GIS-based frequency ratio and Shannon entropy modeling for landslide susceptibility mapping: A case study in Kundah Taluk, Nilgiris District, India

2.1 Study area description

The mountainous Kundah Taluk, located in the Nilgiris district of Tamil Nadu, lies within the vulnerable Western Ghats range. The elevation ranges from 740 to 2,625 m above the mean sea level at the Doddabetta peak. Geographically, Kundah Taluk is positioned at a longitude of 76.6482°E and a latitude of 11.2810°N, covering an area of 331 km². The taluk’s climatic characteristics include wind speeds fluctuating between 5 and 5.4 km/h, with an average relative humidity of approximately 77%. Rainfall varies from 400 to 417 mm. The region’s geology is shaped by tectonic forces, rolling hills, weathering, erosion, valleys and plateaus, predominantly underlain by charnockite associated with granite and gneisses. Colluvial fills, shallow pediments and deep pediments are identified as landslide-prone geomorphological areas. The taluk features diverse land uses, including water bodies, tea plantations, settlements, open forests and agro-horticultural plantations. However, the area is structurally prone to disturbance and faulting, leading to frequent landslides. Consequently, the region experiences property damage, loss of life and regular disruptions in road networks and communication. The location map of the study area is shown in Figure 1.

Figure 1

Location map of the study area.

2.2 Landslide inventory map

Landslide events in a region, including their distribution, distortion patterns, repetition, dominant type and quantitative characteristics associated with slope failures, were assessed using landslide inventory maps. Furthermore, the evaluation of the geographical possibility of landslide initiation and runoff, along with the likelihood of possible landslide sizes for a specific return duration, is made possible using landslide inventory maps [52]. A visual inspection of Google Earth imagery, in-depth field surveys, a review of previous landslide records in pertinent literature, the analysis of satellite photos, technical and scientific papers and databases maintained by the government were all included in creating a landslide inventory. In this study, 581 landslide locations were found and mapped as point features on the inventory map, which were split into datasets for training and testing as depicted in Figure 1. Utilizing the subset features tool in ArcGIS 10.8.2, this study systematically partitions the landslide inventory data into training and testing sets while preserving multipart features, facilitating model development and accuracy assessment. Although there is no universal standard for partitioning landslide occurrences into training and testing datasets, it is common practice in many research studies to allocate 70% (407) of the landslide events for training datasets, which are used to build the weight system for the contributing factor for generating a LSM and 30% (174) of past landslides are reserved for the model’s validation and performance evaluation [53]. Figure 2 shows a flowchart of the methodology.

Figure 2

Methodology flow chart.

2.3 Landslide conditioning influencing factors

A crucial stage in mapping landslide susceptibility is selecting the conditioning factors that will serve as input variables in the model. The accuracy, resolution and reliability of the input data substantially influence the outcomes of the statistical models. Slope, elevation, soil, aspect, Normalized difference vegetation index (NDVI), lineament density, rainfall, land use and land cover (LULC), drainage density, geomorphology and road are the conditioning elements listed in Table 1 used in this study [34].

2.4 Slope

The slope exerts a substantial impact on landslide incidence and significantly influences the water flow on the surface, which directly affects the surface runoff and penetration [54]. The risk of landslides escalates with steeper slopes due to increased shearing stress and gravitational forces. For this investigation, a slope angle map was produced from a digital elevation model (DEM) and categorized into four subclasses: nearly level (0°–3°), moderate slope (3°–10°), steep (10°–35°) and very steep (>35°) (Figure 3a).

Figure 3

Distribution of Landslide evaluative factors: (a) Slope, (b) elevation, (c) aspect, (d) soil, (e) lineament density, (f) drainage density, (g) NDVI, (h) LULC, (i) geomorphology, (j) road and (k) rainfall.

