SNOW COVER AND GLACIER DYNAMICS STUDY USING C-AND L-BAND SAR DATASETS IN PARTS OF NORTH WEST HIMALAYA

The seasonal snow cover and permanent ice in form of Himalayan glaciers provide fresh water to many perineal rivers of Himalayas. The melt water from seasonal snow and glaciers, especially during of 15 March to 15 June acts as important source of water for drinking, hydropower and irrigation requirements of many areas in North India. This work has highlights the use of C-band Synthetic Aperture Radar (SAR) data from RISAT-1, Sentinel-1A and 1B satellites and ALOS-PALSAR-2 PolInSAR data for snow cover and glacier dynamics study for parts of North West Himalaya. Glacier velocity was derived using InSAR based method using 6 day temporal interval images from Sentinel-1 satellites and 14 day interval for PALSAR-2 satellite. High coherence was obtained for main glacier in both the data sets, which resulted accurate line of site (LOS) glacier velocity estimates for test glaciers. These InSAR data glacier velocity results are obtained after a gap of 21 years. Glacier facies was estimated using multi-temporal SAR image composition based classification. All these maps were verified by extensive ground surveys done at these sites during 2014-2017. The time series data of C-band SAR in VV/VH polarisation was also used to map snow cover in test basins of Bhagirathi and Beas River. The VV/VH data clearly shows difference between dry and wet snow, thus helping in improved snow cover mapping using SAR data. This study will help in refining algorithms to be used for such studies using upcoming NASA-ISRO SAR (NISAR) mission.


INTRODUCTION
The seasonal snow cover and glacier are the globally recognised essential climate variables (WMO-2017), which controls the water availability in head water Himalayan watershed and rivers basins Kumar 1997, 2007). The brief about snow cover and glacier mapping and monitoring using remote sensing, the gap areas and objectives are given in this section.

Snow Cover
The Seasonal Snow Cover (SSC) in Indian Himalayan Region (IHR) is fed mainly by North West winter monsoon or Western Disturbances, WDs (Dimri et al., 2015) in mid and high elevation areas (> 1800m) and also in parts by South West Indian Summer Monsoon (ISM) mainly in higher elevation zones (>5000 m), mainly due to elevation induced lapse rate and low temperature in high elevation areas (Pepin and Lundquist, 2008;Thayyen and Dimri, 2018). The snow accumulates during the winter till March April and melts during spring and summer months, with minimum snow cover at the end of ablation season, i.e. last week of August (Thakur et al., 2017a,b). This snow melt is also augmented by glacier melt, which contributes significantly during ablation season (June-August) and minimum glacier melt in winters Agrawal et al., 2017). The vast and high relief areas of IHR including North West Himalaya (NWH) region cannot be accessed using traditional ground based field surveys. Therefore, Remote Sensing (RS) remain one of most important and feasible technology to map and monitor the SSC and its physical properties at various spatio-temporal scales, as proven by many operational and research studies Thakur et al., 2012Thakur et al., , 2017aJain et al., Negi et al., Sharma et * Corresponding author al., Nikam et al., 2017). The RS based snow cover and its properties have direct use as input to snowmelt runoff models (Jain et al., 2009;Prasad and Roy, 2005;Thakur et al., 2009;Aggarwal et al., 2013;Thakur, 2014;Wulf et al., 2016), and as validation and data assimilation in process based hydrological models (Naha et al., 2016;Agarwal et al., 2016). The optical data is capable of giving good quality snow cover maps, but synthetic aperture radar (SAR) data is able to provide, dry/wet snow and snow physical properties as well, due to penetration capability and sensitivity to snow wetness Venkataraman 2007, 2009;Thakur et al., 2012;Snapir et al., 2019). The only major limitation of SAR sensors is no penetration when snow is wet, less temporal resolution and operations in single wavelength (Hallikainen et al., 2001;Thakur et al., 2012).

Glacier Dynamics
The glacier are defined as moving rivers of ice (Paterson, 1994;Pellika and Rees, 2009). The Himalayan glacier are always located above approximately 4000m elevation and are generally difficult in-accessible terrains. The RS remain one of most popular technologies to map and monitor these icy rivers, which provides an important contribution to river flow in head water river basins of NWH such as Indus, Beas, Satluj and Ganga Singh, 2001, Singh andJain, 2002). These glacier are highly sensitive to short and long term variations in temperature and precipitation variations, as well as climate change, and in last 50 years, most of these glaciers have shown negative mass balance and snout retreat (Wagnon et al., 2007;Bhambari and Bolch 2011;Bolch et al., 2012, Dobhal et al., 2008Azam et al., 2012;Sharma et al., 2013;Pratap et al., 2016) as observed by remote sensing and traditional ground based glaciological measurements. The dynamics of these glacier can be effectively observed and mapped by RS based methods (Bhambari et al., 2012), which can quantity its area, length and elevation changes over time, as well as its surface velocity and glacier facies (Joughin, 2010;Berthier et al., 2005;Kumar et al, 2011;Bhardwaj et al., 2015;Thakur et al., 2016Thakur et al., , 2017a. The vast amount of free and multi-temporal optical and SAR data from various space agencies have opened new applications in the field of glacier dynamics monitoring. Therefore, this work is intended to showcase a few of such glaciology applications using free RS datasets, and their utility in monitoring the overall health of Himalayan glaciers

Objectives
The main objectives of the present work are to map dry and wet snow cover using freely available SAR data, to map the glacier facies using multi-temporal freely available C-band SAR data and to estimate glacier velocity using DInSAR (C and L band) and optical datasets. In addition, the tandem-x data was also used for elevation estimation of few selected glaciers.

