Landslide Susceptibility Assessment at a Part of Uttarakhand Himalaya, India using GIS – based Statistical Approach of Frequency Ratio Method

Landslide Susceptibility Assessment at a Part of Uttarakhand Himalaya, India using GIS – based Statistical Approach of Frequency Ratio Method

Binh Thai Pham

1. Department of Civil Engineering Gujarat Technological University

Nr.Visat Three Roads, Visat – Gandhinagar Highway, Chandkheda, Ahmedabad – 382424, Gujarat, India.

2. Department of Geotechnical Engineering University of Transport Technology

54 Trieu Khuc, Thanh Xuan, Ha Noi,Viet Nam.

Dieu Tien Bui

Geographic Information System Group, Department of Business Administration and Computer Science Telemark University College

Hallvard Eikas Plass 1, N-3800 Bø i Telemark, Norway.

Prakash Indra

Department of Science & Technology Bhaskarcharya Institute for Space Applications and Geo-Informatics (BISAG)

Government of Gujarat, Gandhinagar, India.

Dholakia M. B

Department of Civil Engineering, LD College of Engineering , Gujarat Technological University

Ahmedabad – 380015, Gujarat, India.

Abstract – Frequency Ratio has successfully applied as statistical approach for landslide susceptibility assessment in many regions over the world. In the present study, a part of Uttarakhand Himalaya has been selected as a case study to apply the FR model for landslide susceptibility assessment. For this, landslide inventory map was firstly constructed with 430 landslide locations identified from various sources with the help of GIS technology. These landslide locations were then randomly split into two parts (i) for training process (70% landslide locations) and (ii) for validation process (30% landslide locations). Presently, the total of six landslide conditioning factors (slope, aspect, elevation, curvature, land use, and rainfall) has been selected for analyzing the spatial relationship with landslide occurrences. Using training dataset, the FR model was then built to assess landslide susceptibility in the study area. Finally, success rate curve and predictive rate curve have been employed to validate the performance of the FR model. The results show that the FR model indicates fairly well in the present study. Overall, the FR model is an effective method for the landslide susceptibility assessment of hilly areas. It can be applied in other areas of Himalayas for the assessment and management of landslide hazards

Keywords: Landslides; GIS, Frequency Ratio, Uttarakhand, India

Landslide susceptibility map is a useful tool in landslide hazard management via land use planning and decision makings. It shows degree of susceptibility of area to landslide occurrences. Landslide susceptibility map could be produced based on the spatial prediction of landslides that is carried out on the base of an assumption that landslides in the future will occur under same conditions with which occurred in the past [2]. Therefore, landslide susceptibility could be assessed through evaluation of the spatial relationship between a set of conditioning factors and previous landslide occurrences. In recent years, many landslide susceptibility maps have been generated in many regions over the world using Geographic Information System (GIS) technology as a standard tool.

Presently, statistical approach is the most popular for landslide susceptibility assessment. It is known as subjective approach to produce reliable results. Many methods have been applied using this approach such as frequency ratio [3-5], weights of evidence [4, 6, 7], logistic regression [6, 8, 9]. Out of these methods, frequency ratio is used widely for landslide susceptibility assessment with good performance [5, 10].

1. INTRODUCTION

   Landslide is a natural geological phenomenon which is described as a massive movement of materials (soils, rocks, organics, etc.) from up to down slope [1] causing loss of life and properties. It usually occurs under different conditions depending on characteristics of study region such as geology, topography, hydrology, meteorology, vegetation, human activities, etc. Landslide is a complex phenomenon thus researchers all are trying to understand its mechanism in order to mitigate their harmful impaction.

   The main objective of the current study is to create a detailed landslide susceptibility map at a part of part of Uttarakhand Himalaya (India) using the FR model. The performance of the FR model has been evaluated using success rate curve and predictive rate curve.

