The spatial and temporal heterogeneity of seasonal snow and its impact on socio-economic and environmental functionality make accurate, real-time estimates of snow water equivalent (SWE) crucial for hydrological and climatological predictions. Remote sensing techniques allow for a cost effective, temporally consistent and spatially continuous approach to SWE monitoring in areas where in situ measurements are not sufficient. Currently, passive microwave (PM) remote sensing is the most reliable method for estimating large scale SWE variability in a dry snowpack from space. However, its coarse spatial resolution (~25km by 25km) hinders its use for local to regional applications. Downscaling techniques can be implemented to increase the spatial resolution of gridded datasets with the use of related auxiliary datasets at a finer spatial resolution. These methods have been successfully implemented to remote sensing datasets such as soil moisture estimates, however, limited work has applied such techniques to snow related datasets. This study focuses on downscaling the European Space Agency’s (ESA) Globsnow product in Northern United States using 7 satellite derived auxiliary datasets: Sentinel 1 and Sentinel 2 snow cover extent products, ESA’s GlobCover land cover product, Sentinel 3 land surface temperature and normalized difference vegetation index, Sturm snow density and terrain elevation. A hybrid downscaling technique combining regression with area to point interpolation is applied to Globsnow data to estimate SWE at a 1km spatial resolution. Hybrid methods allow for relationships between SWE and auxiliary data as well as the spatial structure of SWE to be incorporated. The National Snow and Ice Data Centre’s (NSIDC) Snow Data Assimilation System (SNODAS) product estimates SWE at a 1km spatial resolution for contingent United States and is used to develop a regression relationship between SWE and the 6 auxiliary datasets. Using a regression-area to point interpolation downscaling method, a SWE estimate for every 1km2 pixel within the original 25km2 pixel is determined. SNODAS is then compared to the downscaled Globsnow product at the pixel level for validation of the downscaling method. The goal of this study is to develop a reliable and reproducible method for downscaling coarse resolution SWE data in areas where finer resolution data such as SNODAS does not exist. Once the methodology is finalized, the same procedure will be applied and tested in an area of Southern Ontario using acquired snow transect data for validation.

Area D - Board 454

Localized, continuous measurements of microwave signatures are instrumental in the development of geophysical retrieval algorithms for satellite sensors. Ground-based instruments allow to relate these signatures to quantified natural processes with increased accuracy and temporal resolution compared to direct use of satellite observations; in particular for coarse resolution passive microwave sensors, understanding the different contributions from heterogeneous terrain components is a prerequisite for interpreting satellite observations, necessitating focused observations over several terrain types. Changes in e.g. snow microstructure, wetness and impurities have different effects at different bands across the microwave and optical spectrums, highlighting the importance of multi-wavelength observations.
The ground-based observation program at the Finnish Meteorological Institute Arctic Space Centre (FMI-ARC) aims to provide season-long, hourly to daily observations of microwave and optical signatures from snow cover on natural terrain types representative of the Northern Boreal forest zone. The focus of the program is on collecting multi-wavelength optical and active and passive microwave observation using ground based instruments, operated throughout the year, capturing the full variety of soil, snow, vegetation and atmospheric conditions, which undergo significant changes throughout the year in Northern latitudes [1]. Measurements in natural surrounding have been complemented by observations of snow samples in controlled, laboratory-like conditions [2], as well as airborne campaigns. An essential part of the Sodankylä observation program is the extensive coinciding automatic collection of (a) in situ data of the processes of the boreal ecosystem and (b) meteorological data. These include in addition to snow cover characteristics e.g. profiles of soil moisture and temperature, diurnal vegetation dynamics and greenhouse gas fluxes between the ecosystem and the atmosphere

The collected data has allowed the development of novel retrieval algorithms for variables such as snow water equivalent (SWE) [3], and refining forward models for snow reflectance as well as microwave emission and backscatter [4,5,6]. In this presentation we summarize the main collected datasets, and describe plans for future campaigns which include quantifying the influence of forest canopy conditions on snow cover remote sensing using different methods.

[1] Lemmetyinen, J.; A. Kontu, J. Pulliainen, J. Vehviläinen, K. Rautiainen, A Wiesmann, C. Mätzler, C. Werner, H. Rott, T. Nagler, M. Schneebeli, M. Proksch, D. Schüttemeyer, M. Kern, and M. Davidson, 2016. Nordic Snow Radar Experiment. Geosci. Instrum. Method. Data Syst., 5, 403-415, doi:10.5194/gi-5-403-2016.
[2] Maslanka, W., L. Leppänen, A. Kontu, M. Sandells, J. Lemmetyinen, M. Schneebeli, M. Proksch, M. Matzl, H-R. Hannula, R. Guerney, 2016. Arctic Snow Microstructure Experiment for the development of snow emission modeling. Geosci. Ins. Methods Data Sys. 5(1), 85-94.
[3] Leinss, S., A. Wiesmann, J. Lemmetyinen, and I. Hajnsek. Snow Water Equivalent of dry snow measured by differential interferometry, 2015. IEEE J Selected T. App. Earth Obs. Remote Sens., 8(8): 3773-3790.
[4] Pulliainen, J., M. Salminen, K. Heinilä, J. Cohen, HR. Hannula, 2015 Semi-empirical modeling of the scene reflectance of snow-covered boreal forest: Validation with airborne spectrometer and LIDAR observations. Remote Sens. Environ., 155, 303-311, 0.1016/j.rse.2014.09.004.
[5] Proksch, M., C. Mätzler, A. Wiesmann, J. Lemmetyinen, M. Schwank, H. Löwe, and M. Schneebeli, 2015. MEMLS3&a: Microwave emission model of layered snowpacks adapted to include backscattering. Geosci. Model Dev., 8, 2611-2626.
[6] Tan, S., W. Chang, L. Tsang, J. Lemmetyinen, and M. Proksch, 2015. Modeling both active and passive microwave remote sensing of snow using dense media radiative transfer (DMRT) Theory with multiple scattering and backscattering enhancement. IEEE J Selected T. App. Earth Obs. Remote Sens., 8(9), 4418-4430.

