Glacial lake changes and outburst flood hazard in Chandra basin, North-Western Indian Himalaya

ABSTRACT Climatic change-induced glacier recession has been accompanied by formation and growth of proglacial lakes in the Himalayan region, which pose an emerging significant threat to the downstream communities/settlements in the form of outburst floods. To understand spatiotemporal evolution patterns, sources and driving mechanism of formation and expansion of glacial lakes, a temporal inventory of glacial lakes (area > 2000 m2) in Chandra basin has been developed from 2000 to 2014 using IRS LISS-III images. From 2000 to 2014, the total number of glacial lakes in Chandra basin increased from 28 to 46 and area expanded from 1.91 ± 0.24 km2 to 3.26 ± 0.24 km2. Glacier recession and increased glacier melt runoff due to climate warming led to the formation and expansion of glacial lakes in space vacated by glacier recession. The increase in number and area of ice-dammed lakes at higher elevations confirms the continued glacier retreat in the basin. Lakes in contact or in the proximity of the mother glacier exhibit higher growth and formation rate. The accelerated growth of glacial lakes has resulted in increased hazard and damage potential of glacial lake outburst floods in Chandra basin. Seven potentially dangerous lakes are identified and analysed qualitatively for outburst probability.


Introduction
Global climate is changing due to various natural processes/phenomena and human activities. High mountain glacial environment and ecosystems with snow, glaciers and permafrost are very sensitive to such climatic changes, thus quickly affected as evidenced by the ongoing worldwide accelerated glacier retreat over the past few decades (Zemp et al. 2008). Climate change-induced glacier recession have triggered the dynamic evolution of glacial lakes in high-mountain areas worldwide, leading to the formation, growth and disappearance of different types of glacial lakes (Frey et al. 2010;Mergili et al. 2013;Wang et al. 2013;Emmer et al. 2015;. Some of these lakes are hazardous and pose a threat to downstream communities and infrastructure because of their potential to outburst and drain suddenly to cause rapid and highly devastating glacial lake outburst floods (GLOFs) (Richardson and Reynolds 2000).
GLOFs have emerged as a serious hazard in the mountain region in recent decades due to increased human settlements, anthropogenic and other developmental activities into areas which were inhabited and were not developed previously (Khanal et al. 2015;Nie et al. 2017). With continued global warming and glacier recession, the frequency and damage potential of GLOFs is anticipated to increase significantly in future (Richardson and Reynolds 2000;Wang et al. 2013). GLOFs evolve as a consequence of series of different processes, for example, mass movement into lakes, glacier/ice front calving into lake, progressive enlargement of lake, rising lake levels leading to overflow, mechanical rupture/failure of dam, hydrostatic failure, degradation of dam or melting of ice cores in dam, earthquakes, a flood wave from lake located upstream and intensive rainfall or snowmelt (Costa and Schuster 1988;Clague and Evans 2000;Richardson and Reynolds 2000;Emmer and Cochachin 2013;Clague and O'Connor 2015).
During last four decades a warming of about 0.15-0.60 C per decade has been observed in the different parts of the Himalaya (Shrestha et al. 1999;Dash et al. 2007;B. Wang et al. 2008;Bhutiyani et al. 2010;Khattak et al. 2011) accelerating the mass loss recession rate of Himalayan glaciers in recent decades (Bolch et al. 2012). Concomitant with the glacier recession, retreating at 10-15 m/year (WWF Nepal Program 2005), glacial lakes formed and developed on/beneath or in front of the glacier leading to an increased GLOF hazard in the Himalaya (Richardson and Reynolds 2000;Quincey et al. 2007;Worni et al. 2013;Shijin and Tao 2014). There are approximately 9000 glacial lakes and about 200 potentially hazardous glacial lakes in the Himalayan region resulting in about 40 GLOF events in over the past four decades (Yamada and Sharma 1993;Ives et al. 2010). Outburst flood from Chog Kumdan Ice-dammed lake in Shyok river basin of Jammu Kashmir, Indian Himalaya, affected 48 villages and land up to a distance of 1000-km downstream in 1929(Gunn 1930. GLOFs from the Cirenamco (1981) in Tibetan Himalaya, Dig Tsho (1985) and Tampokhari (1998) in Nepal Himalayas, Luggey Tsho (1994 in Bhutan and Chhorabari (2013) in Indian Himalaya resulted in considerable loss of life, property and infrastructure in the downstream area (Vuichard and Zimmermann 1987;Xu 1988;Yamada and Sharma 1993;Watanbe and Rothacher 1996;Allen et al. 2015). Similar devastating GLOF events have claimed thousands of lives and caused considerable damage in different regions of the world, including the European Alps, the Andes, the Canadian Cordillera, and the Central Himalaya (Clague and Evans 2000;Huggel et al. 2003;Emmer and Cochachin 2013).
