1. Ice Surface Temperature
Validation of
Ice Surface Temperature
from Metop IASI
- Infrared Atmospheric Sounding Interferometer
Response to EUMETSAT ITT No. 15/137
authors: Gorm Dybkjær, Jacob Høyer, Kristine Skovgaard Madsen‎, Jörg Steinwagner and Rasmus Tage Tonboe
Danish Meteorological Institute
Validation Ice Surface Temperature from Metop Infrared Atmospheric Sounding Interferometer. EUMETSAT, March 30, 2017.

2. Project Aim:
“The overall aim of this project is to gather accurate in situ observations of
Ice Surface Temperatures (IST) for a reference data set and perform a
validation of the EUMETSAT IASI level-2 IST data over sea and land ice.
Sea Ice Drift - Sea Ice Climate Change Initiative, WP 23XX workshop, "Haus der Wissenschaft” Bremen. December 30, 2016

3. Outline: Measuring IST (Ice Surface Temperature)
IASI IST and other satellite IST products
Uncertainties
In situ platforms and instruments
Data collocation
IASI IST performance
Summary
Sea Ice Drift - Sea Ice Climate Change Initiative, WP 23XX workshop, "Haus der Wissenschaft” Bremen. December 30, 2016

4. Measuring IST - challenges
AWS on Sea Ice, Qaanaaq NW Greenland
Large vertical variability
Large diurnal variability
Sensors covered by snow
- large variabilety within
meters.
Sampling sensitivity
Metop AVHRR IST compared with in situ air
and skin temperature measurements: STD
and bias from comparing Metop AVHRR IST
with 2m and 1m temperatures – within
10 min. (solid line) and 30 min. (dashed line).
Bias for 2mT 1mT skinT
STD for 2mT 1mT skinT

5. IASI IST and other satellite and model IST products - applied in this project
IASI IST; PWLR TIR+MW; All sky algorithm
level 2 and 3 analysis
Metop AVHRR IST since 2011; TIR; Clear sky algorithm. Early level 2 production at DMI (used here for level 3, inter-comparison) Comparable to Operational OSI 205, from OSI SAF
level 3 analysis
MODIS Terra/Aqua since (2000/2002); TIR; Clear sky algorithms.
MOD29 and MYD29
NWP; The operational deterministic model from ECMWF
AMSU and IASI assimilated
Applied to the Match-up data set
Training data for IASI PWLR analysis

6. Applied in situ data distribution 2012
Western Arctic Ocean and Greenland ice sheet is relatively well represented by In Situ observation, whereas Eastern Arctic Ocean and in particular Southern Ocean and
Antarctic ice sheet only sparly covered with in situ data.
”Upper” PROMICE stations on Greenland will recieve extra attention here, because they are independent measurements on homogeneous surfaces, with reliable surface temperature.

7. In situ Meta information
Name of dataset
Location
Temporal resolution
Ongoing
Air temperature sensor height
Uncertainty of air temperature
Surface temperature
Distribution
AMRC
Antarctica
10 minutes
Yes
3 m - may vary with snow cover
AWS quality, assessed to 0.5K -
Free with acknowledgements
ARM
Alaska
1 minute
2 m
AWS quality, assessed to 1K
Radiometer
ECMWF dribu
Arctic and Antarctic sea ice
Varying hours - days
About 1.5 m
AWS quality, assessed to 1.5K
SST sensor
Contact ECMWF
IceBridge IAKST1B
Arctic and Antarctic land and sea ice
0.1 second
Yes, until launch of IceSat-2
IABP
Arctic sea ice
1 hour
Unknown
Assessed to 1K
Free
NAACOS
6 hour No
Close to surface
Median of top 5 thermistor sensors, no radiation shield, assessed to 3K
Thermistor
Contact DMI
PROMICE
Greenland
0-2.7 m, varies with snow cover
Calculated from radiation
WMO, GTS weather station
Greenland and Antarctica
3 hour
Assumed 2 m
AWS quality, not assessed
In situ data available in netCDF here: ftp.dmi.dk/iasi_ist/insitu/data

8. Uncertainty IASI IST Qualitative assessment
The uncertainty mainly due to undetected clouds: this uncertainty is difficult to quantify but it normally results in a negative bias because clouds are normally colder than the Earth’s surface.
Instrument noise and how it propagates through the algorithm and affects the temperature estimate is a random uncertainty which is unlikely to be correlated temporally, spatially or with other uncertainty sources.
The geolocation uncertainty is a random uncertainty related to the pointing accuracy of the sensor. Over sea ice it is a function of the sensor spatial resolution, pointing accuracy, sea ice concentration and the ice surface fraction temperature itself.
The snow/ice emissivity uncertainty is related to the spectral emissivity variability - a function of snow grain size, density and viewing angle. The emissivity is wavelength dependent.
The algorithm uncertainty from the regression process, when estimating algorithm coefficients.