Area D - Board 455

Surface albedo is the fraction of incoming solar radiation reflected hemispherically by the surface. It is directly related to the global energy budget. Changes in the surface albedo can be used as an indicator of a climate change, so it is important that all the factors affecting the measurements or calculations of the albedo values are properly taken into account. The boreal forest zone is an important component to the carbon budget in the northern hemisphere and it also affects the surface albedo values in a large area. One of the major factors affecting surface albedo in the boreal forest zone is the snow cover. Determining correctly the snow cover or lack of it in the boreal forest floor is crucial for reliable albedo estimation. In addition, the canopy and the forest floor scattering has to be modelled properly in order to get reliable albedo estimates.

We studied the effect of snow covered and snow-free forest floor to the total black-sky albedo values theoretically by using the extended PARAS albedo model. The model takes into account all the different photon absorption and scattering events of and between canopy and the forest floor. We simulated the broadband black-sky albedo values with varying leaf area index (LAI) values for the latitudes range 40N-80N in the northern hemisphere. The LAI map used was provided by the Copernicus Global Land Service. The canopy type, i.e. distinction between the coniferous and deciduous forests, was provided by the land cover map GLOBCOVER 2009.

We discovered that the snow cover in the forest floor can change the albedo values from around 0.1 (snow-free) up to around 0.65 (snow covered), depending on the latitude and the LAI values. The differences are larger for smaller LAI values and lower latitudes (closer to 40N), where the sun elevation at midday is higher so that the direct forest floor illumination is larger. We also discovered that there are local maxima in the albedo values at the noon when there is snow on the forest floor, whereas in snow-free conditions the opposite takes place. The mean differences of daily albedo values between snow covered and snow-free conditions was in the range of 0.06 ... 0.17 depending on the day of year. Treating the forest as a chess board of dark canopy and bright snow at the forest floor leads into an underestimation of 5% to 43% depending on latitude of the total forest albedo.

Area D - Board 456

Snow is the most reflective natural surface. However, it has been found that there is a tendency on a global scale for snow darkening and snow extent reduction due to pollution and global warming. Therefore, there is a need for better modelling/remote sensing of natural snow using passive and active remote sensing from UV to the microwave regions of electromagnetic spectrum.
In this paper we present the recent advances in theoretical modeling of light transport inside natural snowpack. The light transmission and reflection by optically thick snowpack is considered. Both exact radiative transfer theory and approximation valid in the limit of weakly absorbing snow is conidered. The results of random walk theory as related to light reflectance from snow are also presented. The approximate theory is validated comparing with both radiatiev transfer calculations and ground measurements of light reflectance and transmittance for snowpacks of different thicknesses with different types of underlying surfaces. The question of possibility to determine the concentration of pollutants in snow from ground and satellite measurements is studied in detail. It has been found that optical methods are not suitable for the determination of pollution load for most of natural snowpacks because of generally low load of pollutants. The local optical properties of snowpacks are studied in the framework of the geometrical optics approach. The use of Maxwell theory can be avoided due to large size of grains as compared to the wavelength of incident light. It has been found that Mie thory can not be used for the description of snow reflectance. This is due to the fact that large spherical particles produce angular features never observed for snowpack (such as rainbow and glory). The degree of light polarization by snowpack is studied as well. It has been found that the degree of light polarization reflected by snow is low.

Area D - Board 457

TThe synergistic use of data from different satellites of the Sentinel series offers excellent capabilities for generating high quality products on key parameters of the global climate system and environment. A main parameter for climate monitoring, hydrology and water management is the seasonal snow cover. In the frame of the ESA project SEOM S1-4-SCI Snow, led by ENVEO, we developed, implemented and tested a novel approach for mapping the total extent and melting areas of the seasonal snow cover by synergistically exploiting Sentinel-1 SAR and Sentinel-3 SLSTR data and apply these tools for snow monitoring over the Pan-European domain.

Data of the Copernicus Sentinel-1 mission, operating over land surfaces in Interferometric Wide Swath (IW) mode at co- and cross-polarizations, are used for mapping the extent of snowmelt areas applying change detection algorithms. In order to select an optimum procedure for retrieval of snowmelt area, we conducted round-robin experiments for various algorithms over different snow environments, including high mountain areas in the Alps and in Scandinavia, as well as lowland areas in Central Europe covered by grassland, agricultural plots, and forests. In mountain areas the tests show good agreement between snow extent products during the melting period derived from SAR data and from Sentinel-2 and Landsat-8 data. In lowlands ambiguities may arise from temporal changes in backscatter related to soil moisture and agricultural activities. Dense forest cover is a major obstacle for snow detection by SAR because the surface is masked by the canopy layer which is a major scattering source at C-band. Therefore, areas with dense forest cover are masked out. Based on the results of the round-robin tests we selected for the retrieval of snowmelt area a change-detection algorithm using dual-polarized backscatter data of S1 IW acquisitions over land. The algorithm applies multi-channel speckle filtering and data fusion procedures for exploiting VV- and VH-polarized multi-temporal ratio images. The binary SAR snowmelt extent product at 100 m grid size is combined with the Sentinel-3 SLSTR snow product in order to obtain combined maps of total snow area and melting snow. Complementary to the snowmelt area extent from SAR, the optical satellite images provide information on snow extent irrespective of melting state but are impaired by cloud cover. For generating a fractional snow extent product from Sentinel-3 SLSTR data we apply multi-spectral algorithms for cloud screening, the discrimination of snow free and snow covered regions and the retrieval of fractional snow extent. In order to fill gaps in the optical snow extent time sequence due to cloud cover we apply a data assimilation procedure using a snow pack model driven by numerical meteorological data of ECMWF, simulating daily changes in the snow extent. We present the results of round robin experiments of SAR wet snow algorithms and their performance in different environments and show time series of synergistic snow products using SAR and optical satellite data over the Pan-European domain.

