Seasonal snow is an important component of the global climate system. It is highly variable in space and time and sensitive to short term synoptic scale processes and long-term climate-induced changes to temperature and precipitation. Current snow products derived from different algorithms applied to various satellite data show significant discrepancies in extent and snow mass, a major source of uncertainty for monitoring and climate model verification. The ESA CCI+ Programme addresses seasonal snow as one of nine Essential Climate Variables derived from satellite data.
In the first phase of the snow_cci project (2018 - 2021), reliable and fully validated processing lines for the generation of snow climate data records were developed and implemented. Homogeneous multi-sensor time series of daily snow extent and snow water equivalent with global coverage were generated. Using GCOS guidelines, the product requirements for these parameters are assessed and consolidated using workshops and other engagement with users dealing with different climate applications. The retrieval algorithms for fractional snow extent provide consistent daily products for snow viewable from space (viewable snow) and snow on the surface corrected for forest masking (snow on ground) with global coverage. Input data are medium resolution optical satellite imagery (AVHRR-2/3, A/ATSR-2, MODIS, SLSTR) from 1982 to present. For the snow_cci Climate Research Data Package version 2, an iterative development cycle was implemented to improve and homogenise the snow extent products from different sensors. Independent validation of the snow extent products is performed using high resolution snow maps from Landsat and Sentinel-2 acquired across different seasons and climate zones around the globe from 1985 onwards as well as in-situ snow data following protocols developed within the snow community. Time series of global (non-mountain) daily snow water equivalent (SWE) products are generated from passive microwave data from SMMR, SSM/I, and AMSR from 1979 onwards combined with in-situ snow depth measurements. Long-term stability and quality of the product is assessed using independent snow survey data and by intercomparison with snow mass information from global land surface models.
We will present an overview of the algorithms and systems for generation of snow_cci snow products available at the ESA Open data portal. The 40 years (from 1982 onwards) time series of daily fractional snow extent products from AVHRR with 5 km pixel spacing, and the 20 years timeseries from MODIS (from 2000 onwards, 1 km pixel spacing) as well as the coarse resolution (12.5 km pixel spacing) daily SWE products from 1979 onwards will be presented along with the results of the multi-sensor consistency and validation activities and inter-comparisons with snow products from other sources. The impact of the snow_cci products within the use cases carried out by the Climate Research Group on long term snow extent and mass trends, evaluation of CMIP-6 experiments, and the simulation of long-term changes in Arctic hydrological regimes will be summarized.

Reliable information on snow cover across the Northern Hemisphere and Arctic and sub-Arctic regions is needed for climate monitoring. Warming surface temperatures during the recent decades have driven a substantial reduction in the extent and duration of Northern Hemisphere snow cover. These changes in snow cover affect Earth’s climate system via the surface energy budget and influence freshwater resources across a large proportion of the Northern Hemisphere. In contrast with snow extent, reliable quantitative knowledge on seasonal snow mass and its trend has been lacking.
The ESA Snow CCI project initiated in 2018 strives to further improve the retrieval methodologies for snow cover extent (SCE) and snow water equivalent (SWE) from satellite data and construct long term climate data records (CDRs) of terrestrial snow cover for climate research purposes.
The efforts to improve satellite-based retrieval of snow water equivalent has resulted in the first enhanced resolution SWE record spanning 1979-2020, with 12.5km spatial resolution, also available in 0.10 lat/lon geographic grid. The retrieval applies the GlobSnow approach which combines satellite-based data with ground-based snow depth observations. The new improved dataset will be published in late 2021 as the Snow CCI v2 SWE data record, incorporating dynamic snow density with the improved spatial resolution.
Further, we present an updated bias-correction consideration for the 40+ years climate record of Northern Hemisphere snow mass, incorporating the methodology presented in Pulliainen et al. 2020. The new bias-correction approach includes an increased number of ground-based snow transect data. The bias-corrected data is made for monthly average SWE estimates.
The new SWE data record improves our estimates of the satellite-era snow mass changes and trends for the Northern Hemisphere.

References:
Pulliainen, J., Luojus, K., Derksen, C., Mudryk, L., Lemmetyinen, J., Salminen, M., Ikonen, J., Takala, M., Cohen, J., Smolander, T. and Norberg, J., “Patterns and trends of Northern Hemisphere snow mass from 1980 to 2018”. Nature 581, 294–298 (2020). https://doi.org/10.1038/s41586-020-2258-0.