2.5 Elevation

Elevation measures a landscape’s height above sea level, with landslides more likely to occur at higher elevations than lower ones. Soil strength at higher elevations may be less robust than at lower elevations [8]. Moreover, the elevation of the region affects rock weathering and the effects of snow and rain. Landslide occurrences are influenced by elevation, especially in hilly regions characterized by moderate to steep slopes, abundant rainfall and limited vegetation cover [10]. The elevation map of Kundah Taluk shows values ranging from 740 to 2,625 m (Figure 3b).

2.6 Aspect

Aspect significantly influences a location’s susceptibility to landslides due to the effects of the sun, wind and precipitation [55]. The slope aspect also indirectly impacts soil moisture and vegetation. For this study, the aspect map generated from a DEM was classified into ten directional categories: flat, east, north, south, northwest, southwest, north, southeast, northeast and west (Figure 3c).

2.7 Soil

Soil is a fundamental factor in landslide events, primarily due to its composition and texture. Soil texture, combined with erosion affects properties such as permeability and cohesion, which influence the nature and intensity of slope movement and have a significant impact on landslide susceptibility [56–58]. Due to the region’s steep slopes and heavy rainfall, landslides are common in the sandy clay loam, sandy loam and rock land soil types (Figure 3d). Studies have indicated that soil erosion, particularly in regions with fine-textured soils, accelerates the initiation of landslides. Removing topsoil, particularly in locations where structural integrity is compromised by fractures, exacerbates the landslide risk [59]. Consequently, understanding the interaction between soil texture and erosion is essential for effective landslide mapping and risk assessment in Kundah Taluk.

2.8 Lineament density

Lineament density is a key factor in assessing landslide susceptibility. Lineaments, often associated with fault zones, fractures and other structural features, influence slope stability by creating weak zones. They indicate the surface indicators of underlying structural elements, revealing areas of faulting and fracturing [60]. Elevated lineament densities often correspond to areas with increased geological disturbances, making these regions more prone to landslides [61]. A lineament density map, prepared using the ArcGIS spatial analysis tool, highlights the concentration of linear structural features that influence landslide susceptibility. The map is categorized into five classes, namely, very high, high, medium, low and very low (Figure 3e).

2.9 Drainage density

The drainage density was determined by dividing the cumulative stream length by the total area under study. Regions with high drainage density usually have more intense drainage networks, accompanied by erosion along drainage lines, which increases the risk of landslides [62]. Inadequate drainage systems near roads and infrastructure also elevate landslide risk by increasing water absorption and weakening soil strength. Five drainage density classes were identified as very low, low, medium, high and very high (Figure 3f).

2.10 NDVI

In this study, the NDVI was derived from Landsat 8 imagery collected in August 2023. This period aligns with the region’s growing season, characterized by peak vegetation due to the southwest monsoon rains, providing optimal conditions for accurate NDVI analysis. The NDVI effectively reflects the influence of plant cover on soil stability and slope hydrology, making it a valuable indicator for assessing landslide susceptibility [63] (equation (1)).

The NDVI map of Kundah Taluk was created for this study using the following calculation formula:

where R stands for the electromagnetic spectrum’s red bands and NIR is the Earth’s surface reflectance in the near-infrared channel. The categorization of NDVI is shown in Figure 3g.

2.11 LULC

The LULC is a critical factor in determining the occurrence of landslides. They significantly influence the durability of slope materials and control the water content within the slopes. For instance, plant roots help to stabilize sandy slopes and are generally considered to reinforce them. The land use map for Kundah Taluk was produced through data collection, RS and GIS analysis. Initial data were gathered from Landsat 8 satellite imagery followed by field surveys. The collected information was used to perform supervised classification of the images into distinct land use categories, such as agro-horticulture plantations, open forest, land without scrub, settlements, dense forest, land with scrub, tea plantations and water bodies (Figure 3h). Using GIS software, the data were processed to create land use patterns and their implications for disaster risk management in the region.