Area D - Board 458

Mapping snow cover (SC) on glaciers at the end of the ablation period supports the comprehension how and in which rate a glacier accumulates and ablates, respectively, and which influence climate variabilities have. Furthermore, the SC can be used for the determination of the mass balance. Remote Sensing data are able to provide this information for remote, large glaciers and back in time. With optical satellite imagery it is possible to detect SC on glacier relatively straight forward when no clouds are present and when the bare ice is clearly different from the snow zone on the glacier. Optical satellite data fail, however, to identify whether the snow on the glacier is dry, wet or whether firn is present or not. In contrast, SAR data like TerraSAR-X or Sentinel-1 can discriminate dry snow, wet snow, firn and bare ice.
In this contribution, we compare three types of multi-temporal SC maps derived of optical and SAR data for a set of glaciers in the Ötztal Alps-Austria. For several acquisition times we performed a multiple digitizing experiment by manually mapping the SC area on the glaciers based on high resolution Landsat image which served in the following as reference data. Additionally, we applied the automatic SC mapping tool named ´automated snow mapping on glaciers´ (ASMAG) based on Landsat imagery. The algorithm automatically determines cloud cover and converts raw digital numbers into topographically corrected top of atmosphere reflectance (TOAR). Subsequently, a reflectance threshold for the near-infrared band is automatically selected to map snow on glaciers. The other two snow maps were classified of the SAR data. In a first step, the TerraSAR-X and Sentinel-1 were processed with the Multi-SAR System, which is a pre-processing system developed at the DLR. It includes the geocoding, multi-looking, radiometric calibration and polarimetric decomposition. Subsequently, two snow masks were classified, whereas the first was derived with the help of the coefficient of variation applied on the VH amplitude and the second on the polarimetric information.
A first assessment shows that the manual derived outlines and the SC maps from ASMAG are of high accuracy (~90%). The maps derived from TerraSAR-X are more difficult to interpret. Wet snow maps agree by about 60% with the SC maps from ASMAG, however, dry snow maps, firn and bare ice are difficult to classify. On the other hand, snow facies maps from Sentinel 1 shows a higher agreement in respect to TerraSAR-X, which might be due to the different snow conditions during the acquisitions of TerraSAR-X and Sentinel-1 (5:30 a.m. UTC) and Landsat (11 a.m. UTC).
We conclude, that the combination of different sensors is highly recommendable to support a continuous SC monitoring and to eliminate the disadvantages of each sensor. Also, the round robin experiment resulted in a high agreement. Due to the high repetition time of Sentinel-1 data, the comparison of the SC maps is facilitated in contrast to TerraSAR-X.

Area D - Board 459

Snow on the ground is an important resource for mountain regions to sustain river flow, to provide freshwater input to ecosystems and to support winter tourism. The level of activity, employment, turnover and profit of hundreds of ski resorts in the European Alps primarily depends on meteorological conditions, in particular natural snowfall, but nowadays also conditions favorable for technical snowmaking. In order to mitigate the impact of the large inter-annual variability of snow conditions, and more recently as adaptation measure to the effects of climate change, ski resort managers have massively improved their snow management practice. Nonetheless, there is still a significant potential to increase the efficiency of the snow production by improving the anticipation capabilities at all time scales, spanning from “weather forecast” (up to 5 days typically) to “seasonal prediction ” at longer scales (up to several months).
In this context the PROSNOW project, financed by the European Union, addresses these limitations by building a demonstrator of a meteorological and seasonal prediction and snow management system from one week to several months ahead, specifically tailored to the needs of snow managers. PROSNOW will apply state-of-the-art knowledge relevant to the predictability of atmospheric and snow conditions. In detail, the models AMUNDSEN (Strasser, 2008), Crocus (Vionnet et al., 2012), and SNOWPACK/Alpine3D (Bartelt and Lehning, 2002; Lehning et al., 2006) are applied to compute the expected snowpack conditions. These models comprise sophisticated snow management modules configured to work over 9 representative ski resorts located in the European Alps i.e., France, Switzerland, Germany, Austria and Italy.
The aim of this work is to use the Sentinel-2 satellite images to assess the presence/absence of technical and/or natural snow simulated by the snow models, especially at the beginning and at the end of the snow season, which are the most critical periods for the ski resort snow management. To this purpose, a machine learning algorithm based on Support Vector Machine is implemented to identify the snow-covered and snow-free areas. In order to properly detect the snow management activities on the ski slopes, the spectral information provided by Sentinel-2 is used at 10 meters resolution. The areas covered by a mixture of snow and non-snow, clouds or in bad illumination conditions are classified as no-data (see fig. 1) and only the most certain classified snow or snow-free pixels are considered in the validation framework. This allows the generation of a significant amount of reference pixels to spatially assess the accuracy of the snow simulations.

Fig. 1. Example of assessment of snow cover in the ski resort located in Colfosco (Italy), 11st December 2015. (a) Sentinel-2 false color composition R:NIR-G:RED-B:GREEN. (b) Proposed classified map. (c) AMUNDSEN snow cover map.

Acknowledgement:
This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 730203.
References:
Bartelt, P., Lehning, M., 2002. A physical SNOWPACK model for the Swiss avalanche warning. Cold Reg. Sci. Technol. 35, 123–145. https://doi.org/10.1016/S0165-232X(02)00074-5
Lehning, M., Völksch, I., Gustafsson, D., Nguyen, T.A., Stähli, M., Zappa, M., 2006. ALPINE3D: a detailed model of mountain surface processes and its application to snow hydrology. Hydrol. Process. 20, 2111–2128. https://doi.org/10.1002/hyp.6204
Strasser, U., 2008. Modelling of the mountain snow cover in the Berchtesgaden National Park. Berchtesgaden National Park, Berchtesgaden.
Vionnet, V., Brun, E., Morin, S., Boone, A., Faroux, S., Moigne, P.L., Martin, E., Willemet, J.M., 2012. The detailed snowpack scheme Crocus and its implementation in SURFEX v7.2. Geosci. Model Dev. 5, 773–791. https://doi.org/10.5194/gmd-5-773-2012