As the occurrence of GLOFs and related processes are rare, often singular events, the location, timing, magnitude and probable impact area of such hazard events often hard or impossible to predict (Gruber and Mergili 2013). Even the terrain conditions, governing processes and mechanisms of GLOFs are not well understood due to remote and sometimes inaccessible locations. Hence, it is essential to identify the possible source, location and probable impact area of GLOF hazard at a regional scale in order to prioritize potentially vulnerable areas for risk management and mitigation (Huggel et al. 2002). Several studies have demonstrated the suitability of remote sensing and geomatics-based techniques for the study of glaciers and glacial lakes for broader areas, often inaccessible, on a regional scale due to readily available spatially and spectrally varied temporal data (Huggel et al. 2003;Bolch et al. 2011;Zhang et al. 2015;Cook et al. 2016;. Understanding of climate change impact and induced glacial lake changes is crucial for the evaluation of water resources (Fang et al. 2016), assessment of associated hazard potential (Huggel et al. 2002) and prediction of future spatio-temporal evolution of glacial lakes (Frey et al. 2010). Many scientists and researchers have prepared inventory of glacial lakes in Indian Himalayan region using remote sensing techniques (Govindha Raj 2010;Worni et al. 2013;Chander Prakash & Nagarajan 2017). There are few researches on the regional spatio-temporal evolution, differences and heterogeneity of glacial lakes; hence, we still lack an understanding about impact of climate change and glacier recession on glacier lake dynamics in Indian Himalayan region. Thus, there is a need to analyse the spatio-temporal glacial lake changes in the Indian Himalaya to understand the impact of climate change on glaciers and glacier lakes for future water resource management and potential hazard assessment and risk management.
The aim of the present study is to use remote sensing data and apply geomatics-based approach to up-to-date the knowledge on glacial lakes and analyse their distribution and temporal evolution/ development in Chandra basin located in north-western Indian Himalaya. A further aim is to identify potentially dangerous glacial lake (PDGL) and prioritize the same for detailed field investigation. A multi-temporal glacial lake inventory is prepared using remotely sensed satellite data with attribute information about location, characteristics, development pattern and surrounding conditions of the glacial lake. The information/result shall serve as a baseline for GLOF hazard assessment in Chandra basin.

Study area
The study area for the present study is Chandra Basin a located between 32 05 0 N to 32 45 0 N latitude and 76 50 0 E to 77 50 0 E longitude in Lahul Spiti District of Himachal Pradesh, India ( Figure 1). Chandra Basin is a subbasin of Chenab river basin which is located on the southern slopes of Great Himalayan and the northern slopes of Pir Panjal range of Himalayas.
The region falls in the monsoon-arid transition zone with the wet climate in the south and dry climate to its north and is characterized by low rainfall and severe winters. The aridity increase from south to north in Chandra basin indicated by a much less forested area dominated by shrubs, grassland, and alpine plants with decreasing cover northwards (Singh 1987;Owen and others 1997). The climate is influenced by western disturbances and extra-tropical cyclone that occurs during October and May and southwest monsoon from July to September. The Annual average precipitation and temperature according to monthly data for 1901-2001 are 63.9 mm and 8.9 C, respectively. Annual rainfall varies from about 100 to 400 mm, and snowfall from less than 1 to 6 m or higher at higher altitudes. There are 355 glaciers covering 736 km 2 area in the study area as per Randolph Glacier Inventory 5.0 (Arendt and others 2015). Glaciers in the basin are melting rapidly resulting in the formation of new lakes and expansion of existing lakes (Kulkarni et al. 2006;Pandey and Venkataraman 2013;Allen et al. 2016;Chander Prakash & Nagarajan 2017). Approximately 30 lakes larger than 10,000 m 2 exist in the Chandra-Bhaga basin (Randhawa and Sharma 2014;C Prakash & Nagarajan 2017), where planned developments include major hydroelectric power projects, widening of the existing road network, construction of new roads and railways for better connectivity to the remote villages of Lahul and Spiti district of Himachal Pradesh and the Leh-Ladakh region of Jammu and Kashmir. Many villages and small towns are located alongside streams and rivers that could be affected by GLOFs originating upstream in the Chandra basin.

