The district Ghizer lies in Hindu Kush region of Pakistan in the northern part of Gilgit-Baltistan, between latitude 36.0˚N to 37.0˚N and longitude 73.0˚E to 74.0˚E covering an area of 12,042 km². The estimated terrain elevation above sea level is 3661 meters and the habitat is arid to semi-arid. The region has four Tehsils i.e. Gupis, Ishkoman, Punial and Yasin (Figure 1). The area is prone to different natural disasters although snow avalanches, landslides, and earthquakes

are frequent in the area [7] . Hindukush mountain region is considered as seismically active zone because of the occurrence of low-intensity earthquakes at frequent intervals. Various fault lines spread through entire Gilgit Baltistan region. Low to medium intensity earthquakes are frequent mainly responsible for the occurrence of landslides, Glacial Lake Outburst Flood (GLOF), avalanche, rock fall, and edge failure in the study area. The valleys are present in steeps hills and upper parts of the region get isolated for several months in the winters due to heavy snowfall, landslides and snow avalanches.

The road from district Ghizer to district Gilgit is highly susceptible to landslides because of erosion and rock fall as the slopes are made of muddy dust and loose sediments. There exists sixteen major landslide areas and thirteen small villages prone to rock fall, so as a whole 53 villages are considered to be at high risk to hazards in district Ghizer [7] .

In this study, twelve factors were used to generate the landslide susceptibility map. The factors were totally selected on the basis of their effectiveness and availability. According to Oh and Pradhan [8] , the assessment of the local landslide areas should be convenient and relevant and the factors should be illustrative and effectively available. The factors; slope, aspect, elevation, drainage network, SPI, and TWI were extracted from Digital Elevation Model (DEM) of 30 m resolution acquired from USGS Earth Explorer.

The slope angle ranges from 0˚ - 73.76˚ for the study area (Figure 2(a)). Slopes were reclassified into five classes i.e. <5˚, 5˚ - 15˚, 15˚ - 30˚, 30˚ - 45˚ and >45˚. According to [6] , highest landslide susceptible class is 30˚ - 40˚ slope angle which consists of steep slopes. The aspect map (Figure 2(b)) was divided in to nine classes based on dimensions flat (−1)˚, north (337.5˚ - 360˚, 0˚ - 22.5˚), north-east (22.5˚ - 67.5˚), east (67.5˚ - 112.5˚), south-east (112.5˚ - 157.5˚), south (157.5˚ - 202.5˚), south-west (202.5˚ - 247.5˚), west (247.5˚ - 292.5˚), and north-west (292.5˚ - 337.5˚). According to [9] the southwest and northwest facing slopes are highly susceptible to landslides. [26] suggest assigning higher

weights to southwest, west, and northwest facing slopes. The lowest point of district Ghizer is at elevation of 1662 m and the highest point of elevation is at 6789 m (Figure 2(c)). It was categorized into five classes 1162 - 2894 m, 2894 - 3668 m, 3668 - 4265 m, 4265 - 4829 m, 4829 - 6789 m. Landslide occurrence is linked to certain elevations [10] . [9] investigated that 64% of reported landslides were recorded at elevation of 2000 - 4000 m and 24% were observed at elevation of 1000 - 2000 m. Buffers were created around the drainage network and classified into 0 - 500 m, 500 - 1500 m, 1500 - 2500 m, 2500 - 5000 m and <5100 m (Figure 2(d)). Landslides increase if the distance to streams or rivers is decreased, due to slope instability which leads to erosion. Irregularities and fragmentations are caused by a river’s longitudinal profile due to slope failures [11] . The secondary attribute SPI (Figure 2(e)) tells the net erosion and net deposition in the areas of increased flow rate and decreased flow rate [12] . Furthermore, landslide susceptibility is higher with the higher SPI values. It was reclassified into four class values −13 - 0, 0 - 5, 5 - 10 and 10 - 14. The other secondary attribute TWI (Figure 2(f)) calculates the extent of water accumulation at a place; its higher values show higher accumulation causing more landslide susceptibility [13] . It was reclassified into three classes i.e. 3 - 9, 9 - 28 and <28.

