrences are observed in areas with positive curvatures (52.49%) as compared to negative curvatures (45.69%). They are the least in the flat areas (1.83%). Similarly, LSI is highest for areas with positive curvatures (1.08), followed by negative curvatures (0.95), and the least (0.56) for flat areas. TRI map is classified into nine classes using Jenks natural breaks classification. The largest proportion of area affected by landslides is in the TRI ranges of 41.95–49.20 m (23.70%), followed by 35.37–41.95 m (19.23%) and 49.2–58.3 m (18.62%). Progressively higher LSI values are observed for higher TRI values, with the maximum LSI occurring between 71.95 to 149.99 m (3.04) followed by 58.3–71.95 m (2.35) and 49.20–58.30 m (1.85). More than half (52.85%) of the landslide-affected locations lie within 450 m of the drainage features. However, LSI is observed to be highest (4.86) in areas located between 1050 and 1200 m from the drainage features.

Table 2 Spatial frequency distribution analysis of the landslide locations using the Carto-DEM geodatabase
Full size table

Dataset Derived from ALOS PALSAR DEM

Table 3 summarises the spatial characterization of landslide locations based on the analysis conducted using ALOS PALSAR geodatabase. Elevation of the study area is categorized into nine classes using a similar approach as previously used. Here, it is observed that the maximum landslide-affected area (18.47%) lies in the elevation range of 1310–1448 m, followed by 1162–1310 m (16.48%), and 1448–1595 m (15.57%). However, LSI is highest (1.94) at relatively lower elevation ranges of 712–970 m followed by 1595–1756 m (1.70) and 970–1162 m (1.57). The slope map is divided into eight classes using equal interval classification (Zhao & Chen, 2020). It is observed that 56.83% of the study area has a slope variation between 20° and 40°, encompassing 71.80% of the landslide-affected area. The maximum percentage of the landslide-affected area was observed in the slope range of 20° to 30° (37.53%) followed by 30° to 40° (34.26%) and 40° to 50° (12.24%). The highest LSI values occur in the slope range between 60° and 70° (8.92), followed by the locations having slope values above 70° (7.06) and the slope range between 50° and 60° (4.14). The highest percentage of landslide-affected areas have aspect angles in the range of 180° to 225° (24.72%) followed by 225° to 270° (15.68%) and 135° to 180° (13.37%). The highest LSI value occurs at locations with aspect angles between 180° and 225° (1.54), while the second and third highest LSI values arise in the aspect ranges between 135° and 180° (1.13) and between 90° and 135° (1.11), respectively. In terms of curvature, 45.85% of the landslides occur in areas with positive curvatures, 38.09% occur in areas with negative curvatures, and 16.06% occur in flat areas. LSI value is highest on slopes with positive curvatures (1.13), followed by the slopes with negative curvatures (0.94), and the least for areas that are flat (0.86). As done for the Carto-DEM geodatabase, the study area is divided into nine TRI classes, with most of the study area having TRI values between 0 and 24.86 m (93.53%). The highest LSI value occurs in the TRI range of 47.92–147.78 (8.02), followed by 32.08–47.92 m (3.94) and 24.86–32.08 m (2.32). In terms of the distance to drainage, 55.64% of the landslide locations lie within 450 m of the drainage features. The highest LSI value is observed at a distance between 900 to 1050 m (3.19), followed by 1050–1200 m (2.86) and 750–900 m (2.04) from the drainage features.

Table 3 Spatial frequency distribution analysis of the landslide locations using the ALOS PALSAR geodatabase
Full size table

Discussions

DEM provides the fundamental framework for terrain feature estimation and modeling for landslide-related analyses using Remote Sensing and GIS tools. DEM-derived parameters are often used as input factors for modeling geomorphological conditions in landslide susceptibility, hazard, and risk zonation mapping studies. It is reported that the source and resolution of the DEMs used in such studies influence the prediction accuracy of the employed models (Chow & Hodgson, 2009; Deng et al., 2007). This influence is mostly attributed to the impact of variations in the DEM-derived parameters on the results of the simulation (Sarma et al., 2020). The present study analyzes this impact using two geodatabases of DEM-derived landslide causative factors generated using Carto-DEM Ver. 3R1, which is based on a photogrammetric approach that extracts the elevation data from stereo images, and ALOS PALSAR DEM, which is generated using SAR technology. The variations in terrain feature estimation resulting from these two DEMs are highlighted by conducting a comparative study of the spatial frequency distribution analysis of the landslide locations. Table 4 summarizes these variations in the form of absolute maximum differences observed during the analysis of the two geodatabases for each of the causative factors. The variations are reported as percentages by taking the ALOS PALSAR geodatabase as a reference, as it has a finer resolution.

