Prospects for assimilating microwave sounder
radiances over snow covered surfaces
Stephen J. English and
Patricia de Rosnay
ECMWF, Shinfield Park, Reading, UK.
AMSU Ch. 5 O–B and 50 GHz emissivity atlas at ECMWF
80°N
80°N
60°N
60°N
40°N
40°N
20°N
20°N
0°N
0°N
20°S
20°S
40°S
40°S
60°S
60°S
80°S
80°S
80°E
0.72 0.75 0.78 0.81 0.84 0.87 0.9
Land and sea ice 0.7-1.0
O-B K 1/2/13
120°E 160°E
160°W 120°W 80°W
0.93 0.96 0.99
0.7
60°N
60°N
40°N
40°N
20°N
20°N
0°N
0°N
20°S
20°S
40°S
40°S
60°S
60°S
80°S
80°S
-2.4 -1.8 -1.2 -0.6 -0.1
40°E
0.1
80°E
0.6
1.2
120°E 160°E
1.8
2.4
80°E
120°E 160°E
0.93 0.96 0.99
40°N
30°N
20°W
0°E
20°E
60°E
30°N
20°W
Emissivity 1/2/13
40°N
3
-3
40°W
0°E
-2.4 -1.8 -1.2 -0.6 -0.1
40°E
0.1
80°E
0.6
1.2
120°E 160°E
1.8
2.4
3
30°N
20°W
0°E
0.7
0.1
0.2
0.5
60°E
1
10
0.72
0.75
20°E
0.78
0.81
0.84
40°E
0.87
0.9
0.93
0.6
0.4
60°E
0.96
0
20
40
Snow density %
0.99
60-70N
48-58N
70-80N
70-80S
18-29N
5
0.85
0.8
0.85
0.8
0.75
0.75
0.7
0.7
200
0°N
0.10
20°S
400
Day No. from 1/1/12
600
0
Emissivity
0.06
60°S
4
80°S
40°W
0°E
40°E
80°E
120°E
0.9
0.9
0.85
0.8
0.75
0.75
600
0
0.75
0.8
0.7
200
400
Day No. from 1/1/12
00z
12z
0.94
0.95
0.8
0.94
0.93
0
200
400
Day No. from 1/1/12
600
0.06
Antarctica shows a small rise
in emissivities in the SH
summer period.
12z – 00z
0
–0.01
0.92
200
400
Day No. from 1/1/12
600
0.90
0
–0.01
Greenland shows the well documented
ice-cap melt event around day 200
in 2012. It is interesting to note that
emissivities fell and then took almost a
year to return to their pre-melt values.
Is it well understood why?
12z – 00z
0
–0.04
–0.06
–0.05
–0.08
0
200
400
Day No. from 1/1/12
600
Snow covered in East Europe (00z) for 2012–2013
1
50 GHz emissivity
0.95
0.9
0.85
0.8
0.8
0.85
0.9
0.95
24 GHz emissivity
1
0.95
0.95
50 GHz emissivity
1
0.9
0.85
0.8
0.8
0.85
0.9
0.95
24 GHz emissivity
1
1
0.9
0.85
0.8
0.8
0
200
400
Day No. from 1/1/12
–0.05
600
The emissivity falls steadily over several months, consistent with an increase in grain size
for a dry low density snow pack. These is also an increase in spatial variability, denoted by
the error bars which represent standard deviation of emissivity in the target area. This is
almost certainly due to a combination of vegetation cover and orography. At snow melt the
emissivities initially rise just above the snow-free value, prior to a quick return to a snow-free
Spectral emissivity variations between 24 and 89
GHz are plotted to the left for East Europe. These
show that snow cover introduces large spectral
gradients, which contain additional information
about the snow. In the absence of snow the
correlation in emissivity across the 24–89 GHz
range is very high meaning that there is only
one degree of freedom in the emissivity, and
therefore little information about the surface.
It is encouraging to think that if we can fully
understand the emissivities we may be able to
improve assimilation in these areas using physical
constraints, and also improve snow analyses.