From the geological map of scale 1:5,000,000 fourteen rock types were digitized, which was acquired from Geological Survey of Pakistan (Figure 3(a)). All the lithological units were categorized in view of their capability to trigger landslide Table 1. Each lithological unit has its own susceptibility towards landslides, so it needs to categorize lithological units accordingly [14] . Ranking of the rock types was based on their potential to cause landslide [1] . The rock formations in the study area are categorized into five classes from A to E; each

rock formation is categorized into very highly stable, highly stable, moderately stable, less stable and least stable class; based on the geotechnical properties of the rock units present in the formation. Less stable rock units are highly prone to slope failures which cause landslides. Geological map of scale 1:5,000,000 was used to digitize the fault lines of the study area which were acquired from Geological Survey of Pakistan (Figure 3(b)). Buffers were created for distances of 0 - 3000 m, 3000 - 7000 m, 7000 - 11,000 m, 11,000 - 15,000 m and <16,000 m. Shearing causes rocks weak which are close to fault lines, consequently leading to landslide susceptibility [15] . Monthly average rainfall for different locations of district Ghizer was acquired from Tropical Rainfall Measuring Mission (TRMM) for the years 2006-2015 (Figure 3(c)). Rainfall is an important landslide triggering factor, but it is limited to the monsoon season [9] . Rainfall raster data map was prepared using the Inverse Distance Weighted (IDW) interpolation. The infrastructure of the study area is poor and no such complex road network exists (Figure 3(d)). Road network data was acquired from an online source http://www.mapcruzin.com. Buffers were created for roads in the study area at distance of 0 - 500 m, 500 - 1500 m, 1500 - 2500 m, 2500 - 5000 m and <5100 m. Cutting of slopes for road construction or road widening in the hilly regions could lead to slope failures causing landslide susceptibility [16] . Landsat 8 OLI image for the year 2015 and L4-5TM image for 1999 were acquired from USGS Earth Explorer. Land cover maps were prepared using maximum likelihood supervised classification techniques in ERDAS Imagine 14 (Figure 3(e)). The classes prepared were glacier, vegetation barren soil/ exposed rocks and water. Land cover images for 1999 and 2015 were compared to the land cover change and its impact on a landslide. Accuracy assessment of the classified images (1999 and 2015) was calculated to check the classification accuracy. The accuracy assessment was generated using 50 random points.

Soil sampling for soil texture was performed by taking twelve composite samples from each tehsil of the study area. The samples were air-dried and sieved through 2 mm size sieve. Forty ml of 1% sodium hexa meta-phosphate and 150 ml of distilled water was added to soil sample (40 g) and was kept overnight. The mixture was stirred for almost 10 minutes and was put in a graduated cylinder for readings, which was recorded with Boyoucos Hydrometer method [17] . IDW interpolation method was used to create the raster map of soil texture (Figure 3(f)).

The Analytical Hierarchy Process (AHP) is an adaptable tool which is created by [18] and it is used for various decisions makings such as suitability analysis and susceptibility analysis. It is considered to be a rational decision-making process for multi-criteria as well as for multi-target approach.

In the comparison matrix, the numerical value for each factor was between 1 and 9 (Table 2). The factors were organized hierarchically in the matrix and the Prioritized Factor Rating Value (PFRV) technique was used to assign a numerical value to the factors in the AHP on the basis of their importance as compared with other factors. The numerical value assigned to the factors was based on, expert knowledge, literature, observations, and experiences.

Source: Saaty 1977.

The average of the hierarchically arranged factors was used to calculate the weights and rating value/eigenvalue along with the Consistency Ratio (CR), based on the prepositions of [19] . [20] expressed that the eigenvalue “λ_max” and the total number of factors “n” are same for a consistent comparison matrix.

CI = Consistency Index which is as follow:

C I = λ max − n n − 1 (1)

The consistency of the comparison matrix is checked through CR (Saaty 1977).

C R = C I / R I (2)

where RI = Random Consistency Index.