Snow cover in mountainous areas is one of the most important variables related to biodiversity and ecosystems processes. Optical satellite imagery represents a useful support for snow cover mapping since it allows to evaluate snow cover extent in vast and inaccessible areas in conjugation with an almost continuous temporal coverage. At present, several operational snow cover algorithms and products are available. Although most of them offer an up-to-daily time scale, their spatial resolution is not adequate for studies requiring high spatial detail. Conversely, high resolution products offer the possibility of precise studies, especially in areas with complex topography, but suffer from temporal limitations due to infrequent acquisitions, also worsened by cloud cover. This compromise between high temporal and spatial resolution considerably limits snow cover applications. To address this lack, in this study, we propose a two-stage approach, for daily high resolution snow cover maps retrieval from Sentinel-2. In the first step, we revised the Let-It-Snow (LIS) algorithm, which performs high resolution snow detection based on the normalized difference snow index (NDSI) and a Digital Elevation Model (DEM) to identify the altitudinal height of the presence of snow, below which its presence is excluded, by introducing a new parameter, a threshold in the short wave infrared (SWIR) band, and modifying the overall workflow of the algorithm, in particular in the image pre-processing phase. The revised algorithm has been applied to different mountainous study areas across Europe. Both the original and the modified algorithms have been validated by considering snow data collected by meteorological stations equipped with snow gauges. Moreover, for the Gran Paradiso National Park, we also tested the performances in the presence of cloud cover, in order to evaluate their abilities to reduce the probability of cloud and snow cover misclassification, by comparison with station ground data collected by both snow gauges and solarimeters. The changes introduced in the revised algorithm result in an overall classification accuracy improvement. The output of the previous step will be, then, used as input for training a random forest algorithm to produce daily 20 m snow cover maps, through a site-specific approach based mainly on topographical features. The products will be validated using ground truth data collected daily by weather stations.

Snow cover and timing of snowmelt are key regulators of a wide range of ecosystem functions, e.g. catchment runoff, insulation of permafrost, photosynthetic activity and species distribution. Snow cover and melt are strongly influenced by the amplified Arctic and alpine warming and essential variables to understand environmental changes and their dynamics. The decrease of snow cover duration and extent has been documented for the northern hemisphere over the last 40 years, but changes on local scale do not indicate such clear trends due to high interannual variability and spatial heterogeneity of snow cover duration. Whereas in-situ observations are incapable to cover this high heterogeneity, current state-of-the-art remote sensing methods mostly rely on data from passive multispectral systems (e.g. MODIS, Landsat/Sentinel-2), which lack a sufficient spatio-temporal resolution in the Arctic due to the polar night and frequent cloud coverage. In contrast, weather and sun illumination independent Synthetic Aperture Radar (SAR) data (e.g. from Sentinel-1) has proven high potential to map snow cover depletion and to detect different evolutionary stages of snowmelt at high spatiotemporal resolution. In this contribution, we present a novel workflow relying on the analysis of Sentinel-1 time series data. The approach is based on established concepts and exploits the full Sentinel-1 SAR time series information by making use of the characteristic seasonal SAR backscatter pattern over snow. The tool is accessible via Google Earth Engine and provides weekly information about the state of the snowpack with a spatial resolution of 20 meter. We can derive the snow cover extent as well as seasonal maps of the melt end date. Furthermore, the approach is capable of identifying the extent and timing of the snowpack contributing to runoff as well as the length of different snowmelt phases. We validated the snow cover extent maps against high-resolution orthorectified time-lapse camera imagery at two sites in Greenland and got average accuracies of 90% with all maps reaching above 75% overall accuracy. We will present the results from these two sites as well as novel results from other regions.

Measuring the amount of seasonal snow cover is of paramount importance for a large variety of applications in the cryosphere, greatly influencing energy, water, and bio/geochemical cycling in mountainous ecosystems. Particularly relevant is seasonal snow depth, which is typically performed through in-situ measurements at selected sites. But manual sampling of snow depth at large scale is time-consuming and potentially dangerous for field operators. Remote sensing represents therefore a valuable solution to overcome such limitations and several techniques have been developed in recent years for estimating seasonal snow depth from lidar and Synthetic Aperture Radar (SAR) sensors. Specifically, airborne lidar sensors are able to measure the height of the surface with high accuracy and spatial resolution. Snow depth can be retrieved by differentiating surface models acquired at snow-free and snow-covered conditions (Deems et al., 2013). Although very accurately, spaceborne imaging lidar sensors required for regular large-scale monitoring of snow accumulation are still not available. In the AlpSnow project (2020-2022) we are investigating the suitability and limitations of single-pass bistatic TanDEM-X acquisitions to map the accumulation of seasonal snow on Alpine glaciers and high alpine areas with gentle topography building on the work by Leinss et al (2018).

In order to map snow depth, we are differentiating two coregistered TanDEM-X-derived digital elevation models (DEM), acquired with and without snow cover. Critical aspects to be carefully considered are the penetration of radar waves into the snow pack, the imaging geometry, and the snow conditions. At X-Band the penetration of the radar signal is low for bare ice and wet snow surfaces, while it can reach several meters in presence of dry snow (Rizzoli et al., 2017). This last case would result in a considerable bias within the estimation and has therefore to be excluded from the analysis.

