Northern hemisphere Fractional Snow mapping with VIIRS - First experiments in ESA DUE GlobSnow-2 Sari Metsämäki 1), Kari Luojus 2), Jouni Pulliainen 2), Mwaba Hiltunen 2), Andreas Wiesmann 3) 1) Finnish Environment Institute, SYKE 2) Finnish Meteorological Institute, FMI 3) Gamma Remote Sensing 7 th EARSeL Workshop on Remote Sensing of Land Ice & Snow 3-6 February, 2014 Bern, Switzerland

 AATSR is not operating any more  NPP Suomi/VIIRS employed in GlobSnow NRT SE-production  Gap-filler for Sentinel-3  Northern Hemisphere covered daily (2013 onwards)  Processing software by Gamma, operated in NRT at FMI, Sodankylä  FSC retrieval algorithm (SCAmod) and auxiliary data analogous to GlobSnow SE from ATSR-2/AATSR  VIIRS band 550nm (+ 1.6 µm for NDSI) employed: corresponding to AATSR/ATSR-2  Cloud masking with SYKE-made algorithm SCDA 2.0, employing BT3.7, BT11, BT12, R550, R1.6 GlobSnow SE-product from VIIRS

 Developed for GlobSnow FSC purposes; works for ATSR-2 /AATSR, MODIS and VIIRS  No need to identify all (small/semi-transparent) clouds in confident snow-free areas  Cloud screening is targeted, not cloud classification  binary information, no classes, no propabilities  Must be relatively simple and computationally fast  Must be applicable for Northern Hemisphere, no regional/local tuning  Must not confuse between fractional snow and clouds (it’s the fractional snow we want to see!)  Should work particularly over seasonally snow-covered areas and throughout the potential snow season Cloud Detection SCDA2.0 for VIIRS

 Simple Cloud Detection Algorithm  Designed to work with only a few spectral bands  common for Terra/MODIS, Envisat/AATSR, ERS- 2/ATSR-2, NPP Suomi/VIIRS  R550, R1.6, BT3.7, BT11, BT12  Based on empirically determined thresholds for single bands and their ratios  Driven by BT11-BT3.7  Ratio NDSI / R550 important in avoiding false cloud commissions  Several other test for BT12, R550 and NDSI Major features of SCDA2.0

28 Feb, 2003

Cloud Snow-covered forest and tundra

AATSR cloud (operational)

Cloud Snow-covered forest and tundra

MODIS cloud (from MOD10_L2)

Cloud Snow-covered forest and tundra

GlobSnow-2 cloud SCDA v2.0

Weather stations and snow courses in Finland.  FSC from the three 8-days period plus one day in September compared with in situ Snow Depth  First binary snow data is generated:  IF Snow Depth > 1cm  ’snow’  IF FSC>0  ’snow’ OR  IF FSC>0.15  ’snow’  Statistical measures provided  Hit Rate, Probability of detection, False alarm rate, Kuiper skill score VIIRS SE-product vs. in situ Snow depth

Threshold for assigning the pixel as ‘snow’ : FSC >=0.15 (15%). From WS: snow depth >1 cm, N=1418 estim. in-situ snowNon- snow #in situ PODFARHRKSS Snow Snow Non-snow Non- snow Threshold for assigning the pixel as ‘snow’: FSC > 0. From WS: snow depth >1 cm, N=1418 estim. in-situ snowNon- snow #in situ PODFARHRKSS Snow Snow Non-snow Non- snow VIIRS SE-product vs. WS Snow depth

Comparison against Landsat-8 data  VIIRS Fractional snow compared with Landsat fractional snow  RMSE, bias  Data set preparation in progress

Motivation: MOD10C2 commonly used  interesting to see what new can VIIRS and SCAmod bring into NH Fractional snow mapping VIIRS provides good gap-filler data for Sentinel-3 SLSTR Earlier: GlobSnow SE (AATSR-based) difficult to evaluate due to data gaps Problems with weekly and monthly product: artifacts due to very limited number of observations Now: VIIRS provides global coverage  possible to produce 8- days composites comparable with MOD10C2 Demonstration for three 8-day periods in melting season 2013 Doy (March30 - April06) Doy (April15 - April22) Doy (May01 - May08) VIIRS FSC vs. MOD10C2

VIIRS SE March 30 – April

MOD10C2 FSC in Climate modeling grid (CMG, 0.05º× 0.05º) based on binning the 500m 8-day maximum snow cover (MOD10A2) to the CMG MOD10A2 based on binary algorithm  FSC biased at the transitional zone? VIIRS FSC vs. MOD10C2

GS-2 VIIRS FSC NASA MOD10C2

 MOD10C2 and GlosSnow SE mostly agree well: o RMSE 12% (first period) to 15% (last period)  MOD10C2 shows less snow for forest areas, even 10% less for % forest-covered areas VIIRS FSC vs. MOD10C2

 Distinctive difference between GS SE and MOD10C2: width of the melting zone MOD10C2 based on aggregation of binary snow data  FSC is also ’binarized’ i.e. underestimations of low snow fractions and overestimations of high snow fractions Snow ablation during spring VIIRS FSC vs. MOD10C2

Fusion of GlobSnow SE and SWE GlobSnow SWE NRT-product may have difficulties in detecting snow- free areas. Solution: snow line identification from SE-product Will be in use in 2014 SWE-production