2.12 Geomorphology

The geomorphological map offers valuable insights into the processes of lithology, constructions and geographical elements that influence landslide susceptibility by highlighting critical geomorphic units, landforms and fundamental geological features. In Kundah, shallow pediments comprise 20% of the region, whereas colluvial fills make up 50% of the overall area. Some of the distinctive features observed in Kundah Taluk are dissected plateaus, undissected plateaus, residual hills, debris slopes, pediments, shallow pediments, floodplains, deep pediments, moderate pediplains, colluvial fills, crust lines, gullied valleys, pediplain and erosional plateaus (Figure 3i).

2.13 Road

Landslides frequently occur near and along road slopes. Particularly in hilly areas, uncontrolled road excavation weakens the surrounding rock mass. An important human component to consider when assessing a site’s susceptibility to landslides is its proximity to roads. The existence or lack of road networks in an area influences the probability of landslides [64]. Roadside excavation and vegetation removal during road construction can trigger landslides. Consequently, proximity to the roads is a significant factor in landslide studies [8]. The road proximity was categorized into three buffer distances: <100 m, 100–200 m and 200–300 m with intervals of 100 m (Figure 3j).

2.14 Rainfall

Rainfall triggers landslides. Excessive rainfall saturates the soil, increasing its weight and reducing shear strength owing to higher pore water pressure within the slope materials [3]. Rainfall is the primary trigger of landslides on hilly slopes. Rainfall quantity and intensity are influenced by topographical features, which vary across space and time, thus impacting landslide susceptibility in the region [65]. Monthly rainfall data for Kundah Taluk were obtained from rain gauge stations (Kil Kundah, Kil Kundah 2, Balacola 2, Ithalar and Ithalar 2) provided by the Indian Meteorological Department’s database. Five different categories were created from the data by applying the inverse distance weighted approach (Figure 3k) to map the amount of rainfall.

2.15 LSM

LSM involves the creation of maps that identify areas at risk of landslides based on a range of influencing factors. The two bivariate models used in this study are the FR and the SE models. The resulting LSM is used to pinpoint high-risk regions for focused intervention and mitigation measures. Both models aim to enhance land use planning and disaster risk reduction strategies.

2.16 FR model

The FR model was used to quantify the likelihood of a relationship between several influencing factors. This technique helps to predict the development of landslide trends and better identify the key factors contributing to their occurrence [23]. A ratio greater than 1 indicates a strong relationship between the factors, while a ratio less than 1 suggests a moderate correlation. For each contributing factor, the FR is computed for every subclass of causative factors by dividing the percentage of landslide occurrences by the percentage of pixel area occupied [66,67], as shown in equation (2).

Following equation (3), the FRs of all causative factors considered were added to determine the landslide susceptibility index (LSI).

A higher LSI value indicates a greater susceptibility to landslide, while a lower value signifies a reduced likelihood of landslide occurrence.

2.17 SE model

Shannon’s entropy has been widely applied in evaluating landslide susceptibility, utilizing knowledge-based weightings for various parameter subcategories to account for regional variations [8,15]. This method provides a quantitative measure of a system’s instability, inequality and unpredictability, while also offering insights into the future trends of specific systems.

To evaluate the environmental dissimilarity and examine the potential impact of each element on landslide occurrence, the present study has employed Shannon’s entropy technique. Higher values of the entropy index indicate that a factor has a larger effect on landslides [68].

Entropy values are instrumental in analyzing natural hazards such as landslides. Specifically, entropy values indicate environmental dissimilarities, highlighting factors that are potential triggers for landslides. The entropy index calculation involves the following steps to determine the weighting of the causative factors:

where P i j is the probability density for each class I in factor j, H j is the entropy, H j max is the maximum entropy value, S j is the number of classes, I j is the information coefficient of factor j, W j is the final weightage of each factor and FR is the frequency ratio.

3.1 Assessment of FR

Landslide incidences and the 11 causative factor components were mapped and analyzed spatially using a FR model. The following variables were included: slope, soil, elevation, aspect, drainage density, NDVI, LULC, geomorphology, rainfall, lineament density and road. The FR of each class of landslide conditioning factor was calculated. According to Pradhan [69], values greater than 1 imply a higher risk of landslide events, whereas values below 1 indicate a lower likelihood. Table 2 presents the FRs of all classes and their subclasses which can effectively assess the correlation between the landslide spatial distribution and their triggering factors in a study region [18].