Material and methods
3.1. Data source Linear Imaging Self Scanning (LISS)-III sensor multispectral images of 24-m resolution from Indian Remote Sensing (IRS) satellites IRS-1C, Resourcesat-1(IRS-P6) and Resourcesat-2 (IRS-R2) for the years 2000, 2004, 2007, 2011, 2013, and 2014 procured from National Remote Sensing Centre (NRSC) Hyderabad, India were used for mapping and classification of glacial lakes. All the IRS images selected for glacial lake delineation were from last week of August to the second week of October. During this period, there is minimum or no perennial snow coverage, availability of cloud free data as monsoon season is over and lakes have a larger area following snow/glacier melt monsoon precipitation-induced runoff. Hence, it is suitable for identification and mapping of glacial lake/glaciers. Global Digital Elevation Model (GDEM) version 2 of Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) downloaded from US Geological Survey (http:// earthexplorer.usgs.gov) is used to derive the topographical data and information such as slope, elevation, extraction of the catchment, and identifying shadowed region using hillshade, etc., in the Chandra basin. High-resolution Google Earth images were used as auxiliary data for the classification of the glacial lake and assessing lake surrounding condition through visual interpretation. These images were mainly from GeoEye and SPOT-5 with a spatial resolution of 1.65 m and 2.5 m, respectively.

Conversion of DNs to top-of-atmospheric (TOA) reflectance
The multispectral LISS-III images are already corrected for radiometric and geometric correction by National Remote Sensing Centre (NRSC) but need to be corrected for atmospheric effects. Atmospheric corrections are necessary to remove effects of scattering and transmittance. After the atmospheric corrections, image data are composed of ground reflectance only which helps in identification of water pixels more accurately and differentiating the same from shadowed area having similar spectral characteristics. Atmospheric corrections involved two steps: (1) converting digital numbers (DNs) into at-sensor radiance or top-of-atmospheric (TOA) radiance and (2) to convert the apparent at-sensor or TOA radiance to surface reflectance.
DNs of each image band are converted to at-sensor spectral or TOA radiance (L l ) and can be computed as follows (Lillesand et al. 2004): where L l is spectral radiance at-sensor, L max and L min are scaled spectral radiance (provided in the header file) in Wm-2 sr ¡1 mm ¡1 , and Q calmax is maximum possible DN value. The above formula is also expressed as follows: where Gain can be computed using L max and L min and Bias is equal to L min for respective bands. The At-sensor or TOA radiance can be further converted to surface reflectance which involves the correction for solar angle and atmospheric effects using the following formula from Landsat 7 Science Data Users Handbook and Govindha : where r λ is planetary reflectance in a particular band, d is Earth-Sun distance in astronomical units (AU), E sunλ is solar exo-atmospheric spectral irradiance (Wm ¡2 mm ¡1 ), and u s is solar zenith angle (degree). The values required for converting DNs to radiance and radiance to reflectance for each band w.r.t. Satellites are obtained from the header file of the image and tabulated in Table 1.
A model was prepared in ERDAS Imagine's model maker for the above calculations and LISS-III multispectral images were converted to TOA surface reflectance from DNs for further image processing and analysis for extracting details about glacial lakes and its surrounding features.