400
Day No. from 1/1/12
600
There is no reason to expect an annual
cycle in emissivity for the Sahara.
At 00z there is almost no variation, but at
12z there is a large variation.
This is almost certainly an impact of model
Tskin errors. Can we trust Tskin in snow
regions anymore?
0
200
400
Day No. from 1/1/12
600
and almost invariant value. For Antarctica there is also an annual cycle with a brief midsummer rise in emissivity. During the polar night the 00z emissivities are higher, for reasons
which are not understood. At present its not straightforward to understand which changes
are real, and which are driven by biases in the system, such as poor skin temperature, cloud
screening and errors in background atmospheric fields.
microwave sounders. Poor estimation of surface emissivity
and skin temperature are a likely explanation for the high
This gives reason to believe biases in background skin
temperatures may bias the emissivity estimation.
3. Snow surfaces show strong spectral de-correlation in the
departures in these regions, despite the dynamic emissivity
estimated emissivities. This means different channels give
estimation technique used at ECMWF.
independent information.
2. There are day minus night differences in emissivity that
4. Collaboration between the land surface model and data
are difficult to explain and may be the result of several
assimilation community and the ATOVS assimilation
factors. This is also seen in snow-free areas e.g. the Sahara.
community is strongly encouraged.
References
0.85
0.9
0.95
31 GHz emissivity
1
Acknowledgements
We acknowledge Fatima Karbou, Marie Dumont and Samuel Morin of Météo-France/CEN for useful discussions and to Ghislain Picard for
making the DMRTML code freely available. Anabel Bowen is gratefully acknowledged for her help in editing this poster.
email: stephen.english@ecmwf.int
200
Conclusions
0.85
0.9
0.95
24 GHz emissivity
0
12z – 00z
1. Snowy regions show very large background departures for
0.85
600
–0.04
Snow free in East Europe (00z)
0.9
1
400
Day No. from 1/1/12
–0.03
0.95
0.8
0.8
200
–0.02
–0.04
1
31 GHz emissivity
0
–0.02
Four case studies of 50 GHz emissivity variation 2012-13 in different snowy regions are
shown above, and one for the Sahara for comparison with a snow-free region. The changes
in analysed emissivity with season are clear. In addition there are day:night differences in
analysed emissivity. The Greenland ice melt event in 2012 is also very prominent, around
day 200. The Siberia and East European cases show the timing of the onset of snow cover.
89 GHz emissivity
600
0.02
Emissivity
Emissivity
–0.03
400
Day No. from 1/1/12
0.04
The 12z–00z annual cycle
for Greenland and Antarctica
is similar, despite being in
opposite hemispheres? Same
pattern, different reasons?
–0.02
200
0.92
0.91
0.7
0
0.93
Emissivity
0.01
600
600
600
0.91
400
Day No. from 1/1/12
400
Day No. from 1/1/12
Further west the snow is marginal
and the winter shorter. This is clearly
seen in the analysed emissivities,
though in mid-winter the values are
not very different to Siberia. However
the 12z–00z differences are less
obvious than for Siberia, but do
become large when emissivities are
at their lowest values
0.02
0.96
0.9
Emissivity
0.8
200
50 GHz Emissivity Sahara 10W–25E 18-29N in 2012–13
12z
0.9
0
12z – 00z
50 GHz Emissivity Greenland 30–50W 70–80N in 2012–13
00z
600
0.04
–0.04
400
Day No. from 1/1/12
400
Day No. from 1/1/12
0.06
–0.04
200
200
0.08
–0.02
0
0
0.10
1/8/13
0.85
0.8
–0.02
Emissivity
Emissivity
0.75
0.95
0
0.85
0.8
As grain size grows the
emissivity falls. We also
start to reject more
data and emissivity
spatial variability rises.
0.02
12z
0.85
600
0.04
50 GHz Emissivity Dome C Antarctica 80–160E 70-80S in 2012–13
00z
12z
0.95
0
160°E
12z = 12z ECMWF data assimilation window, that covers the period 09z-21z and 00z = 00z
ECMWF data assimilation window, that covers the period 21z-09z.