The method to map snow accumulation consists in differentiating two DEMs from Bistatic Tandem-X data, acquired at snow free conditions in late summer, when bare ice is still visible on the glacier’s surface in the ablation area and only wet snow is present at higher altitudes, and the second one acquired during spring time, in presence of wet snow on top of the glacier. The main challenges to be tackled are: signal penetration into the surface, single-DEM accuracy, and mutual DEM calibration. Radar signal penetration primarily depends on the surface type and conditions and the SAR imaging geometry. In order to avoid penetration biases in the DEM generation we are properly selecting the dates of image acquisition. Another aspect is the strong topography in the Alps, which leads to significant geometric distortions in SAR images and might cause phase unwrapping errors, resulting in a lower DEM accuracy. Moreover, the low radar backscatter in presence of wet smooth snow surfaces results in a lower interferometric coherence and increases the phase noise in the interferogram. In a first step we mitigate these aspects by properly masking out regions affected by extreme topography and by radar shadow and layover, as well as by optimizing the interferometric processing and, in particular, the estimation of the interferometric phase. This can be done by either increasing the standard boxcar multi-looking, at the cost of a lower resolution, or by applying advanced denoising methods during the phase filtering process (Sica et al.; 2020).
An important step is the mutual calibration of the two DEMs which compensates for residual horizontal shifts and vertical offsets. This is solved, on the one hand, by performing a precise coregistration of the geocoded DEMs, and on the other hand, it requires the identification of reliable tie-points which can be used to vertically adjust the DEMs. Such points are required to show high interferometric coherence and to be stable in time. This operation is not trivial, given the fact that we are dealing with extremely difficult terrain and geometric distortions. To this end, we decided to select tie-points by relying on persistent scatterers candidates (PSC) which, in presence of high signal-to-noise ratio, can approximate well temporal stability in both amplitude and phase (Ferretti et al., 2001). Since no long-enough TanDEM-X time-series was available, we used a continuous time series of 1-year of Sentinel-1 data to identify PSC, which are mostly small villages and man-made structures in the valley floor and a few exposed solid rock boulders or ridges in high Alpine terrain. Given the similarity between X and C bands, we used the locations of PSC at C band for inter-calibration of the various TanDEM-X DEMs.
We will present the first preliminary results for Rofen valley in the Austrian Alps where in-situ measurements and regular terrestrial laser scanning data are available for validation. Preliminary results are promising and, given the availability of suitable data, could allow for an effective measurement of seasonal snow depth at least at regional scale. This aspect is also relevant in view of future bistatic/multi-static InSAR missions, such as the ESA Harmony Earth Explorer or the German High Resolution Wide Swath (HRWS)/Mirror-SAR.

Deems et al., 2013. Lidar measurement of snow depth: a review. Journal of Glaciology, 59(215), 467-479
Leinss et al. 2018. Wet snow depth from TanDEM-X single-pass InSAR DEM differencing, IEEE International Geoscience and Remote Sensing Symposium, 8500-8503.
Lievens et al., 2019. Snow depth variability in the Northern Hemisphere mountains observed from space. Nature Communications, 10(4629).
Rizzoli at al., 2017. Characterization of Snow Facies on the Greenland Ice Sheet Observed by TanDEM-X Interferometric SAR Data, Remote Sensing, 9(4), 315.
Sica et al. 2020. Φ-Net: Deep Residual Learning for InSAR Parameters Estimation, IEEE Transactions on Geoscience and Remote Sensing, 59(5), 3917-3941.

In the limited terrestrial ecosystems of Antarctica, all photosynthetic organisms will make a significant contribution to the ecology of their habitat. Blooms of snow algae are known to occur annually in Antarctica but neither their scale nor contribution have been quantified. We present the first estimate of the green snow algae community biomass and distribution along the Antarctic Peninsula and our work since to scale up observations across Antarctica. Using Sentinel-2 satellite imagery from 2017-2019 and field data collected over the same period, we identified 1679 blooms on the Peninsula, covering approximately 1.95km² of snow in a season. Field measurements enabled us to calculate that this represents a dry biomass of 1327 tonnes (479 tonnes of carbon). Spatial analysis suggests that snow algal range is limited to areas with average positive degree days in the austral summer, and that their distribution is strongly influenced by nutrient inputs from the ocean via marine vertebrates, with 60% of the blooms identified found within 5 km of penguin colonies. Our findings suggest a warming Antarctic Peninsula will likely to lose a majority green snow algae blooms, as 62% of these were on small islands with no high ground for upward range expansion. However, as bloom area and elevation were observed to increase towards the north of the Peninsula, we suggest a parallel expansion of blooms on larger landmasses, close to bird or seal colonies. This increase is predicted to outweigh significantly the biomass lost from small coastal blooms and result in a net increase in snow algae extent and biomass as the Peninsula warms.