2. STUDY AREA

   The study area (Fig 1) is located between Pauri Garhwal and Tehri Garhwal districts in Uttarakhand state of India (longitudes of 78o2901E to 78o3706E and latitudes 29o5638N to 30o0937N) covering an area of of

   323,815m². The study area is a mountainous region with high mountain ranges. Elevation ranges from 380m to 2180m above the standard sea level. Terrain surface of the

   study area is very steep with slope angles ranging from 0 to 70 degrees. About 85.45% of the study area belongs to slope angles of 15 to 45 degrees.

   Fig. 1 Landslide inventory map of study area

   In the study area, there are four main land use patterns such as dense forest, open-forest, non-forest, and scrub land. Non-forest occupies the biggest area (39.02%). The study area occupies by two types of soil namely silt and loamy. Loamy soil is predominant in the area (73.73%).

   The study area is situated in subtropical moon soon region having three separate seasons including winter (October to February), summer (March to June), and moon soon (June to September). Rainfall usually occurs heavily in moon soon season with annual mean rainfall ranging from 770mm to 1684mm. The temperature in this region varies from 1.3°C to 45°C whereas the humidity varies between 25% and 85%.

3. METHODOLOGY

   3.1. Data collection and interpretation

   Regional scale topographical and land use map on the scale of 1:1000.000 have been used in the present study (http://www.ahec.org.in/wfw/maps.htm). Meteorological data was studied for 30 years from 1984 to 2014 obtained from Global Weather data for SWAT [11]. Landslide susceptibility assessment has been done using GIS software

   10.2 versions.

   1. Preparation of landslide inventory map

      Landslide inventory map has been constructed with 430 landslide locations identified using interpretation of Google Earth images up to 10m spatial resolution in Google Earth pro 7.0 (Fig 1). These landslide locations have been then validated from historical landslide reports, newspaper records, and extensive field data. Landslide inventory has

      been then divided into two parts to generate training dataset (70% landslide inventory, i.e 301 landslide locations) and testing dataset (30% remaining landslide inventory, i.e 129 landslide locations)

   2. Development of various thematic layers

   Landslide conditioning factors such as slope angle, slope aspect, elevation, curvature, land use, rainfall have been taken into account to evaluate the spatial relationship between them and landslide occurrences in the study area. Slope angle map, slope aspect map, elevation map, and curvature map have been constructed using DEM with 20m generated from regional scale topographic map. Land use map has been extracted from state land use map. Rainfall map has been generated based on spline interpolation method [12] using meteorological data.. All classes of these maps are shown in Table 1. Also, Fig 2 shows the slope angle map, Fig 3 shows land use map.

   Fig. 2 Slope map with landslide locations

   Fig. 3 Land use maps with landslide locations

   1. Background of the Frequency Ratio metod

      Frequency Ratio (FR) is a statistic approach that has been applied to evaluate landslide susceptibility in this study. The main principle of this method is based on assessment of observed spatial relationship between past landslides and a set of landslide conditioning factors [13]. FR is carried out based on the frequency ratio values that are a ratio of the probability of present and absence of landslide occurrences for each landslide conditioning factor class. Higher FR value indicates stronger observed spatial relationship between the landslide occurrence and landslide conditioning factor [14]. FR values are calculated by applying the following equation:

      P Npix /N

      conditioning factor class, NL is the number of all landslide pixels in total the study area.

   2. Development of the Frequency Ratio model for landslide susceptibility assessment

      In order to assess landslide susceptibility in the study area using the FR model, landslide inventory map (70% landslide location) has been first overlaid separately with thematic data layers to calculate the frequency ratio values (FR). Thereafter, the FR values have been converted into Normalized Frequency Ratio values (NFR) in the range from 0.01 to 0.99 to facilitate the final analysis and interpretation [15]. The NFR values were then used to

      reclassify all landslide conditioning factors for landslide susceptibility analysis. The results are shown in Table 1.

