Frozen Surface Classification Scheme for ATMS and GMI

Within the development of passive microwave precipitation retrieval techniques, and, in
particular, of snowfall detection and retrieval techniques, the possibility to characterize the
frozen background surface (snowcover and sea ice conditions) at the time of the overpass
appears to be a relevant task. As demonstrated by many recent studies (e.g., Tabkiri et al.,
2019, Ebtehaj and Kummerow 2017, Panegrossi et al., 2017), the microwave signal
related to snowfall is strongly influenced by the surface conditions, and the response of the
observed brightness temperatures to the presence and intensity of snowfall depends on
complex interconnections between environmental conditions (surface temperature, water
vapor content, snow water path, cloud depth, presence of supercooled droplets) and the
different surface conditions (wet or dry snow cover, sea ice concentration and type, etc.).
The use of surface classification climatological datasets results inadequate for the purpose
because of the extreme variability of the frozen surface conditions. It is therefore
necessary to be able to identify the background surface condition as close as possible (in
space and time) to that of the observation. The conically scanning GPM Microwave Imager
(GMI) and cross-track the Advanced Technology Microwave Sounder (ATMS) are the most
advanced currently available microwave radiometers. They are both equipped with
channels at several different frequencies that can be exploited both for the identification of
the frozen surface conditions and for snowfall detection and retrieval at the time of the
overpass over a precipitation event (i.e., Rysman et al., 2018). Moreover, they can be
used to analyze the potentials of future radiometers with similar characteristics such as the
EPS-SG Microwave Sounder (MWS) and Microwave Imager (MWI), which represent the
future in terms of European operational radiometers that can be exploited for precipitation
retrieval at all latitudes (including the Polar Regions). In the last years we have developed
two frozen surface classification schemes based on the use of GMI and ATMS low
frequency channels (from 10 GHz up to 36 GHz) and on ancillary near-surface
temperature and columnar water vapor data (obtained from ECMWF global ERA5
reanalysis). The algorithm is able to identify 9 classes of soil including different type of
snow and sea ice. The results of such classification have been compared with other
products, such as the NASA-GPROF soil type classification, and with snowcover and sea
ice global datasets (such as GMASI- Autosnow, and SNODAS from NOAA, and ECMWF
ERA5). In particular, the comparison with SNODAS over Northern America region shows
that the probability of detection of snow-covered surfaces varies between 86% - 98%
(79%-95%) for GMI (ATMS) with a relatively small false alarm ratio (10%-30%). The
analysis evidenced the main factors limiting the detection capability, such as the moisture
content, the presence of orography, the snow cover beam filling and the snow depth.