Glacial lake mapping and inventrorization
A glacial lake inventory is a prerequisite for examining the spatial distribution and temporal evolution of glacial lakes which are necessary to identify potentially hazardous lakes. Glacial lakes were detected and lake boundaries extracted from remote sensing images by automated image processing methods complemented by visual image interpretation and Google Earth high-resolution images. Here, we applied a semiautomatic three-step-based approach to develop the glacial lake inventory ( Figure 2). In the first step, glacial lake pixels were automatically mapped by first computing the normalized difference water index (NDWI) and then applying manually chosen appropriate thresholds for image segmentation and extracting water pixels as proposed by Li and Sheng (2012). Based on the principal of high reflection of water bodies in the visible spectrum (maximum in green wavelength) and strong absorption in near-infrared wavelength, NDWI was calculated using the TOA reflectance images of the green band (Band 2) and NIR band (Band 4) IRS LISS-III images (Huggel et al. 2002) NDWI ¼ Green À NIR Green þ NIR : In the second step, mountain shadows misclassified as glacial lakes due to the similar spectral characteristic to water bodies are eliminated using the DEM-based terrain analysis. A shadow mask developed using sun azimuth angle and sun elevation information of image data in ArcGIS is used to remove any false lake due to mountain shadow. A slope threshold was applied to extract the boundaries of glacial lakes. A water pixel with a slope less than 5 was assumed to be a water pixels as water surface slope is theoretically much smaller than mountain shadow and glacial lakes mostly develop on 2 -6 surface gradients (Reynolds 2000).
In the third step, manual examination and correction of automatically extracted lakes were carried out using with false colour composites, high-resolution Google Earth imageries. ArcGIS was synchronized with Google Earth for interpreting, examining and manual improvements in the extraction of lake boundaries. Synchronization was necessary as it assured relatively accurate information about lake dam type, glacier lake distance and lake drainage type. For excluding glacier lakes beyond 10 km from glacier terminus, Randolph Glacier Inventory (RGI v 5.0) was used with a buffer polygon of glaciers in the study area with a buffer distance 10 km (Wang et al. 2013;Zhang et al. 2015). Finally, lakes with area greater than 2000 m 2 and within 10 km from nearest glacier terminus were catalogued in the glacial lake inventory with following set of attributes assigned to each lake: spatial location of the lake in latitude and longitude and elevation (z) in m, the area of the lake in m 2 ; lake dam type: the lakes were classified into three classes based on dam type namely: (i) icedammed which are either impounded by a glacier or embedded on glacier surface as supra glacier pond/lake, (ii) moraine-dammed lakes impounded by either recessional terminal moraine or lateral moraine, (iii) bedrock-dammed lakes impounded by a solid rock mass or lakes without any clear dam structure in the flat terrain of glacier fore-fields (Worni et al. 2013;Emmer et al. 2015); glacier lake distance in m; lake drainage type: divided into (i) surface drainage, and (ii) closed lakes, based on recognizable surface or no surface outflow assessed mainly from Google Earth.
The accuracy of glacial lakes area and extracted related information depends on cloud and snow coverage, pixel resolution, image quality, pre-processing/processing of the image, the experience of expert and others (Hall et al. 2003). The uncertainty and errors in glacial lake boundary extraction and area computation using satellite data are difficult to evaluate quantitatively due to lack of ground-based field data. The errors induced by spatial resolution and pre-processing/processing of data are systematic and have limited impact and are not important for spatiotemporal changes of lakes at a regional scale ). The error due to seasonal discrepancies was minimized as the images were obtained for September and October month only.
In the present study, we estimated the uncertainty in mapping glacial lake area following the Hanshaw and Bookhagen (2014) approach based on the assumption that errors in mapping are Gaussian distributed, i.e. 68% (1s) of the mapped pixels are subject to errors. The uncertainty in lake area is computed as follows: where P is the perimeter of mapped lake and G is the grid cell size. Potentially half of an individual pixel is included or excluded during vectorization as lake boundaries were mapped along pixel diagonals and shoreline of the lakes passed through the centre of the pixel. Hence, only half of the single pixel area is considered for error computation.