      FR = i = i

      PL NLpix /NL

      (1)

      i i

      i

      Where Pi is the percentage of pixels in each landslide conditioning factor class, PLi is the percentage of landslide

      pixels in each landslide conditioning factor class.

      Npix is

      the number of pixels in each landslide conditioning factor class, N is the number of all pixels in total the study area.

      N

      Lpix i

      is the number of landslide pixels in each landslide

      Table 1 Landslide conditioning factors and its Normalized Frequency Ratio values

      Data layers	Class	Pixels	Landslide pixels	% Class Pixels	% Landslide Pixels	FR	NFR
      Slope angle (degree)	0-8	23380	0	2.89	0	0	0.01
      	8-15	51036	182	6.31	2.97	0.47	0.161
      	15-25	172508	587	21.34	9.57	0.449	0.154
      	25-35	307836	1752	38.08	28.57	0.75	0.25
      	35-45	210478	2608	26.03	42.52	1.633	0.533
      	> 45	43250	1004	5.35	16.37	3.06	0.99
      Slope aspect	Flat (-1)	2995	0	0.37	0	0.000	0.010
      	North (0-22.5 and 337.5-360)	91903	823	11.37	13.42	1.181	0.670
      	Northeast (22.5-67.5)	110190	505	13.63	8.23	0.604	0.348
      	East (67.5-112.5)	103550	403	12.81	6.57	0.513	0.297
      	Southeast (112.5-157.5)	99163	661	12.27	10.78	0.879	0.501
      	South (157.5-202.5)	102376	986	12.66	16.08	1.270	0.720
      	Southwest (202.5-247.5)	110327	1467	13.65	23.92	1.753	0.990
      	West (247.5-292.5)	93966	749	11.62	12.21	1.051	0.597
      	Northwest (292.5-337.5)	94018	539	11.63	8.79	0.756	0.433
      Elevation (m)	< 600	69962	1745	8.65	28.45	3.288	0.990
      	600 – 750	87839	804	10.86	13.11	1.207	0.370
      	750 – 900	111735	694	13.82	11.32	0.819	0.254
      	900 – 1050	120840	1042	14.95	16.99	1.137	0.349
      	1050 – 1200	119901	953	14.83	15.54	1.048	0.322
      	1200 – 1350	105343	511	13.03	8.33	0.639	0.201
      	1350 – 1500	86345	192	10.68	3.13	0.293	0.097
      	1500 – 1650	57348	121	7.09	1.97	0.278	0.093
      	1650 – 1800	34539	71	4.27	1.16	0.271	0.091
      	> 1800	14636	0	1.81	0	0.000	0.010
      Curvature	Concave (<-0.05)	368974	3572	45.64	58.24	1.276	0.990
      	Flat (-0.05 – 0.05)	71506	0	8.84	0	0.000	0.010
      	Convex (>0.05)	368008	2561	45.52	41.76	0.917	0.714
      Land use	Dense Forests	258794	1730	68.07	56.25	0.826	0.275
      	Non Forests	315891	1931	7.36	22.49	3.057	0.990
      	Open Forests	181011	1637	15.09	20.76	1.376	0.451
      	Scrub Land	53964	835	4.5	0.5	0.110	0.045
      Rainfall (mm)	< 900	68200	914	8.44	14.92	1.768	0.990
      	900 – 1000	127765	1288	15.8	21.02	1.330	0.739
      	1000 – 1100	123612	1275	15.29	20.81	1.361	0.757
      	1100 – 1200	111966	1064	13.85	17.36	1.254	0.695
      	1200 – 1300	104849	970	12.97	15.83	1.221	0.676
      	1300 – 1400	94066	486	11.63	7.93	0.682	0.368
      	1400 – 1500	81107	89	10.03	1.45	0.145	0.060
      	> 1500	96923	42	11.99	0.69	0.057	0.010