Study Area
The study area for SAR based snow cover analysis is Beas River and Upper Ganga River basins in Himachal Pradesh and Uttarakhand states of India. The Figure 1 shows the location of these basin in India. The study area for Glacier dynamics is few major glaciers of NWH, such as Siachen, Gangotri and Bara Shigari glaciers, as shown in Figure 2. These three glacier are the longest and largest in terms of length and total area, in Indian Himalayas, and they fall in Indus, Ganga and Chenab river basin head watersheds, respectively. The total area of Beas basin upto Thalot and Ganga basin upto Haridwar is 4975 km 2 and 23, 177 km 2 , respectively. The maximum and minimum snow cover area in these basins varies from 20 to 80 % for Beas and 20 to 60 % for Ganga. The relative contribution of snow melt to total river discharge is highest in the headwater catchments and it decreases significantly as we move further downstream.
Figure 2: Selected glaciers (Siachen, Gangotri and Bara Shigari) for glacier dynamics study in NWH region (The background RS data is from Digital Globe and USGS).

Data Used
The free ALOS-PALSAR-2 data was provided by JAXA announcement of opportunity science project, RISAT-1 data was procured from National Remote Sensing Centre's (NRSC's), India national data centre (NDC) and Sentinel 1A &1B data was taken from Alaska SAR facility. Earlier studies using Interferometric SAR (InSAR) data for Himalayan glaciers has suffered from loss of coherence due to low temporal resolution of SAR images and high glacier velocity. In addition, TANDEMx data of year 2012 and 2013 was also used to derive temporal elevation changes due to dynamic snow/ice condition of Gangotri glacier. The time series of SAR data from RISAT-1 and Sentinel-1 satellites was used for glacier in Chenab and Gangotri basins, for creating SAR image composites, identify and map radar glacier zones such as debris ice, percolation-refreeze zone, wet snow and dry snow zones. The ascending and descending passes of Sentinel-1 SAR data has given temporal coverage of 4 to 6 days to map and study the snow cover and the glacier dynamics in test basins of NWH. All SAR data was processed using Sentinel-1 toolbox (SNAP) of ESA (ESA, SNAP, 2017).

Snow Cover Dynamics Using C-Band SAR Data
The main approach of estimating the dry, wet and total snow cover is similar to the earlier studies (Nagler andRott, 2000, Thakur et al., 2016), however, the input data is C-band SAR from Sentinel-1A/1B satellites and algorithm uses Google earth Engine (GEE) codes to access, and analyses time series data, as recently demonstrated by (Snapir et al., 2019). The VH polarization is used with Sentienl-1 data, as it has shown higher sensitivity to dry and wet snow ) as compared to the VV polarization. The brief methodology flow chart is given in figure 3.

Glacier Dynamics Using SAR/Optical Data
The glacier dynamics is studied by estimating the surface glacier velocity and mapping of glacier radar zones using SAR datasets. The DInSAR based glacier velocity method employed in the present study is given in Figure 4. In both cases of co-registration and re-flattening, SRTM-x band 30 m DEM (Farr et al., 2007) was used to remove topographic phase due to its better resolution and vertical accuracy.  Sood, 2015).
The multi-temporal SAR image based composite method is used for identification of the glacier facies ( Figure 5), this approach was first used by Patrington (1998) and has proven its utility using RISAT-1 based datasets for Gangotri glacier (Thakur et al., 2017a). The support vector machine (SVM) method of supervised image classification is used with SAR colour composite image for final glacier facies mapping.

RESULTS AND DISCUSSIONS
The results section is divided into two sub-sections, the first subsection gives SAR based dry, wet snow cover mapping results and next sub-section gives main results of glacier velocity and facies mapping for selected glaciers of NWH.