Outburst probability assessment
A remote-sensing-based qualitative approach was used to identify PDGLs and assess the outburst probability using available Google Earth high-resolution imagery. The lakes were categorized as potentially dangerous according to (i) their size (>0.1 km 2 ) and potential to grow in future (increasing area) (Bolch et al. 2011;Iribarren Anacona et al. 2014), (ii) lake water source (glacier fed lakes) (O'Connor et al. 2001;Wang et al. 2013), and (iii) distance from nearest glacier terminus (1 km) (L€ u et al. 1999;. Lakes with the steep surrounding terrain, in direct contact or proximity of glacier with potential glacier calving, are considered highly susceptible to outburst Iribarren Anacona et al. 2014). For morainedammed lakes, unstable dam geometry, lower freeboard, and steep distal face slope are additional characteristics which increase the probability of outburst of such lakes (Huggel et al. 2004;Wang et al. 2012). Hence, the identified lakes were further assessed qualitatively in detail for outburst probability by evaluating five key indicators: (i) lake and glacier characteristics, (ii) dam type, (iii) dam geometry, (iv) freeboard, and (v) the potential for lake impacts. The five key indicators were selected based on three criteria. First, the indicators have been used worldwide for assessing glacial lake outburst susceptibility (Huggel et al. 2004;McKillop and Clague 2007;Bolch et al. 2011;Worni et al. 2013;Emmer and Vil ımek 2014). Second, the five key indicators selected could be measured and interpreted using readily available remote sensing satellite data. Third, the indicators data, being continuous and nominal in nature, could be utilized as qualitatively or semi-quantitatively for assessing potential outburst probability. Hence, it was possible to assign each indicator one attribute out of three lake outburst probabilities i.e. low, medium and high. The key indicators and different parameters used to assess the outburst probability qualitatively are summarized based on the previous studies (Huggel et al. 2004;Bolch et al. 2011;Wang et al. 2013;Worni et al. 2013). The general decision approach used for final evaluation of outburst probability of glacial lakes is shown in (Figure 2).

Glacial lakes inventory
A total of 46 lakes with area greater than 0.002 km 2 were detected and inventoried in Chandra basin using IRS LISS-III 2014 images. The overall area of glacial lakes in 2014 was 3.26 § 0.33 km 2 . The area of mapped glacial lakes in Chandra basin ranges from 0.002 km 2 to 1.26 km 2 . The majority of the lakes in Chandra basin are small (area < 0.01 km 2 ), comprising 29 lakes (63%) of the total. 12 lakes are medium-sized (0.01 area < 0.1 km 2 ), 4 lakes are large-sized (0.1 area < 1.0 km 2 ) and one is very large sized (area 1.0 km 2 ) glacial lake (Figure 3). Though the large and very large sized lakes are less in number, these account for 2.8 km 2 (86%) of the total glacial lake area in the basin (Table 2). There are total 35 (76%) proglacial lakes (i.e. within 500 m of glacier terminus) in the study area. The majority of glacier lakes are glacier fed (82%). Non-glacier fed lakes are only 8 (18%) and are sparsely distributed. We found 23 lakes (50%) are very close or in direct contact with the glacier, while 13 (28%) are located in the fore-fields of the retreating glacier (100-500 m from glacier terminus), and 10 (22%) beyond 500 m distance from glacier terminus. Of the 46 lakes, 20 (43%) are bedrock-dammed lakes, 16 (35%) are ice-dammed, and 10 (22%) are moraine-dammed lakes covering area 2.5%, 20%, and 77.5%, respectively. Moraine-dammed lakes (average area 0.25 km 2 ) in the basin are larger as compared to bedrock-dammed lakes (average area 0.032 km 2 ) and ice-dammed lakes (average area 0.008 km 2 ). Surface drainage/outflow is clearly observed for 61% lakes, with highest for bedrock-dammed lakes (85%) and lowest for ice-dammed lakes (25%) ( Table 3).
Glacial lakes are situated within elevation range from 3020 to 5420 m with an average elevation of 4735 m. The majority of the lakes (46%) are located at 5000-5500 m elevation zone, 24% are at 4500-5000 m, 21% at 4000-4500 m, and only 9% at 3000-4000 m (Figure 4a). In general, lakes dominate at the higher elevations and are underrepresented at lower elevations. Moraine-dammed lakes are located at medium elevation with elevation varying from 4070 -5245 m (average elevation 4602 m), bedrock-dammed lakes with average elevation 4777 m (3020-5370 m) and ice-dammed lakes having average elevation 4716 m (4120-4370) dominate at the higher elevations (Table 3). The majority of the lakes are in the proximity of the glacier. About 76% lakes are within 500-m distance from the glacier, 56.5% within 100-m distance and 19.5% between 100 and 500 m from the glacier (Figure 4b). These observations about lakes w.r.t elevation and distance from glacier confirm the ongoing retreat of glaciers in the basin, as modern glaciers are at higher elevations, and lakes are emerging and developing in the forefields of glaciers vacated due to glacier recession.