   3. Landslide Susceptibility Map

      Landslide susceptibility map has been constructed by calculating and classifying landslide susceptibility indexes (LSI) for whole study area. LSI indicates the degree of susceptibility of area to landslide occurrences. Areas with smaller LSI indicate less susceptiblity to landslide occurrence. LSI has been calculated based on the NFR values that have been determined in training process (Table 1). The calculation of LSI is shown in E.q (2):

      Many methods can be employed for classification of landslide susceptibility indexes such as the equal interval, the natural break and the standard deviation [16]. Out of these, the natural break method is the most widely used

      [17] thus it has been selected for classifying the landslide susceptibility indexes in this present study. Using this method, landslide susceptibility indexes were classified into 5 intervals with respective susceptible classes as: (1) Very low (LSI = 0.06 1.905), (2) Low (LSI = 1.905 2.481), (3) Moderate (LSI = 2.481 3.035), (4) High (LSI

      6

      LSI = NFRi

      i=1

      (1)

      = 3.035 3.703), (5) Very high (LSI = 3.703 5.94).

      Landslide susceptibility map developed using the FR model in the study area is shown in Fig. 4.

      Where NFRi are the normalized frequency ratio values of slope, aspect, elevation, curvature, land use, and rainfall, respectively

      Fig. 4 Landslide susceptibility map (LSM) of the study area using the FR model

   4. Validation of the Frequency Ratio Model

   The performance of the FR model has been evaluated using the success rate and predictive curves which were proposed by Chung and Fabbri [18]. Success rate curve indicates the relationship between the percentage of landslide susceptibility map and the percentage of landslide pixels used for training process. In contrast, predictive rate curve presents the relationship between the percentage of landslide susceptibility map and the percentage of landslide pixels employed for testing process. The area under success rate curve (AUC) illustrates the degree of fit of the

   Frequency Ratio model with the training dataset whereas the area under predictive rate curve (AUC) shows prediction capability of the Frequency Ratio model [18]. Higher AUC values indicate better performance of the FR model.

   The results are shown in Fig. 5. It can be observed that the AUC of success-rate curve is 0.75 indicating quite good degree of fit of the Frequency Ratio model with the training dataset. Whereas, the AUC value of prediction rate curve is

   0.70 indicating that prediction ability of the Frequency Ratio model are also fairly good.

   Success rate curve AUC = 0.75

   Predictive rate curve AUC = 0.72

   Fig. 5 The performance of the FR model using success rate curve and predictive curve in this study

4. DISCUSSIONS AND CONCLUSIONS

Landslide susceptibility assessment at a part of Uttarakhand Himalaya, India has been carried out in this study using the Frequency Ratio (FR) model which has been applied widely in literatures. A total of 236 landslide locations have been utilized to construct landslide inventory map. Six landslide conditioning factors (slope angle, slope aspect, elevation, curvature, land use, rainfall) have been taken into consideration for evaluation of the spatial relationship between them and landslide occurrences. The performance of the FR model has been validated using success rate and predictive rate curves. The results show that the FR model is applicable for landslide susceptibility assessment. Its performance is fairly good (AUC = 0.72). The results of the present study are comparable with other studies [5, 19, 20].

Overall, the FR model is an effective method for landslide susceptibility assessment of hilly and mountainous areas. It can be applied in other landslide prone areas for assessment and management of landslide hazards.

ACKNOWLEDGEMENT

Authors are also thankful to Mr. T.P Singh, Director, Bhaskarcharya Institute for Space Applications and Geo- Informatics (BISAG), Gujarat, India for providing facilities to carry out this research work.

REFERENCES

1. Van Westen, C., T. Asch, and R. Soeters, Landslide hazard and risk zonation-why is it still so difficult. Bull Eng Geol Environ 2006. 65: p. 67184.

2. Pradhan, B., A comparative study on the predictive ability of the decision tree, support vector machine and neuro-fuzzy models in landslide susceptibility mapping using GIS. Computers & Geosciences, 2013. 51(0): p. 350-365.