Area D - Board 460

Optical remote sensing of the seasonal snowpack in mountainous regions provides valuable information for numerous applications including mountain hydrology, natural hazard forecasting, and climate modelling. Satellite retrievals of the extent and physical properties of the snowpack have previously been shown to be robust for flat terrain configurations. However, in mountainous areas, the radiometric effects caused by the rugged topography introduce large uncertainties in satellite observations and complicate the estimation of the physical properties of the snowpack. Indeed, changes in surface topography increase the variance in spaceborne measurements due to 1) the change in the local illumination angle as a function of the relative geometry between the Sun and the target owing to the surface slope and aspect, 2) the reduced portion of the diffuse sky irradiance that reaches the surface owing to surrounding mountains, as well as shadows blocking direct irradiance, 3) the highly reflective surfaces that contribute strongly to the multiple reflections of radiation between the target and the surrounding slopes, which are exacerbated by grazing sun angles in winter, often associated with the maximum snow cover.

Over the two last winter seasons (2016/2017 and 2017/2018) two extensive field measurement campaigns have been carried out by the team at the Col du Lautaret validation site (2100 m a.s.l) in the French Alps, to better characterise the effects of rugged terrain on optical satellite retrievals. In order to address the issues that arise when converting a Top-of-Atmosphere (TOA) signal to information about the snow surface in mountainous regions, specific efforts were dedicated to measuring the snowpack’s physical (spectral albedo, specific surface area) and chemical properties (impurities content, aerosol species in the snowpack) for a wide variety of topographic conditions. Regular sampling throughout both seasons allowed to track the temporal evolution of the snow properties, with high-resolution time-series spanning two years. The collected dataset is invaluable for the Sentinel-3 OLCI calibration activities carried out by the team, as well as providing input data for the development of a forward model designed to predict TOA radiances from in situ snow properties.

This presentation will provide a brief review of the different methods used to characterise the snowpack at Col du Lautaret, as well as an overview of the different satellite validation approaches carried out at the site, with emphasis on the issues linked to complex terrain.

Area D - Board 461

Knowledge about the snow cover distribution is of high importance to be considered for climate studies, weather forecast, hydrological investigations, irrigation and tourism, respectively. The key aspects are the pronounced albedo change between snow cover and ground but also water storage influencing run-off in the following spring and summer season. Distribution of snow on the globe is highly variable in space and time. Especially the annual variability of snow extent of the northern hemisphere ranges from 3 to 50 million square kilometers from summer to winter, respectively. Considering the high dynamic and the vast spatial distribution only satellite data can provide the needed information. Related to climate studies a time series of more than 30 years is needed for statistical sound analysis as defined by WMO. Regarding these general conditions the only usable sensor is the Advanced Very High Resolution Radiometer (AVHRR) on the polar orbiting satellites NOAA and MetOp providing data from 1981 – 2019 (most likely until 2025) with daily resolution. Based on the developments of ESA CCI cloud project we have used the calibrated and geocoded AVHRR-GAC (Global Area Coverage) data set with a spatial resolution of 4km including a consistent cloud map. The retrieval of snow extent considers the high reflectance of snow in the visible spectra and low reflectance values in the near infrared combined in the Normalized Difference Snow Index (NDSI). Additional thresholds are included, related to topography and land cover, to allocate every pixel as snow or no-snow. First results will be shown covering Europe, Asia and North America in comparison with medium-resolution data from AVHRR (1km) and high-resolution images from Landsat (30m). The presentation will also include first results on gap-filling considering the snow line and temporal evolution of snow cover. Based on the preliminary results the final processing will be adapted to start the global processing of a time series of almost 40 years fulfilling the requirements of WMO. The final product will support the climate community to compare model results using an independent data set, which has a daily resolution and is consistent over time with an uncertainty measure per pixel.

Area D - Board 462

In the last decades, monitoring the cryosphere with remote sensing data was successfully established as a complementary approach to in-situ measurements and field surveys. The major advantage of remote sensing is its spatial explicit nature of the data and the large spatial coverage. On the other side, numerical snow models can provide information on snow pack characteristics such as snow wetness and snow water equivalent which are hardly directly observable from satellite data. However, especially at large scales, a proper spatial representation of snow and ice related processes in models remains challenging e.g. due to a) the strong dependency of models on reliable meteorological forcing data; b) improper process representations at the scales of interest (i.e. simulation scale vs. process scale), c) an inadequate reflection of the spatial heterogeneity of snow and ice related processes by model parameters.
Indeed, both models and remote sensing data do have limitations in describing accurately the cryosphere. However, the nature of these limitations differs fundamentally. A combined approach, in which model and remote sensing data are opportunely integrated, is thus one of the possible solutions to increase the representativeness and to reduce the uncertainties associated to the estimation. A well-known family of techniques for data integration between remote sensing data and models is data assimilation. These techniques are massively used in atmospheric science and hydrology, but the main drawback is in dynamically incorporating the observation into the model, which may be complex.
Changing the paradigm, with this work we propose to jointly use the outputs of a physical model and the remote sensing products to improve the estimation of some of the most relevant snow and glacier parameters e.g., snow water equivalent and glacier mass balance. This new fusion paradigm has the advantage to be independent from the model and the remote sensing technique employed. In detail, within the CRYOMON SciPro project we developed advanced fusion approaches based on machine learning techniques. As proof of concept, we present two examples of newly developed methodologies for the integration of the AMUNDSEN hydro-climatological model outputs with a) the MODIS snow cover product to improve the estimation of snow cover area on a regional scale; b) the high-resolution Landsat and Sentinel-2 snow cover products to better estimate the glacier mass balance. The snow cover fusion method is tested on the alpine Euregio region of Tyrol, South Tyrol and Trentino giving an improvement of the estimation accuracy of 6% for the MODIS and 4% for the AMUNDSEN model snow cover products with respect to in-situ snow depth measurements. The glacier mass balance fusion method is validated on three glaciers in Tyrol (i.e. Varnagt, Kasselwand, Hintereis glaciers) giving an improvement of 14% in terms of water equivalent root mean square error between the model simulations and the in-situ glacier mass balance measurements.
With the proposed integration concept of hydrological model outputs and remote sensing products we set the basis for an innovative operational monitoring approach of the cryosphere in mountain regions.

