SNOW COVER MAPPING AND ICE AVALANCHE MONITORING FROM THE SATELLITE DATA OF THE SENTINELS

In order to monitor ice avalanches efficiently under disaster emergency conditions, a snow cover mapping method based on the satellite data of the Sentinels is proposed, in which the coherence and backscattering coefficient image of Synthetic Aperture Radar (SAR) data (Sentinel-1) is combined with the atmospheric correction result of multispectral data (Sentinel-2). The coherence image of the Sentinel-1 data could be segmented by a certain threshold to map snow cover, with the water bodies extracted from the backscattering coefficient image and removed from the coherence segment result. A snow confidence map from Sentinel-2 was used to map the snow cover, in which the confidence values of the snow cover were relatively high. The method can make full use of the acquired SAR image and multispectral image under emergency conditions, and the application potential of Sentinel data in the field of snow cover mapping is exploited. The monitoring frequency can be ensured because the areas obscured by thick clouds are remedied in the monitoring results. The Kappa coefficient of the monitoring results is 0.946, and the data processing time is less than 2 h, which meet the requirements of disaster emergency monitoring. * Corresponding author


INTRODUCTION
Snow cover mapping plays a very important role in the rapid and accurate assessment of the extent and degree of a snow disaster, such as an ice avalanche. Snow cover can be mapped from remote sensing data at a large scale and with high precision (Immerzeel et al., 2009). Snow cover mapping methods based on multispectral remote sensing have been developed for a long time (Fily et al., 1997;Painter et al., 2009).
Based on the high reflectance of snow in some wavelengths compared with other natural targets, optical and near-infrared sensors can distinguish between snow-covered and snow-free ground (Warren, 1982). Landsat-7 ETM+ data were used for mapping snow and ice automatically (Sirguey et al., 2009;Taccardi, 2012), and an estimated error was given (Paul et al., 2013). The fractional snow cover can be provided by a method based on the Normalized Difference Snow Index (NDSI) of Moderate-resolution Imaging Spectrometer (MODIS) (Salomonson and Appel, 2004;Fraser et al., 2010). Sub-pixel snow cover mapping using spectral unmixing was studied, particularly for alpine snow (Painter et al., 1998;Painter et al., 2003).
Because microwaves have the characteristics of penetrating clouds, in the process of disaster emergency monitoring, microwave imaging is an effective data source. As an active microwave remote sensing technology, Synthetic Aperture Radar (SAR) is often used in snow cover mapping. Passive microwave remote sensing technology is also used in snow cover information extraction; however, the resolution of passive microwave remote sensing is low (Robinson et al., 1984).
For snow cover mapping, using the coherence of the repeat-pass Interferometric SAR (InSAR) data is a very effective method.
An InSAR coherence image is a cross-correlation product derived from two coregistered complex-valued SAR images. A loss of InSAR coherence is often referred to as decorrelation.
InSAR has been widely applied to measure glacier topography and displacements (Joughin et al., 1998). Snow cover has considerable impacts on InSAR coherence values; specifically, the coherence values of snow-covered grounds are lower than those of snow-free grounds, based on which a threshold slicing algorithm (TSA) has been developed for snow cover mapping (Kumar and Venkataraman, 2011). The measurement of InSAR coherence between two repeat passes of C-band SAR offers a way to acquire shallow dry snow areas (Zebker and Villasenor, 1992). The research results show that the coherence of the snow cover is approximately 0.31 and the coherence of a thick snow layer is lower (Kumar and Venkataraman, 2011). Lakes and forests also cause decorrelation; however, the snow cover can be distinguished from them (Shi et al., 1997;Strozzi et al., 1999). Compared with the real snow cover results, the accuracy of the snow cover mapping based on InSAR data coherence can reach more than 80%.

STUDY AREA
On July 17, 2016, an ice avalanche, the volume of which was approximately 60-70 million cubic meters, occurred in the western Aru Co lake, Rutog county, Ngari prefecture, western Tibet, killing 9 herders and hundreds of animals. The center of this ice avalanche was located at 82°23'21"E, 34°0'3"N. On September 21, 2016, another ice avalanche occurred southeast of the first avalanche. A sketch map of the delineated study area for these ice avalanches is shown in Figure 1.

Sentinel-1 Satellite Data
The Sentinels are a fleet of European Space Agency (ESA) satellites designed specifically to deliver data and imagery that are central to the European Copernicus program. Sentinel images can be acquired from the Copernicus Open Access Hub (https://scihub.copernicus.eu/dhus/#/home).

The first Sentinel satellite in the European Copernicus program,
Sentinel-1, carries a C-band SAR instrument to provide an all-weather, day-and-night supply of Earth's surface imagery (Malenovský et al., 2012). The Sentinel-1 mission benefits numerous services that relate to the monitoring of sea ice, the surveillance of marine environments, and the measurement of land surfaces for motion risks and for the management of soils .
Five phases of predisaster and postdisaster images were acquired for the study area, which were used for the snow cover mapping experiment based on the threshold segmentation of InSAR coherence images. The data list is shown in Table 1.

