Snow Cover in Europe derived from historical AVHRR Data - a TIMELINE thematic processor

Abstract

The AVHRR (Advanced Very High Resolution Radiometer) mission offers a data set that goes back continuously to the early 1980s. These data are processed and examined as part of the TIMELINE (TIme series processing of Medium resolution Earth observation data assessing Long-term dynamics In our Natural Environment) project, which focuses on a 30-year homogenized time series of NOAA and METOP-AVHRR data from Europe and North Africa. This time series is generated using the historical data archive of the German Remote Sensing Data Center (DFD) at the German Aerospace Center (DLR), which reaches back until 1981. The main goal of TIMELINE is to develop consistent, reproducible, transparent, reliable and generic geoscientific variables for research related to global change. These outputs are used to answer climate-related questions, enable the detection of changes and identify geoscientific phenomena and trends. A wide range of geoscientific products is generated within TIMELINE and made available to the public via a free and open data policy. The processing of these products follows a sophisticated sequence of individual steps, including preprocessing (calibration, chip matching, orthorectification), atmospheric and BRDF correction, cloud masking and the generation of geophysical and thematic products on level 2 and level 3. All processing steps are based on solid operational algorithms and are thoroughly documented and validated. One of these thematic products is the snow cover, which is calculated on the basis of the atmospherically corrected Level 1B data. Snow has the largest spatial extent of the entire cryosphere, but at the same time also has the greatest seasonality. To determine the water content of the snow cover, passive microwave remote sensing was used early on, but for a precise investigation of the variability of the spatial extent of the snow cover, its geometric resolution is insufficient. By using multispectral sensors with a higher geometrical resolution, one takes advantage of an important property of snow: Snow has a very high reflection in the visible spectral range (VIS) and a very low reflection in the short-wave infrared (SWIR). This property is used in the Normalized Difference Snow Index (NDSI), whereby the difference between the reflection of VIS and SWIR is divided by their sum. The index can have a value between -1 and +1, whereby an NDSI greater than 0.4 stands for a snow cover of over 50%. In areas with dense vegetation cover, the NDSI can adapt lower values (up to 0.1) even when the ground is completely covered with snow; these are determined by the additional calculation of the Normalized Difference Vegetation Index (NDVI). In the TIMELINE project, AVHRR bands 1 and 3A are used for the NDSI calculation, AVHRR bands 1 and 2 for NDVI. If only band 3B exists, the reflective component is calculated from the radiation measurements and an artificial band 3A is formed. The NDSI is also suitable for distinguishing between snow and clouds, but fails when it comes to ice clouds. Therefore, ERA5 skin temperature data is included and the temperature difference is used as a differentiation criterion. The resulting level 2 product contains a thematic snow classification with an associated quality layer for each AVHRR scene in swath projection. The Level 2 products are first brought into the TIMELINE projection (ETRS89-extended / LAEA Europe) and divided into 4 tiles in order to create daily composites. The pixel-based value assignment is based on the quality layers. The percentage of snow cover over 10 days and per month is then calculated from the daily products (except for cloud-covered pixels). For better comparability with the 8-day MODIS snow product, an aggregated 8-day snow cover is also calculated. The daily data is also used to obtain cloud-free images using the DLR Global SnowPack processor. From this, the snow cover duration, the beginning and end of a snow cover season can be calculated for each pixel. This is used for trend analysis of the entire time series and for assessing the accuracy with the help of the MODIS product, which has existed since 2000.