[20] have created RI by utilizing scales 1/9, 1/8, 1/7… 1… 8, 9. The average RI of 12 matrixes is given in (Table 3).

The calculated CR from the comparison matrix for the 12 factors was 0.028. This value demonstrates that the matrix of the factors is acceptable. The result of AHP showing weights of causative factors (Wj) and the factor rating values (wij) are given in the (Table 4).

Source: Saaty 1977.

Weighted Linear Combination (WLC) is comprised of both subjective and quantitative strategies and depends on the qualitative map combination approach (heuristic analysis) [21] . It is the last step in making the landslide susceptibility map in which all the weighted layers were combined using weighted overlay technique in ArcGIS 10.1. All the layers were reclassified to a typical scale and the vector layers were rasterized. The weights of the factors were linearly combined (WLC) to obtain the Landslide susceptible Index (LSI) according to the formula:

L S I = ∑ j = 1 n W j w i j (3)

where LSI is Landslide susceptibility index, Wj is weight value for parameter j, wij is rating value or weight value of class I in parameter j and n is no. of classes.

The weights of the factors; slope, aspect, elevation, drainage network, SPI, TWI, lithology, fault lines, rainfall, roads, land cover land use and soil were derived using AHP by Prioritized Factor Rating Value (PFRV) (Table 3). The final landslide susceptibility map was generated using these weights in the WLC. The resultant map showed that the pixel ranking value for landslide susceptibility varies from very low (1.53) to very high (4.43). The areas with high pixel values have more chance of landslide as compare to the low pixel values. The categorization of the pixel ranking values was obtained by natural breaks in GIS.

Based on the above categorization, the area and percentage of the five susceptibility classes were also determined (Figure 4). Very low susceptibility class covers an area of 8.66% while; low susceptibility class covers 16.96% of the area. In addition, a larger extent of the area lays in the moderate category i.e. 28.14%. Furthermore, the high susceptibility class is the one which covers a larger area in the district Ghizer i.e. 28.22%. The very high susceptibility class in the district falls over an area of 18.02%. Hence, in district Ghizer, a total of 74.38% of the surface area falls into the moderate to very high landslide susceptible zones whereas 25.62% of the area falls into low to very low landslide susceptible zones.

The topographic, geologic, and hydrologic factors causing landslides are considered as stationary, while land cover is the factor that can change within a short time; therefore it is in a direct relation to landslide occurrence [22] . In this regard, Temporal assessment of land cover change was studied for the years 1999 and 2015, to analyze the difference in the land cover change over sixteen years in the district Ghizer and its impact on landslides. Hence, between the years 1999 till 2015 a number of landslide events have occurred and significant changes in

land cover have been observed. The changes are visible in the classified maps (Figure 5). It showed a major decline in the glaciers from 9.79% to 6.63%. As the district Ghizer is largely covered by barren soil/exposed rocks, it poses more vulnerability to landslides. Barren slopes have more chances of erosion as compared to areas with vegetation so they are more susceptible to landsliding [23] . However, the barren soil/ exposed rocks have reduced to 78.46% from 81.465. Vegetation cover has increased from 8.33% to 12.09%, while water class which was least area covering class in 1999, increased from 0.41% to 2.82%. The result of overall classification accuracies for the year 1999 and 2015 from the accuracy assessment were 80.0% and 80.01% respectively. In most of the studies overall classification accuracies target below of 85% [24] .

There is number of methods to validate a susceptibility map. One such method is computing landslide frequency/density in the susceptibility classes [25] . In this study, landslide susceptibility map validation is prepared by computing landslide frequency in the susceptibility classes. For this, 34 observed landslide sites were considered (Figure 6).

The observed landslides in the very high susceptible zone were 38.2% with a landslide frequency of 0.0059, which was found to be the largest among other

susceptibility classes. The high, moderate, low and very low classes showed frequencies of 0.0035, 0.001477, 0.001471 and 0.00096 respectively. The overall validation result shows that 88.1% of the landslides have occurred in the moderate to very high susceptibility zones (Figure 7).