Slopes with angles between 0° and 3° had a very low impact on landslide occurrence. However, with an increasing slope between 3° and 10°, the FR (1.4107) increased suddenly, indicating a higher likelihood of landslides. Most landslides in this study occurred in areas with 3°–10° slopes. The FR for 10°–35° was also high (0.9915) [12,70]. Elevated slope values intensify gravitational pull and escalate shear stress, thereby exacerbating the risk of landslides. Elevation ranges of 2,625–1,439 m and 740–1,311 m, have respective FR values of 0.2401 and 0.4188, indicating a low likelihood of landslides. The greatest FR value is 3.8053 for the elevation range of 1,664–1,936 m, followed by 0.6826 for 1,936–2,188 m and 0.6582 for 1,311–1,664 m [26].

Landslide occurrences were more heavily influenced by the southeast, south, east and west aspects than any other direction. The southeast aspect exhibited the highest FR of 1.6901, then the south, east and west aspects with FRs of 1.5471, 1.2164 and 1.0097, respectively. These areas are subject to increased erosion and rainfall, which increase the risk of landslides. FR values are below 1 for other aspect classes, on the other hand, indicating a reduced likelihood of landslides.

As for soil factors, sandy clay loam has a high FR value (2.5277), indicating an increased likelihood of landslide events. The soil classes with sandy loam and rock land had the lowest FR values (0.1368 and 0.2359), indicating low odds of landslides in the region. Areas most affected by landslides, with a low NDVI range (0.107783–0.20765), exhibited the highest FR of 1.5308.

The results show that for land use, the areas with the settlement, tea plantations and open forests had FR values of 4.1788, 1.6232 and 0.6488, suggesting a varying likelihood of highest landslide occurrence. According to Solaimani et al. [71], exposure to soil moisture and erosion causes this increased sensitivity. Regarding geomorphology, the classes of colluvial fills, shallow pediments and deep pediments had the greatest FR values (19.3626, 1.5041 and 0.8391) highlighting the high likelihood of landslide events. Low FR values of 0.0988 and 0.1176 for floodplains and pediments, respectively, imply that landslides are unlikely to occur in these regions [72].

Furthermore, road distance classes of 100–200 m and 200–300 m had a negligible impact on landslide occurrence (FR values of 0.4432 and 0.3589), whereas distances less than 100 m had a substantially higher influence (FR = 1.7397). This suggests that landslide occurrence is more likely along shorter routes.

The maximum FR (1.2798) for drainage density occurred in the medium range (1.61382–2.87488), implying a high chance of landslides, while the minimum FR value (0.4835) for drainage density in the low range (0.64538–1.61382) suggests a lower probability for landslide occurrence [73]. According to the correlation between lineament density and landslide probabilities, the very low range exhibited the maximum FR value of 1.1505. Conversely, the extremely high range showed the minimum FR value of 0.4211 [74].

Precipitation is a key triggering factor for landslides, with higher FR values observed in areas with higher rainfall. The greatest FR values, 1.5715 and 1.1856, were found in locations with rainfall in the range of 405–410 mm and 410–417 mm, respectively [12,75]. Generally, the most significant factors influencing landslide occurrence include slope classes of 3°–10°, elevation ranging from 1,664 to 1,936 m, south-eastern facing aspects, soil composed of sandy clay loam, extremely low density of lineaments, moderate drainage density, low NDVI classification, land use categories such as settlement, tea plantations and open forests, geomorphological features like deep pediments and colluvial fills, proximity to roads within 100 m and precipitation levels between 405 and 410 mm, resulting in a high susceptibility to landslides in the region. The LSI values of the FR model range from 3.38656 to 40.1292. The zonation of the LSM (Figure 4) was generated in ArcGIS using the Jenks natural breaks classification method [27]. These zones indicate areas with very low, low, moderate, high and very high susceptibility to landslides. It is noted that 31.10% of the region falls within the low susceptibility class, while 4.729% is classified under the extremely low susceptibility category. A total of 34.70% of the area is categorized as having moderate susceptibility category. Furthermore, Table 3 illustrates that 29.46% of the taluk is highly prone to landslides.