Glacial lakes changes 2000-2014
Evolution and development process of glacial lakes in Chandra basin was very complex, consisting of new emerging lakes, growth and disappearance of existing glacial lakes from 2000 to 2014. The number and area of glacial lakes increased significantly from 2000 to 2014 in Chandra basin. In total, 18 new glacier fed lakes appeared in the basin, exhibiting 64% increase in the total number of lakes from 28 lakes in 2000-46 lakes in 2014. The total area rose by 70% from 1.91 § 0.24 km 2 in 2000-3.26 § 0.24 km 2 in 2014. The rate of growth is 1.29 a ¡1 (4.6% a ¡1 ) and 0.095 km 2 a ¡1 (5% km 2 a ¡1 ), respectively, for number and area of lakes. The rate of formation and disappearance of the ice-dammed lake was higher as compared to other types of lakes. The number of ice-dammed lakes increased at 0.72 lakes per year (12% a ¡1 ) from 6 in 2000 to 16 in 2014, while bedrock-dammed lakes grew at 0.57 lakes per year (4.8% a ¡1 ) from 12 to 20. There was no increase in the number of moraine-dammed lakes for the same period. The areal expansion was highest for moraine-dammed lakes from 1.37 § 0.17 km 2 in 2000-2.51 § 0.19 km 2 in 2014 with an average rate of growth 0.082 km 2 a ¡1 (6.3% a ¡1 ) from 2000 to 2014 (Table 4).
Based on the observation of evolution patterns and processes of lakes from 2000 to 2014 lakes are grouped into three categories: emerging lakes (newly formed lakes), growing lakes (significant increase in area) and constant lakes (lakes without any major change in the area). Some ice-dammed lakes and bedrock-dammed lakes emerged from 2000-2014, while no new moraine-dammed lake appeared for the period. The majority of growing lakes in direct contact with the mother glacier are  expanding towards the direction of the glacier, in the space provided by glacier retreat. Bedrockdammed lakes and moraine-dammed lakes beyond 500-m distance from glacier are constant lakes as there is no significant change in the area of such lakes. Hence, lakes with closer hydrological connection with glacier are expanding more rapidly as compared to the lakes with remote or no hydrological connection.

Potentially dangerous glacial lake
Based on preliminary assessment seven lakes were identified as PDGLs in the Chandra basin in 2014, with total area 2.9 § 0.21 km 2 . Lake and surrounding characteristics of these seven lakes are tabulated in Table 5 for qualitative assessment of outburst probability. Among the seven identified PDGLs, two lakes expanded moderately by 64% and 39%, two expanded only by 23% and 16% while for three lakes the expansion was insignificant from 2004 to 2014 (decadal). Three lakes were in direct contact to the glacier with potential glacier calving. Based on the outburst probability and   prioritization scheme outlined in Figure 3, two lakes are classified as having high-outburst probability, one with medium and four with low-outburst probability. The temporal growth of lakes with high-outburst probability is shown in Figure 5.