References:
Beniston, M. (2012a): Impacts of climatic change on water and associated economic activities in the Swiss Alps. Journal of Hydrology, 412-413, 291–296. doi:10.1016/j.jhydrol.2010.06.046
Beniston M (2012b): Is the snow in the Alps receding or disappearing? WIREs Clim Change, 3, 349- 358, doi: 10.1002/wcc.179.
Dalponte, M., Bruzzone, L. and Gianelle, D. (2008): Fusion of hyperspectral and LIDAR remote sensing data for the classification of complex forest areas, IEEE Transactions on Geoscience and Remote Sensing, Vol. 46, No. 5, May 2008, pp. 1416-1427.
Gaudard, L., Romerio, F., Dalla Valle, F., Gorret, R., Maran, S., Ravazzani, G., Stoffel, M. and Volonterio, M. (2014): Climate change impacts on hydropower in the Swiss and Italian Alps. The Science of the Total Environment, 493, 1211–21. doi:10.1016/j.scitotenv.2013.10.012
Pasolli, L., Bruzzone, L. and Notarnicola, C. (2011): A novel hybrid approach to the estimation of biophysical parameters from remotely sensed data, Proc. of the IEEE-International Geoscience and Remote Sensing Symposium.
Rixen, C., Teich, M., Lardelli, C., and Gallati, D. (2011): Winter tourism and climate change in the Alps: an assessment of resource consumption, snow reliability, and future snowmaking potential. Mountain Research and Development, 31(3), 229–236. doi:10.1659/MRD-JOURNAL-D-10-00112.1

Area D - Board 463

Precipitation in Antarctica is poorly known but it plays an essential role in the context of current global warming and sea-level rise. The APRES3 project (Antarctic Precipitation, Remote Sensing from Surface and Space) deployed a battery of radar instruments during the Antarctic 2015/2016 summer campaign that provided unprecedented solid precipitation data over the French Dumont d'Urville station. This precipitation is characterized by the sublimation of snow in the lower layers of the atmosphere [Grazioli et al., 2016; Grazioli et al., 2017], but also by a complex vertical structure poorly represented by current climate models. These aspects must be understood to improve our knowledge of Antarctica's climate and thus to obtain reliable estimates of snowfall over coastal areas.

Apart from these field data, CloudSat observations are spatial remote sensing data, the only ones that observes solid precipitation at continental scale (North of 82°S). All snowfall events recorded by both CloudSat and ground radars were compared, and it has allowed the estimation of uncertainties on the satellite observations to be reduced from nearly 200 % to less than 25 % [Lemonnier et al., 2018]. This has opened up the opportunity to use this new dataset with more trust and confidence. Following recent studies [Palerme et al., 2014; Lemonnier et al., 2018; Souverijns et al., 2018], CloudSat snowfall retrievals were computed on a 3D grid with a 1° of Latitude per 2° of Longitude, and a 240 m vertical bin, representing the first three-dimensional climatology of the Antarctic precipitation [Lemonnier et al., 2018, in preparation].

Using this new dataset and the LMD/IPSL climate model, we simulated events with a zoomed configuration over the Dumont d'Urville area and a resolution of 25 km. Our simulations predict that precipitation sublimation becomes dominant at an altitude too high compared to radar observations. In addition, the sublimation process of the modelled snowfall is not as strong as in the observations. We also analyzed the precipitation pattern across the Antarctic continent obtained with the CloudSat satellite, and found that the LMD/IPSL model generates precipitation at too high altitude in the atmospheric column, underestimates snowfall amounts, and overestimates the occurrence of snowfall events. In this study, we will present the results of spatial remote sensing observations as well as the comparisons between our simulations and CloudSat data. We will finally discuss how the current CloudSat observations are being used to improve the physics of our model in the polar regions, and the future observations that could be useful to climate modelers.

References

Grazioli, J., Genthon, C., Boudevillain, B., Duran-Alarcon, C., Del Guasta, M., Madeleine, J. B., & Berne, A. (2017). Measurements of precipitation in Dumont d'Urville, Adélie Land, East Antarctica. The Cryosphere, 11(4), 1797-1811.

Grazioli, J., Madeleine, J. B., Gallée, H., Forbes, R. M., Genthon, C., Krinner, G., & Berne, A. (2017). Katabatic winds diminish precipitation contribution to the Antarctic ice mass balance. Proceedings of the National Academy of Sciences, 114(41), 10858-10863.

Lemonnier, F., Madeleine, J.B., Claud, C., Genthon, C., Duran-Alarcon, C., Palerme, C., Berne, A., Souverijns, N., van Lipzig, N., Gorodetskaya, I., L'Ecuyer, T., & Wood, N. (2018). Evaluation of CloudSat snowfall rate profiles by a comparison with in-situ micro rain radars observations in East Antarctica. Under submission in The Cryosphere.

Lemonnier, F., Madeleine, J.B., Claud, C., Genthon, C., Palerme, C., L'Ecuyer, T., & Wood, N. (2018). Studying Antarctic vertical structure of precipitation and its associated microphysical properties by using CloudSat snowfall retrieval. Publication in progress.

Palerme, C., Kay, J. E., Genthon, C., L'Ecuyer, T., Wood, N. B., & Claud, C. (2014). How much snow falls on the Antarctic ice sheet?. The Cryosphere, 8(4), 1577-1587.

Souverijns, N., Gossart, A., Lhermitte, S., Gorodetskaya, I. V., Grazioli, J., Berne, A., Duran-Alarcon, C., Boudevillain, B., Genthon, C., Scarchilli, C., & van Lipzig, N. P. M.: Evaluation of the CloudSat surface snowfall product over Antarctica using ground-based precipitation radars, The Cryosphere Discussions, 2018, 1–21, https://doi.org/10.5194/tc-2018-111, https://www.the-cryosphere-discuss.net/tc-2018-111/, 2018b.