Figure 4

LSMs of FR and SE models.

3.2 Assessment of SE

The results of Shannon’s entropy model are listed in Table 2, which presents the weighted contribution of each causative element to landslide susceptibility. According to the P i j data, a slope class of 10°–35° and a slope degree interval of 3°–10° has the greatest likelihood of experiencing landslides. Conversely, the probability is smaller for other slope classes [76]. Compared to other elevation classes, the range between 1,664 and 1,936 m shows a higher risk of landslides. Field observations verify that a higher elevation of this region has a substantial impact on the ongoing landslide incidents of the taluk. In terms of aspect, most landslide-prone locations in the study area are slopes facing southeast, followed by facing south and east.

The most likely soil types for landslides are sandy loam and sandy clay loam, the probability of landslides on rock land is often lower. Landslide occurrences are significantly affected by lineament density, as seen by the highest P i j value (0.4208) reported in the very low range [77]. An elevated risk of landslides is indicated by the high P i j value of 0.2810 for the medium drainage density region (1.61382–2.87488) [78]. It is evident from the highest P i j value (0.3375) in areas with low NDVI, showing a greater likelihood of landslides in regions with less vegetation. As NDVI values increase, a decrease in landslide probability is observed, revealing a lower risk of landslides in the study area.

Regarding LULC, tea plantations and settlements have the highest possibility of experiencing landslides, whereas water bodies have the lowest chance [79]. The habitat area was found to be the primary concentration in the landslide-prone area. The geomorphological analysis reveals high P i j values of 0.8090 and 0.0628 for deep pediments and colluvial fills, respectively, indicating their proneness to land collapses [80]. Less than 100 m road has the highest P i j value (0.6844) across proximity to road classes, followed by 200 m (0.1744) [81].

Similarly, the 405–410 mm rainfall class has the highest P i j value (0.3519). The SE model assigns more weight to geomorphology, soil, elevation and land use factors. It is essential to recognize that landslide causes vary significantly with location characteristics and data quality [82]. The LSI values of the SE model fall from 0.506763 to 27.9323. As shown in Figure 4, these values were classified into five classes: very high, high, moderate, low and very low, based on the Jenks natural breaks classification method. In Kundah taluk, 35.99% area are less susceptible to landslides. The taluk’s area is further divided into high (20.97%), very high (17.58%) and moderate (25.43%) susceptibility classes, as presented in Table 3.

3.3 Assessment and comparison of model results

In risk prediction modeling, validation is crucial, as data quality and model selection substantially influence the reliability of outcomes. This study developed statistical models incorporating various causative factors and validated using the receiver operating characteristic (ROC) method’s area under the curve (AUC) [83]. The ROC curve effectively illustrates the LSI, with the Y-axis displaying the downward-ordered index and the X-axis indicating the overall percentage of landslide occurrences relative to the prediction rate [84,85]. Models with higher AUC values, ranging from 50 to 100%, are considered more accurate for prediction. The study utilized a test dataset (30% or 174 sites) to generate the prediction rate using LSMs created by the FR and SE models, as depicted in Figure 5. The AUC analysis revealed that the SE model achieved an AUC value of 0.82, indicating good accuracy.

Figure 5

ROC curve for FR and SE models.

In comparison, the FR model demonstrated lower accuracy with an AUC of 0.816. Both models consistently identified landslide susceptibility zones within the study area, with the SE model showing a notable advantage in overall predictive accuracy. These AUC values align with numerous studies conducted in various landslide-affected regions, including Karnataka [32], Kerala [51], Morocco [85], Western Ghats [32], Himalayas [86], Meghalaya [34,65], Ethiopia [15,53] and the Union Territory of Jammu and Kashmir [87].