Table 4 Absolute maximum differences (%) between corresponding values of landslide causative factors from the two geodatabases used in this study for (1) percentage of the study area (2) percentage of landslide affected area (3) LSI values in a particular variable class
Full size table

The findings show a difference in the evaluated landslide severity between the two geodatabases for elevation, as measured by the highest LSI values. Comparing the two geodatabases, it is observed that the greatest LSI value is marginally lower, and the second and third highest LSI values are marginally higher (Tables 2 and 3) in the Carto-DEM geodatabase. Most landslides occurred in moderate-sloped regions characterized by slope gradients between 20° and 40° in both the geodatabases. Such observations are interesting because it is intuitive to think that landslide activity should be elevated in areas characterized by higher slope gradients. However, similar observations have been reported in other studies in different parts of the world where landslides occur in regions characterized by low to moderate slope gradients (Naseer et al., 2021; Xu et al., 2014). This is because of the distribution of the slope-forming materials where steeper slopes are usually composed of rocky materials that effectively transfer all their weight to the toe of the slope. Approximately two-fifths of the landslide-affected areas exhibited an aspect angle between 180° and 270°, although the distribution of these areas between the intervals 180°–225° and 225°–270° varied between both the geodatabases. Similar observations are also reported in other studies of the Indian Himalayan region, which report that slopes oriented in the south-southwestern direction are more prone to landslides (Sarkar et al., 2013).

Significant variations were observed in the characterization of the study area for all the causative factors, with the maximum variation being observed for curvature (Table 4). Also, the landslide occurrences in flat terrain were found to be significantly more prominent (8.79 times) in the ALOS PALSAR geodatabase. In both the geodatabases, progressively higher LSI values were observed at locations with higher values of TRI, which is logical because the higher values of TRI indicate that the locations have a higher elevation difference from the surrounding locations. Furthermore, more than half of the landslides occurred within 450 m of the drainage features for both geodatabases. The literature supports similar observations of landslides occurring near drainage features (Chawla et al., 2019; Sarkar & Kanungo, 2004). However, the greatest LSI values are observed farther away from the drainage features for both the geodatabases, in contradiction to the general observations by other researchers (Asmare, 2023; Jones et al., 2021). Moreover, as compared to the observations from the Carto-DEM geodatabase, the areas of higher LSIs were situated relatively nearer to the drainage features in the ALOS PALSAR geodatabase. On further investigation employing the interpretation of remotely sensed satellite images, it was concluded that higher LSI values farther from the drainage features may have resulted from anthropogenic activities in the form of residential construction and other infrastructure development. The variations in the estimated distances from drainage features of these high LSI values were attributed to the effects of the two DEMs on the characterization of the landslide regions.

Conclusions

The accuracy of a DEM in representing the topographical characteristics of a given location is crucial for landslide-related studies. In particular, the accuracy of the results in such analyses is significantly influenced by the variations in the DEM-derived parameters. Thus, this study investigated the variations in the spatial characterization of landslide locations, employing Carto-DEM Ver. 3R1 and ALOS PALSAR DEM. Differences were observed in terrain feature estimation by conducting spatial frequency distribution analyses of the geodatabases generated from these DEMs. The results of the analyses reveal substantial variations between the two geodatabases employed.

Though both DEMs exhibited comparable topographical features, discrepancies were observed upon closer scrutiny. Although the derived causative factors from both the selected DEMs successfully captured the general characteristics of the landslide locations, the distribution of landslide-affected areas varied significantly. A comparative analysis of the generated geodatabases showed that the slope gradients in the Carto-DEM geodatabase between 20° and 30° varied by + 7.45% and 30°–40° varied by − 19.41%, respectively as compared to the ALOS PALSAR geodatabase. A notable difference was observed in the landslide areas with flat terrain, primarily because of the difference in the resolutions of the two DEMs. Terrain ruggedness and drainage features were captured fairly well by both the DEMs with progressively higher values of LSI being observed for increasing TRI values. A comparative analysis of the LSI values revealed significant differences in terms of slope, with the Carto-DEM exhibiting LSI values over 18 times larger than the ALOS PALSAR DEM.

The results of this study suggest that the source and resolution of the selected DEMs may influence terrain feature estimation, thus affecting their usability in landslide-related studies. The choice of a suitable DEM for terrain modeling and surface feature estimation may be impacted by the availability of data. However, the choice greatly depends upon the specific setting of the investigation, and it is crucial to acknowledge that topographic characteristics exhibit considerable variation across diverse landscapes. Consequently, the selection of an appropriate DEM depends upon the objectives of the research, the characteristics of the study location, and a comprehensive assessment of the employed models.