Sentinel-2 Satellite Data
The Sentinel-2 satellite is the first multispectral Earth observation satellite in the European Copernicus program (Malenovský et al., 2012). Sentinel-2 carries a wide swath high-resolution multispectral imager with 13 spectral bands.
The span of the 13 spectral bands, from the visible and the near-infrared to the shortwave infrared, are at different spatial resolutions from 10 to 60 meters. The combination of the wide swath of 290 km and the frequent revisit times of Sentinel-2 provides continuous views of Earth.
Three phases of predisaster and postdisaster images were acquired by Sentinel-2 for the study area, which were used for the snow cover mapping experiment based on the NDSI method, as well as the visual interpretation process. The data list is shown in Table 2 States of America). The backscattering coefficient image is shown in Figure 2  The snow cover is mapped by using the decorrelation phenomenon in the InSAR data process. Decorrelation is expressed as the low numerical value in the coherence image.
The ice avalanches in the study area occurred in the local snowmelt period; therefore, the causes of decorrelation in this experiment include the changes between the snow cover before and after the ice avalanche and the melting process of the snow.
According to these points, the coherence image of Sentinel-1 InSAR data can be segmented by a certain threshold to map the snow cover. The coherence image is shown in Figure 2 (b). It can be seen that the snow cover has low coherence values.
However, because of the absorption characteristics of water to electromagnetic waves, low coherence values are also exhibited in the water bodies. Therefore, the water bodies extracted from the backscattering coefficient image must be removed from the coherence segment result. The coherence of the newly formed ice avalanche is low (as shown in the yellow frame), which is beneficial to the extraction of the ice avalanche body.

Snow Cover Mapping from the Sentinels
Based on the statistical analysis of the snow confidence map with the visual interpretation result of the snow cover in the areas without clouds as a reference, the threshold of the snow confidence map was set to 0.8 to map the snow cover. The snow cover mapping result from Sentinel-2 is shown in Figure   5(a). result of the snow cover in the areas without clouds as a reference, the backscattering coefficient threshold was set to -15 db, and the coherence threshold was set to 0.43 to map the snow cover. The regional average elevation, which was calculated from the New Global Digital Elevation Model (GDEM V2), was used as the elevation threshold. The snow cover mapping results from Sentinel-1 is shown in Figure 5(b).
The snow cover maps were obtained from the comprehensive utilization of Sentinel-1 and Sentinel-2 data. The final snow cover mapping result is shown in Figure 5(c). It can be seen from the yellow frame that the incomplete mapping of the snow cover caused by the thick cloud obscuration has been remedied.

Ice Avalanche Monitoring
The snow cover mapping results from the predisaster data, the first ice avalanche data and the second ice avalanche data were obtained to monitor the ice avalanches, as shown in Figure 6. It clearly shows the influence extents of the two ice avalanches.
From the yellow frame in Figure 6, it can be seen that the foot of the first ice avalanche, which had slid into the lake, melted before the second ice avalanche.  Table 3. The processing of the snow cover maps could be completed in 2 h The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-3, 2018 ISPRS TC III Mid-term Symposium "Developments, Technologies and Applications in Remote Sensing", 7- Table 3. Snow cover area statistics for the mountain where the ice avalanches were located

DISCUSSION
The accuracy of the snow cover mapping was further evaluated by the Kappa coefficient. The Kappa coefficient is shown in Equation (1): The accuracy evaluation is shown in The accuracy of the proposed method is similar to or better than that of most of the other methods using remote sensing images with medium resolution, except that of the subpixel snow cover mapping algorithm using an airborne hyperspectral image (as good as 95%) (Jebur et al., 2014;Chen et al., 2014).
However, mapping snow cover with satellite images from the Sentinels has the advantage of wide monitoring coverage.
Moreover  It can be seen that the result of the Sentinel-1 after the removal of water bodies cannot effectively display the foot of the ice avalanche that slid into the lake. Therefore, it cannot reflect the phenomenon of ice melting in the lake, as shown in Figure 6 of Section 5.2.

CONCLUSIONS
Snow cover mapping results were generated from the InSAR coherence image of Sentinel-1 and the atmospheric correction result of Sentinel-2 for the study area in Rutog county, Ngari prefecture, western Tibet. The comparison of the extracted snow cover and the visual interpreted snow cover showed that the accuracy of the mapping result was equivalent to that of the visual interpretation result. The applicability of the method and the accuracy of the results were analyzed and evaluated. The extraction accuracy and time consumption met the application requirements for ice avalanche emergency monitoring.
The innovations of this paper were as follows. First, an appropriate method of snow cover mapping from images of the Sentinel satellites was proposed, and the application potential of the Sentinel data in the field of snow cover mapping was exploited. Second, the snow cover mapping method based on the microwave features of SAR images and that based on the spectral features of multispectral images have been applied in collaboration, which not only avoids the interference of thick cloud obscuration on the mapping results but also ensures the accuracy of the mapped boundary.