400
Day No. from 1/1/12
Emissivity
80°W
200
On onset of
melting emissivity
rises rapidly and
the emissivity
variability reduces.
12z – 00z
0.08
40°S
00z
0.9
0
200
80
50 GHz Emissivity East Europe 25–50E 48–58N in 2012–13
12z
0.95
0.9
40°N
20°N
00z
0.95
2
0
60
(wet snow) has generally high emissivity. Snow pack models combined with
physical emissivity models may help constrain emissivity and skin temperature
analysis in some snow regions. The DMRTML results also show that snow can
increase or decrease emissivity, a result consistent with the emissivity retrievals.
50 GHz Emissivity Siberia 90–150E 60–70N in 2012–13
90-150E
25-50E
30-50W
80-160E
10W-25W
1
60°N
temperature, or another factor. In any case its clear we have a problem in the snow
covered regions to use satellite sounder radiances. Model runs such as DMRTML
(above right) show that low density (dry) snow has highly variable emissivity,
depending on snow grain size and snow depth. By contrast high density snow
Emissivity
Siberia
East Europe
Central Greenland
Antarctica Dome C
Sahara
3
120°W
0.05
40°E
Land and sea ice 0.7-1.0
80°N
160°W
20°E
Both higher and lower emissivity values
(compared to snow free conditions) are found
in the area where the ECMWF snow analysis
has more than 0.01m snow. Therefore snow
cover can both increase or decrease emissivity,
depending on the snow pack characteristics.
50°N
160°W 120°W 80°W
0°E
0.01
70°N
In winter large background departures are seen at ECMWF for AMSU-A channel 5.
In the summer these large departures are not seen. They are caused by problems
with the quality of the emissivity and skin temperature analysis over snow. This may
be use of specular assumption [see discussion in Harlow 2009], poor surface skin
Emissivity
40°E
60°N
Land and sea ice 0.7-1.0
1.
2.
3.
4.
5.
40°N
0.8
Emissivity
0°E
40°E
0.72 0.75 0.78 0.81 0.84 0.87 0.9
Land and sea ice 0.7-1.0
O-B K 1/8/13
80°N
40°W
0°E
0.7
80°N
160°W 120°W 80°W
40°W
50°N
Emissivity
40°E
50°N
Emissivity
0°E
60°N
60°N
Emissivity
40°W
1
70°N
70°N
Emissivity
160°W 120°W 80°W
-3
Emissivity 1/8/13
Emissivity 1/8/13
Dmrtml-1.6 (Picard et al GMD 2012): simulations
Snow grain size and snow pack depth matter when snow density is low only.
High density snow is almost “invisible” at 50 GHz
Emisssivity 50 GHz
Emissivity 1/2/13
snow surfaces using information from snow pack models and snow analysis, combined with
radiative transfer models?; (b) can we use the information in the emissivities to improve
snow analysis?
Varying snow grain size and snow depth
The ECMWF dynamic emissivity scheme produces instantaneous estimates and a Kalman
filter analysis of emissivity as described by Karbou (2005) and Krzeminski (2008). This poster
is thinking about two questions: (a) can we constrain better the emissivity analysis over
Harlow C. 2009: Millimeter Microwave Emissivities and
Effective Temperatures of Snow-Covered Surfaces: Evidence
for Lambertian Surface Scattering. IEEE Trans Geosci Remote
Sensing. 47.
Karbou, F. ; Prigent, C. ; Eymard, L. ; Pardo, J.R. 2005:
Microwave land emissivity calculations using AMSU
measurements. . IEEE Trans Geosci Remote Sensing.
43, 948-959.
Krzeminski B. N. Bormann, F. Karbou, P. Bauer, 2007:
Towards a better use of AMSU over land at ECMWF. Proc.
ITSC-16.
Picard G., L. Brucker, A. Roy, F. Dupont, M. Fily, A. Royer
and C. Harlow, 2013: Simulation of the microwave emission
of multi-layered snowpacks using the dense media radiative
transfer theory: the DMRT-ML model, Geoscientific Model
Development, 6, 1061-1078.