Area D - Board 464

The extent and changes of snow and ice are key factors playing a role in climate change. Due to its high reflectivity w.t.t solar radiation, it has high albedo and bidirectional reflectance. Changes in snow have significant climate feedbacks. Consequently, an accurate knowledge of snow bidirectional reflectance is of fundamental importance for assessing climate change. In addition, a variety of satellite remote sensing applications involving snow covered surface (such as retrieving atmospheric variables over Polar Regions) require precise modeling of snow reflectance to decouple the highly reflecting surface from atmospheric scattering in the Top Of Atmosphere (TOA). Uncertainties in our knowledge of the microphysical and macroscopic properties of snow are an unresolved issue for radiative transfer, ray-tracing and chemistry climate models [1, 2].
In this study, to improve our understanding of snow reflectance properties we derived and analyzed Bidirectional Reflectance Factor (BRF) and the Bidirectional reflectance Distribution Function (BRDF) values from measurements made by the spring 2008 Arctic Research of the Composition of the Troposphere from Aircraft and Satellite (ARCTAS) campaign conducted at Barrow, Alaska [1]. These measurements were acquired by the Cloud Absorption Radiometer (CAR) on the NASA P-3B aircraft. The unique design of CAR provides measurements in fourteen spectral bands between 0.3 and 2.30 μm. The broad spectral range covering important wavelengths for remote sensing applications such as aerosol retrieval is of great importance for further application of the results of this study. The high spatial resolution of 1° in both viewing and azimuth angles allowed us to estimate the anisotropy of snow BRF with high accuracy. Besides, measurements obtained at different flight altitudes (~0.2-2 km) provided an opportunity to test and investigate atmospheric parameters in the model used to be compared with measured snow BRF and BRDF. To assess the performance and accuracy of the Radiative Transfer Model (RTM) SCIATRAN in simulating snow reflectance, the properties and changes of snow BRF with wavelength (from UV to shortwave infrared region), sun-satellite geometry and physical properties of snow (such as snow grain size and shape) have been investigated. The results of the SCIATRAN simulation are compared with CAR measurement. The accuracy of the SCIATRAN RTM in simulating snow BRF, including both cases of good agreement with airborne measurements but also discrepancies will be presented and discussed in detail.


This work was supported by the DFG funded Transregio-project TR 172 “Arctic Amplification (AC)3“.

References
[1] Lyapustin, A., Gatebe, C. K., Kahn, R., Brandt, R., Redemann, J., Russell, P., King, M. D., Pedersen, C. A., Gerland, S., Poudyal, R., Marshak, A., Wang, Y., Schaaf, C., Hall, D., and Kokhanovsky, A.: Analysis of snow bidirectional reflectance from ARCTAS Spring-2008 Campaign, Atmos. Chem. Phys., 10, 4359-4375, https://doi.org/10.5194/acp-10-4359-2010, (2010).
[2] Peltoniemi, J. I., Gritsevich, M., Hakala, T., Dagsson-Waldhauserová, P., Arnalds, Ó., Anttila, K., Hannula, H.-R., Kivekäs, N., Lihavainen, H., Meinander, O., Svensson, J., Virkkula, A., and de Leeuw, G.: Soot on Snow experiment: bidirectional reflectance factor measurements of contaminated snow, The Cryosphere, 9, 2323-2337, https://doi.org/10.5194/tc-9-2323-2015, (2015).