Snow Cover Results
The C-band SAR based results for study area are shown in Figures 6 and 7. The variable and multiple threshold approach is used to classify the snow as dry and wet snow in both the case. The wet snow is taken below -20 db and dry snow above -5 db, with 0 and above values taken as layover or double bounce features. The Google Earth Engine (GEE) is used (Gorelick et al., 2017) to select and download the Sentinel-1 SAR data. The Threshold and final mapping is completed in the ArcGIS software. The both Ganga and Beas basin are covered with a single full scene of Sentinel-1 data in descending mode, hence snow maps of this region can be generated for every 6, 12 or 24 day time interval. Total 37.06% of Ganga basin was covered with snow, out of which, 23.61% are had dry snow and 13.45% was covered with wet snow for 28 March 2017. The Beas basin has total snow cover of 56.85%, with dry snow of 29.38 % and wet snow of 27.47 % on 14 April 2017. The wet snow in Beas basin was higher as compared to Ganga basin, mainly due the late spring time period. The accuracy of SAR based snow maps is more than 95 %, when compared with MODIS based snow maps. The layover mask will also reduce some of the usable area (>10%) in The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-5, 2018 ISPRS TC V Mid-term Symposium "Geospatial Technology -Pixel to People", 20-23 November 2018, Dehradun, India the case of Sentinel-1 SAR datasets, as they have fixed incidence angle range (32.9 o for IW1, 38.3 for IW2 and 43.1 for IW3) These are initial results and work is in progress to create long duration weekly or 10 daily time series of dry and wet snow cover maps using Sentinel-1 SAR data and reference image threshold approach (Nagler and Rott, 2000;Thakur et. al., 2013). This can be an improvement over monthly snow maps as derived by Snapir et al., (2019).

Glacier Dynamics Results
The glacier dynamics, the variation in the glacier surface velocity and glacier facies have been estimated using SAR based data and presented in this section.

Glacier velocity estimation:
The DInSAR data from Sentinal-1A satellite have been processed for three major glaciers of NWH. The results are shown from Figures 8 to 13. The coherence for main parts and trunk glacier of these test sites was higher than 0.8, mainly due to the 6 day interval Sentinel-1 InSAR datasets. Similarly 14 day interval ALOS-APLSAR-2 data was used for glacier LoS velocity estimation in Chandra Basin of Chenab River, covering Smudra Tapu, Batal, Chota and Bara Shigri glaciers. The LoS glacier velocity for Chandra basin glacier is shown in Figure 13, with value ranging from -0.176 m to +0.13 m/day in 14 days. The 6 day interval InSAR data from Sentinel-1 is available for more winter dates of 2016-2017 and 2017-2018, and can give seasonal to annual estimation of glacier velocity in Himalaya.

Glacier facies mapping:
The results of multi-temporal SAR image colour composite based glacier facies identification and mapping are given in this sub-section. In all the colour composites, the red colour is assigned to early summer (April-May) image, Green to late summer (August end) and Blue to winter (January-February) SAR images, as demonstrated in previous studies (Sood, 2015;Thakur et al., 2016Thakur et al., , 2017a    The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-5, 2018 ISPRS TC V Mid-term Symposium "Geospatial Technology -Pixel to People", 20-23 November 2018, Dehradun, India Figure 19: Multi-temporal SAR image composite of Bara Shigri glacier (2015).
Glacier facies can be easily identified in all the SAR based temporal composites (Figure 16 to 18), with light grey, purple and blueish showing presence of percolation-freeze zone in higher, middle and lower elevations of a glacier. Green, mixed green blue showing clean or bare ice, and dark grey as dry snow zone (Patrington, 1998;Thakur et al., 2016Thakur et al., , 2017a. The SVM based supervised classification method was used to map glacier facies classes such as debris cover glacier ice, bare ice, percolation-refreeze radar zone in various elevations, wet snow and supra-glacier lakes with Sentinel-1 datasets for these glacier, with sample map shown for Gangotri glacier (Figure 18). The annual time series maps of glacier facies can be easily used to identify and map accumulation and ablation zones, along with equilibrium line altitude (ELA) of a glacier, thereby giving direct input or indicating about heath of mapped glacier (Thakur et al., , 2017a, in a given water year.

CONCLUSIONS
The study has again proven significant improvement in application of SAR data for snow cover mapping and glacier health monitoring, if systematic and long duration time series data is available (Sentnel-1 in present case), as also proven in the previous study with RISAT-1 data (Thakur et al., 2017a). The snow cover in terms of dry and wet snow was successfully estimated using C-band SAR data with VH backscatter images for Ganga and Beas River basins. The Sentinel 1A/1B data with 6 to 12 day interval can be used for operational wet/dry snow mapping in Himalayas, if we combine both descending and ascending pass data at 8 daily time scale, and this will be complimentary to optical images based operational snow cover maps.
This study is also significant, as it shows for the first time use of freely available InSAR/SAR data of Sentinel-1 for glacier velocity and glacier facies mapping in all major glaciers of NWH. The derived glacier velocity is reported in LoS glacier values only, and further work is required to covert this LoS data into actual glacier velocity using horizontal glacier flow assumption and use of external Digital Elevation Model (DEM) (Wangensteen et al., 2005). The 3D glacier velocity can also be estimated if both Descending and Ascending pass InSAR data is made available (e.g., Kumar et al., 2011) by European Space Agency (ESA) at 6 day interval, in near future, as at present, only descending mode data is available with 6 day interval for selected time and glacier of Himalaya. The results generated and experience gained from the present study will help in improving algorithms to be used for such studies using upcoming NASA-ISRO SAR (NISAR) mission.