3. Mohammady, M., H.R. Pourghasemi, and B. Pradhan, Landslide susceptibility mapping at Golestan Province, Iran: a comparison between frequency ratio, DempsterShafer, and weights-of- evidence models. Journal of Asian Earth Sciences, 2012. 61: p. 221- 236.

4. Ozdemir, A. and T. Altural, A comparative study of frequency ratio, weights of evidence and logistic regression methods for landslide susceptibility mapping: Sultan Mountains, SW Turkey. Journal of Asian Earth Sciences, 2013. 64: p. 180-197.

5. Poudyal, C.P., et al., Landslide susceptibility maps comparing frequency ratio and artificial neural networks: a case study from the

   Nepal Himalaya. Environmental Earth Sciences, 2010. 61(5): p. 1049-1064.

6. Devkota, K.C., et al., Landslide susceptibility mapping using certainty factor, index of entropy and logistic regression models in GIS and their comparison at MuglingNarayanghat road section in Nepal Himalaya. Natural Hazards, 2013. 65(1): p. 135-165.

7. Kayastha, P., M.R. Dhital, and F. De Smedt, Landslide susceptibility mapping using the weight of evidence method in the Tinau watershed, Nepal. Natural hazards, 2012. 63(2): p. 479-498.

8. Bai, S., et al., GIS-based rare events logistic regression for landslide-susceptibility mapping of Lianyungang, China. Environmental Earth Sciences, 2011. 62(1): p. 139-149.

9. Yilmaz, I., Comparison of landslide susceptibility mapping methodologies for Koyulhisar, Turkey: conditional probability, logistic regression, artificial neural networks, and support vector machine. Environmental Earth Sciences, 2010. 61(4): p. 821-836.

10. Shahabi, H., et al., Landslide susceptibility mapping at central Zab basin, Iran: A comparison between analytical hierarchy process, frequency ratio and logistic regression models. CATENA, 2014. 115(0): p. 55-70.

11. NCEP, Global Weather data for SWAT. http://globalweather.tamu.edu/home, 2014.

12. Kawamura, H., T. Sasaki, and T. Otsuki, Spline interpolation method. 1992, Google Patents.

13. Choi, J., et al., Combining landslide susceptibility maps obtained from frequency ratio, logistic regression, and artificial neural network models using ASTER images and GIS. Engineering Geology, 2012. 124(0): p. 12-23.

14. Pradhan, B. and S. Lee, Regional landslide susceptibility analysis using back-propagation neural network model at Cameron Highland, Malaysia. Landslides, 2010. 7(1): p. 13-30.

15. Umar, Z., et al., Earthquake induced landslide susceptibility mapping using an integrated ensemble frequency ratio and logistic regression modelsin West Sumatera Province, Indonesia. CATENA, 2014. 118(0): p. 124-135.

16. Ayalew, L. and H. Yamagishi, The application of GIS-based logistic regression for landslide susceptibility mapping in the Kakuda-Yahiko Mountains, Central Japan. Geomorphology, 2005. 65(1): p. 15-31.

17. Irigaray, C., et al., Evaluation and validation of landslide- susceptibility maps obtained by a GIS matrix method: examples from the Betic Cordillera (southern Spain). Natural hazards, 2007. 41(1): p. 61-79.

18. Chung, C.-J.F. and A.G. Fabbri, Validation of spatial prediction models for landslide hazard mapping. Natural Hazards, 2003. 30(3): p. 451-472.

19. Choi, J., et al., Combining landslide susceptibility maps obtained from frequency ratio, logistic regression, and artificial neural network models using ASTER images and GIS. Engineering Geology, 2012. 124: p. 12-23.

20. Yalcin, A., et al., A GIS-based comparative study of frequency ratio, analytical hierarchy process, bivariate statistics and logistics regression methods for landslide susceptibility mapping in Trabzon, NE Turkey. Catena, 2011. 85(3): p. 274-287.