Area D - Board 465

In recent years, in the framework of ESA studies, several campaigns were carried out with the purpose of exploiting the capabilities of multi-frequency SAR data to derive snow information. In this paper we present the results obtained from the ESA SnowSAR airborne instrument on different test areas, selected as representative of different snow covered areas, including boreal forest, tundra and alpine environment, with the aim of exploiting the SnowSAR capabilities in estimating the snow parameters, namely Snow Depth (SD) and Snow Water Equivalent (SWE).
SnowSAR [1] is a side looking, dual polarized (VV/VH), X/Ku band SAR, funded by ESA, that acquired data during several airborne campaigns in Northern Finland, Austrian Alps, Northern Canada and Alaska, from 2011 to 2013.
The analysis of the SnowSAR backscattering sensitivity to SD/SWE, carried out for each test site in dry snow conditions, clearly pointed out that a general relationship between single frequency, single polarization backscattering (σ°) and the observed parameters of snow cannot be established, since many other factors concur to the total signal measured by SnowSAR. Although some increase of the measured σ° when SWE increases was found at both frequencies and polarizations, the spatial variations of soil background and the temporal variations of other snow parameters, mainly snow microstructure and snow density, are in general able to mask out completely this sensitivity.
Model simulations obtained using the Dense Medium Radiative Transfer (DMRT) model [2] also confirmed these behaviors. Moreover, both model simulations and experimental results demonstrated that signal dynamic due to SWE variations is generally smaller than the one produced by the spatial variation of soil background signal, which has the dominant effect, especially at X band.
However, each soil and snow parameter affects differently the measured backscattering at different frequencies and polarizations. This fact suggested exploiting the synergy of the available polarizations and frequencies from SnowSAR dataset for addressing the retrieval.
In this respect, the potential of machine learning applications and in particular of Artificial Neural Networks (ANN) was evaluated: ANN have been selected among other retrieval techniques because they can combine easily input data from different sources, thus being particularly suitable for merging the two frequencies and two polarization of SnowsSAR in the same retrieval algorithm [3-4].
While the backscattering is more related to SWE than SD, the present study was focused on SD retrieval because the amount of SWE data available from the in-situ measurements was not sufficient for completing successfully the ANN training and testing operations, while a large number of SD samples was available both from LiDAR and manual measurements.
It is well known that ANN algorithms need a proper training for obtaining satisfactory retrievals. In this case, the algorithm was trained on 50% of the entire dataset (i.e. over the three tests site) of SnowSAR acquisitions and corresponding in situ measurements of snow parameters, while the remaining 50% of data was used for validating the trained ANN, by predicting SD from SnowSAR acquisitions not involved in the training.
The possibility of having dedicated ANN for each test site was also evaluated: other ANN, have therefore been implemented, considering for training and validation only the data collected in boreal forests, tundra and Alps.
Finally, the possibility of using data simulated by electromagnetic models has been also evaluated, for overcoming the intrinsic limitation of experimental driven training [5]. A dataset simulated by DMRT was generated for the ≃800 points for which the mandatory SWE information was available, by running DMRT as single layer configuration, with fixed stickiness of 0.1, grain size variable between 0.1 and 2.5 mm and assuming the soil below snow as frozen. All the remaining DMRT inputs were derived from the experimental data.
This activity demonstrated that the algorithm can estimate SD by exploiting the synergy of SnowSAR acquisitions at two frequencies and two polarizations with R ~= 0.77, RMSE ~= 13 cm, and bias negligible. Better results have been obtained with the “dedicated” ANN for each test site, while the model driven ANN resulted in R≃0.74 and RMSE ≃12 cm. By improving the DMRT input parametrization and especially by better accounting for the background contribution of soil under snow, this approach might provide even more accurate results. This approach has the additional advantages of reducing the amount of in-situ data needed for training and of limiting the “site dependence” of the experimental driven training.
References
[1]. Di Leo D, Coccia A, Meta A (2015) Technical Assistance for the Development and Deployment of an X-and Ku-band MiniSAR Airborne System (SnowSAR). ESTEC No. 4000106761-CCN1. (https://earth.esa.int/web/guest/campaigns).
[2]. L. Tsang, J. Pan, D. Liang, Z. X. Li, D. Cline, and Y. H. Tan, 2007. Modeling active microwave remote sensing of snow using dense media radiative transfer (DMRT) theory with multiple scattering effects, IEEE Trans. Geosci. Remote Sens., 45, no. 4, pp. 990–1004.
[3]. Hornik K., M. Stinchcombe and H. White, “Multilayer feedforward network are universal approximators,” Neural Networks, vol. 2, no. 5, pp. 359–366, 1989.
[4]. Linden A.and J. Kinderman, “Inversion of multi-layer nets,” in Proc. Int. Joint Conf. Neural Networks, vol. 2, pp. 425–430.
[5]. Santi E., S. Paloscia, S. Pettinato and G. Fontanelli. “Application of artificial neural networks for the soil moisture retrieval from active and passive microwave spaceborne sensors,” Int. J. Appl. Earth Observ. Geoinf., vol 48, pp. 61–73, Jun. 2016.

Area D - Board 466

Changing the initial snow fields in models shows a clear impact on the modeled sea ice volume. Improving model estimates by using additional data has been a hot topic for the past years with sea ice concentration being routinely assimilated in operational system and more recently promising results shown also with assimilation and initialisation using sea ice thickness. For snow we did not find that many studies, which might be partly due to the lack of such data, which works as a motivation for our research. After all, snow is an important factor in the formation and evolution of sea ice as it insulates and controls the ice-atmosphere interface.

The novel snow products in question were created for the ESA Support To Science Element Arctic+ Improving observations and understanding of snow on Arctic sea ice project. Two types of snow products were created for this project. The first type is the Dual-altimeter Snow Thickness (DuST), where the method is based on different penetration depths of two altimeters, which enable the retrieval of snow depths. The second type is Snow on Drifting Sea Ice (SnoDSI), where the snow is handled through Lagrangian ice parcel tracking by a sea ice model. These products in turn replace the original snow fields of which one is a result of NCEP Reanalysis-2 atmospheric reanalysis data (for CICE) and the other is a result of ER40+ERA-Interim atmospheric forcing driving model hindcast from 1958 until the beginning of each model run (for NEMO-LIM3).

We run two models with the two snow products by replacing the original snow field with the Arctic+ product snow. One of the models is CICE, a dynamic-thermodynamic sea ice model, designed to be included in a global climate model. The other model is NEMO-LIM3, a coupled ocean-ice global model, also designed to be included in a global climate model. The snow depths were perturbed in November 15th and the model was let run for a full year after that. These steps were repeated for 2010-2014 for SnoDSI and 2013-2015 for DuST, according to the product availability.

The Arctic+ snow depths were in general higher than the model snow for the starting time of the run. This resulted in decreases in sea ice volume, already during the first month of the run when compared with the reference runs. Both models showed that thicker snow resulted in lower sea ice volumes throughout the year and that the change to reference run was in general greatest on the Russian side of the Arctic. The results were validated against sea-ice thickness observations such as the ones obtained from Upward Looking Sonar (ULS) and Airborne Electromagnetic (AEM) flights. If the novel snow products capture the actual truth better than the original product, and the actual sea ice volume is thinner than current estimates, this could mean the ice is changing more rapidly than currently predicted.

Area D - Board 467

Snow depth at the top of sea ice is a key parameter of the climate change due to its isolation and albedo properties, which condition the sea ice growth and melt, the heat absorption and the primary production under the ice. Moreover, for several reasons that will be described in this presentation, the lack of knowledge about the snow depth can impact sea ice thickness retrieval using altimetry with an error that can reach up to 100% in the worst cases.

Nevertheless, there exists currently no reliable snow depth product over Arctic and Austral sea ice. Indeed, the Warren (1999) climatology, frequently used to convert freeboard to sea ice thickness, has been built with data obtained decades ago, before the first impacts of the climate changes, and the meteorological re-analyses fail to faithfully reproduce snow falls in polar regions.

A study published in [Guerreiro et al., 2016] has shown that the difference of the scattering properties of the Ku-band and the Ka-band can provide a good proxy of the snow depth using an adapted processing chain, Alti Snow Depth (ASD). As a first demonstrator, the ASD chain has been applied on the Ka measurements, issued from Saral/AltiKa, a French-Indian satellite, and on the corresponding Ku measurements, issued from the European satellite altimeter CryoSat-2/SIRAL. These two altimeters are based on different technics and the ratio between the surface areas illuminated by each radar is about one order of magnitude.