Discussion
The global climatic change influences regional climate which in turns has a major impact on the glaciers. Glacier dynamics are closely related to the evolution and development of glacial lakes. The retreat of glaciers reveals further basins resulting in the formation of new pro-glacial lakes and increase in the size of existing pro-glacial lakes worldwide (Carrivick and Tweed 2013;Hanshaw and Bookhagen 2014;Zhang et al. 2015;Cook et al. 2016;Govindha Raj and Kumar 2016). Climatic variations like increase in temperature, precipitation, and evaporation in the glaciated region affect the mass balance of glaciers and results in enhanced snow and glacier melt runoff leading to expansion of glacial lakes due to increased water supplies to the lake. Recent researches indicate increasing trends in mean temperature and glacier mass balance while a decreasing trend in precipitation in north-western Himalaya. An overall warming of 1.6 C has been observed in the western Himalaya during last century (Bhutiyani et al. 2007) The Pir-Panjal and Greater Himalayan range, bounding the Chandra basin on south and north, show an increase of 0.8 C and 1 C in maximum temperature and 0.6 C and 3.4 C minimum temperature during the past two decades, respectively, (Shekhar et al. 2010). Winter snowfall shows a decreasing trend in the region with a warming climate (Bhutiyani et al. 2010;Dimri and Dash 2012) and a decrease of 280 cm in Pir Panjal and 440 cm in Greater Himalayan range has been observed from 1988 to 2007 (Shekhar et al. 2010). Concomitant with the warming and decrease in snowfall, the glacier in Chandra basin have been retreating significantly at an average rate of about 0.1-0.5% per annum (Kulkarni et al. 2011;Gardelle et al. 2013;Pandey and Venkataraman 2013). As 82% lakes in the basin are glacier-fed and 62% have surface drainage. Lakes in contact with the glaciers in the basin have highest expansion rate indicating glacial meltwater as the dominant source of water for lake expansion. The excess flow of water into Lake due to precipitation overflows through the surface drainage, thus, can be ignored. Hence, glacier retreat in the basin is the main source of water and extra space for the formation and expansion of glacial lakes in the basin.
With the ongoing climate change and continued warming in future, the glacier ablation rate will accelerate leading to formation and expansion of glacier lakes in Chandra basin. Additionally, glacial lakes in contact with glaciers strongly influence the glacier dynamics. Such lakes modify the stress regime of the glacier ice, increase mass loss through glacier calving, and enhance the melting of glaciers in contact through warm water circulation and transmittance of thermal energy (Basnett et al. 2013;Gardelle et al. 2013). Hence, lakes in contact with glacier will further accelerate the glacier retreat by thermal feedback mechanism and result in expansion of glacial lakes in the basin.
Ice-dammed lakes or supraglacial lakes represent the initial phase of glacial lake evolution (Quincey et al. 2007;Sakai 2012) and are more dynamic in nature due to a strong influence of climatic factors (e.g. temperature, precipitation, etc.,) and non-climatic factors (e.g. debris cover, glacier surface slope, speed of glacier and condition of glacier) . The size, shape, and locations of such lakes change rapidly as these are unstable and temporary which may survive from a few months to several years (Wang et al. 2012). Such lakes accelerate the ablation of glacier ice by draining water through hydro-fracture/cracks to subglacial channels accelerating the melting and disintegration of ice at glacier bottom (Watanabe et al. 1995;Reynolds 2000;Benn et al. 2001). The change in ice-dammed lakes in the Chandra basin shows characteristics of disappearance and formation of new lakes simultaneously. Overall there is an increase in the number and area of ice-dammed lakes from 2000 to 2014. Highest changes in ice-dammed lakes are on Bara Shigri glacier in the basin, largest glacier in the basin (28 km long, 3 km wide with 131 km 2 area). The thick debris cover and relatively flat slope of <4 0 of Bara Shigri glacier represent the most conducive features for the formation of supraglacial lakes. About 60 lakes of size >500 m 2 were detected by Schauwecker et al. (2015) in 2014 using highresolution SPOT images, which is in accordance with the observation of present study ( Figure 6).
Bedrock-dammed lakes in the Chandra basin are expanding at a relatively lower rate of 1.8% per annum, and most of the lakes remain identical in extent and size from 2000 to 2014 (Figure 7). These lakes develop either through disconnection of ice-dammed lakes from a glacier or via collection of meltwater in the glacier eroded over-deepened parts of the bed covered previously by glaciers. Emmer et al. (2015) recognized three phases of bedrock-dammed lake evolution in Austrian Alps. Phase I is the initial phase of lake formation from the first appearance (in contact with glacier) to complete evolution, characterized by shorter time spans and high-outburst hazard probability. Phase II is characterized by continued glacier retreat, stable, constant, and decreased outburst probability, while Phase III is a non-glacial-fed lake developed over a longer span of time and are usually constant in volume with very low outburst probability and hazard. The bedrock-dammed lakes in Chandra basin show similar evolution pattern as new bedrock-dammed lakes emerged at higher elevations and are in proximity of the glacier. Lakes with no change in areal extent are either non-glacial or located at a distance from the glacier.
Though less in number, Moraine-dammed lakes constitute 77% of the total glacial lake area in Chandra basin. Moraine-dammed lakes in contact with the glacier are expanding rapidly in accordance with the glacier recession ( Figure 5), while detached moraine-dammed lakes exhibit a smaller rate of growth in the basin. The absence of formation of new moraine-dammed lakes indicates the movement of glacier terminus to higher elevations and steeper slopes as a consequence of glacier retreat, a condition unfavourable for the development of moraine-dammed lakes.
The analysis of temporal glacial lake inventory reveals that rate of formation of supraglacial lake and rate of expansion of proglacial lake have increased significantly from 2002 to 2014. The analysis and inferences about formation and expansion of glacial lakes are concomitant with other studies in the Himalayan region (Zhang et al. 2015;Govindha Raj and Kumar 2016;Song et al. 2016;Nie et al. 2017). Rapid expansion of proglacial moraine-dammed lakes is a well observed phenomenon throughout the Hindu-Kush-Himalayan region contributing most of the increased glacial lake area Zhang et al. 2015;Govindha Raj and Kumar 2016;Song et al. 2016). Most of the bedrock-dammed lakes are stable growing at lesser rate as compared to moraine-dammed lakes and  Ice-dammed lakes. Moraine-dammed existed as perennial lakes with rapid expansion rate. The icedammed lakes or supra-glacier lakes/ponds are highly variable in terms of formation, spatial location, size and growth rate in the study area. The characteristics exhibited by the different types of glacial lakes in Chandra basin are consistent with the various other studies in Himalayan region (Benn et al. 2001;Nie et al. 2017).
The methods presented in the study enable rapid development of a database of glacial lakes, identification of PDGLs, and qualitative assessment of outburst probability of critical glacial lakes in remote, inaccessible mountain regions using remote sensing data. As the identification of PDGLs, evaluation of outburst probabilities and associated hazard assessment is a systematic process for which detailed data and knowledge about climatic conditions, glacier and lake dynamics, lake surrounding conditions, dam geometry and material properties, trigger events and downstream valley characteristics are required (Richardson and Reynolds 2000;McKillop and Clague 2007;Bolch et al. 2011;Wang et al. 2013;Emmer and Vil ımek 2014;Chander Prakash & Nagarajan 2017). Hence, the present approach is efficient for only first order evaluation of outburst probability of glacial lakes using selected physical indicators derived from remote sensing data at the regional scale.
Based on our glacial lake inventory of 2014, we carried out a preliminary assessment of lakes for potentially damaging outburst flood hazard. A total of seven PDGLs were identified and evaluated qualitatively for first order estimates of outburst probability based on selected indicators.
The number and area of glacial lakes increased in the Chandra basin by 64% and 70% from 2000 to 2014, respectively. The rate of formation of new lakes was highest for ice-dammed glacial lakes followed by bedrock-dammed lakes with averages 0.72 lakes per year and 0.57 lakes per year, respectively. The moraine-dammed lakes with average growth rate of 0.082 km 2 per year expanded more rapidly as compared to other types of lakes. The formation and expansion of glacial lake were in concert with the climatic change (increase in temperature and decrease in precipitation) induced glacier retreat in the basin. The rate of growth of glacial lakes varied inversely with the proximity to the parent glacier, closer the glacial lake to mother glacier higher was its growth rate. Hence, glacier meltwater flowing into the lakes is the main source of water to the areal expansion of lakes in the region.
As a consequence of the continuous expansion of lakes and glacier retreat, GLOF is an emerging threat in the Chandra basin. Seven PDGLs were identified and evaluated qualitatively for outburst probability, among which two are highly susceptible to outburst, one medium and four are having low-outburst probability. The lakes with high-and medium-outburst probability are recommended for further detailed hazard and risk assessment. For future research, we suggest (i) a continuous monitoring of glacier and glacial lakes for future hazard assessment, (ii) GLOF modelling and potential damage assessment, and (iii) detailed investigation about development of new supraglacial lakes, bedrock-dammed lakes in contact with glacier, expansion of proglacial moraine-dammed lakes and its impact on the glacier mass balance.

Disclosure statement
No potential conflict of interest was reported by the authors.