In order to make the AltiKa LRM measurements comparable to the CryoSat-2 SAR measurements, these last ones have been pre-processed by the CNES in a degraded PLRM mode. The evaluation of the ASD product in regards with the Operation Ice Bridge snow radar data shows a good correlation (R=0.67), much better than all previous solutions. Nevertheless the results were limited to two winter campaigns.

This talk will present the results for the period extended to the five CryoSat-2/Saral common years (2013-2018) and to Austral sea ice, the GOP product dedicated to ocean from ESA. Their performance in regards with several in-situ datasets will be analysed.

Acknowledgement: this work has been supported by the CryoSeaNICE ESA project and TOSCA SICKAyS CNES project.

Area D - Board 468

Following the launch of NASA’s ICESat-2 laser altimeter, it will be possible to obtain basin-scale, time-varying snow depth estimates based on the difference between ICESat-2 total (snow + sea ice) freeboard measurement and ESA’s CryoSat-2 (CS-2) radar altimetry estimate of sea ice freeboard. The methodology is proven feasible at cold snow temperatures (Kwok and Markus, 2018). However, a complication arises for first-year sea ice (FYI). Due to salinity in the basal snow layers on FYI, radar scattering could occur above the snow–ice interface, which is potentially a significant source of error for sea ice thickness estimation and thus snow depth estimation (Kwok and Kacimi, 2018). We propose to implement a snow depth retrieval scheme over Arctic FYI that explicitly accounts for snow salinity.

During FYI formation, notable amount of brine is expelled upwards, forming a thin surface brine layer. When snow accumulates on the FYI this brine is wicked up into the overlying snow cover, producing brine-wetted snow layers. These occur predominantly within the bottom 6 to 8 cm of the snow cover. Snow salinity in these layers commonly range from 1 to 20 parts per thousand (Barber et al., 1995; Crocker, 1992; Geldsetzer et al., 2009). For snow depths of less than 8 cm, the snow cover is often completely brine-wetted (Nandan et al., 2017). Due to their higher dielectric properties, brine-wetted snow layers can cause an upward shift in the Ku-band radar main scattering horizon, particularly at snow temperatures only a few degrees below zero (Geldsetzer et al., 2009; Nandan et al., 2017). This is contrary to the common assumption that Ku-band microwaves penetrate the entire snow cover and are scattered at the snow-sea ice interface, which forms the operational basis of sea ice freeboard measurement by CS-2.

The feasibility and accuracy of our proposed snow depth retrieval method depends on reliable measurements of total and ice-only freeboards by ICESat-2 and CS-2, respectively. Although several snow properties have been reported to impact CS-2 signal penetration (Willat et al., 2011; Laxon et al., 2013; Kwok, 2014), the effects of snow salinity have only recently received attention (Nandan et al., 2017). At present, the role of snow salinity has yet to be explicitly considered in current retrieval models of sea ice freeboard. Our approach will address the effect of varying snow volume properties on microwave penetration depth at CS-2 frequencies, in order to obtain more accurate sea ice thickness estimates and thus enable reliable snow depth estimates. We will apply penetration depth correction to CS-2 sea ice freeboard data for FYI. The degree of the correction will be determined using a priori modelled snow depth estimates from the Regional Ice Ocean Prediction System (Dupont et al., 2016). Validation uses in situ snow measurements from the Canadian Arctic Archipelago. It is anticipated that our method will be applicable to all FYI in the Arctic.


Reference

D. G. Barber, S. P. Reddan, and E. F. Ledrew. Statistical characterization of the geophysical and electrical properties of snow on landfast first-year sea ice. Journal of Geophysical Research, 100(C2):2673–2686, 1995.

G. Crocker. Observations of the snowcover on sea ice in the Gulf of Bothnia. International Journal of Remote Sensing, 13(13):2433–2445, 1992.

F. Dupont, A. Caya, and J.-F. Lemieux. Regional Ice Ocean Prediction System: pseudo-analysis and forecast components. Technical Note. Environment and Climate Change Canada, 2016.

T. Geldsetzer, A. Langlois, and J. Yackel. Dielectric properties of brine-wetted snow on first-year sea ice. Cold Regions Science and Technology, 58:47–56, 2009.

R. Kwok. Simulated effects of a snow layer on retrieval of CryoSat-2 sea ice freeboard. Geophysical Research Letters, 41(14):5014–5020, 2014.

R. Kwok and S. Kacimi. Three years of sea ice freeboard, snow depth, and ice thickness of the Weddell Sea from Operation IceBridge and CryoSat-2. Cryosphere, 12:2789–2801, 2018.

R. Kwok and T. Markus. Potential basin-scale estimates of Arctic snow depth with sea ice freeboards from CryoSat-2 and ICESat-2: An exploratory analysis. Advances in Space Research, 62(6):1243–1250, 2018.

S. W. Laxon, K. A. Giles, A. L. Ridout, D. J. Wingham, R. Willatt, R. Cullen, R. Kwok, A. Schweiger, J. Zhang, C. Haas, S. Hendricks, R. Krishfield, N. Kurtz, S. Farrell, and M. Davidson. CryoSat-2 estimates of Arctic sea ice thickness and volume. Geophysical Research Letters, 40(4):732–737, 2013.

V. Nandan, T. Geldsetzer, J. Yackel, M. Mahmud, R. Scharien, S. Howell, J. King, R. Ricker, and B. Else. Effect of snow salinity on CryoSat-2 Arctic first-year sea ice freeboard measurements. Geophysical Research Letters, 44(20):10419–10426, 2017.

R. Willatt, S. Laxon, K. Giles, R. Cullen, C. Haas, and V. Helm. Ku-band radar penetration into snow cover on Arctic sea ice using airborne data. Annals of Glaciology, 52(57):197–205, 2011.

Area D - Board 469