9. Uncertainty anticipated challenges using mw for sea ice temperatures
Spatial variations in ice concentration, surface melt, ice age and snow propeties makes it
an extra challenge to estimate surafce temperatures from MW data.
Arctic Sea-ice-with-snow-pack Emissivity at
50 GHz on a March day.
Svendsen (1993)
Snow/ice emissivity variability is largely as a function of snow grain size and density, viewing angle and sensor band (Wavelength). Melt onset change the MW emissivity drastically.

10. Data collocation and Match-Up data
Match-up file content
In situ skin temperature (K)
In situ air temperature (K)
In situ air pressure (hpa)
IASI surface temperature
IASI surface temperature quality
IASI total column water vapour (mm)
IASI total column water vapour quality
IASI Cloud cover signal. “OmC”
IASI satellite land fraction inside IASI IFOV (percent)
IASI elevation (m)
IASI sun zenith angle
IASI satellite zenith angle
IASI Mean surface temperature within 3x3 area
IASI STD surface temperature within 3x3 area
Distance between buoy position and IASI pixel centre (km)
Time difference between buoy IASI time stamp (min.)
Sea Ice Concentration
nwp skin temperature (K)
nwp air temperature (K)
nwp wind speed in 10m closest nwptime -3hours -2h … +3hours
*Cloud cover fraction estimated from PROMICE
Match-Up Criterias:
Position of in situ platform must be within 50 km of centre IASI pixel.
Recording time of in situ measurement must be within 50 minutes of IASI level 2 segments.
*Cloud cover is estimated from longwave radiation/near-surface
air temperature relationships. Using a clear sky and a cloudy algorithm
Match-Up data available as text files here:
ftp.dmi.dk/iasi_ist/mudb

11. Validation Strategy stratification with many possibilities
Rough estimate of possible data splits from this data set
Ice type(2): sea (2: fyi and myi) and land ice
Hemisphere(2): nh and sh
Observation type(3): air temp, surface temp and radiometric temp
Satellite temperature representation (2): nearest value and area mean
Associated variables (11); Time inter-annual, Time diurnal, cloud, elevation, temperature quality, distance to obs., sun-zen angle, sat-zen angle, ice concentration, wind, temperature.
…that gives more than 250 ways to analyse the data set.
There is only sufficient data to perform analysis for selected combination.

12. Land Ice Validation and sensitivity analysis

13. Performance on land ice – level 2 General performance
General All-Sky performance of
IASI IST on land ice.
Parameter
Bias
Standard deviatio n
Correla­ tion
Root mean square diff.
Number of matchups
Northern hemisphere
IASI IST vs. aws air temp
-2.5 K
5.4 K
0.9
5.98 K
315140
IASI IST vs. aws surf temp
-0.8 K
5.3 K
5.36 K
298816
IASI IST vs. NWP air temp
-3.2 K
6.2 K
333885
IASI IST vs. NWP skin temp
0.3 K
Southern hemisphere
3.78
5.1
0.95
6.36
8889
4.25
6.8
0.45
8.0
694
-2.2
2.9
0.98
3.6
10790
-0.07
2.93
10774
Following slides will illustrate where IASI IST performs best and worse by stratifiying the validation using associated information from the MUDB
Metop-AVHRR
Bias
STD r
Counts
IST, AWS Summit
-3.22
3.14
0.95
590
”Cloud-free” Metop AVHRR IST performance on Summit, 2011 Dybkjaer et al. (2012)

14. IASI IST performance on land ice – level 2 IASI IST quality indicator
STD and bias
IASI temperature quality indicator
Works well!
IASI IST – in situ IST
counts

15. IASI IST performance on land ice – level 2 Water vapour - total column
STD and bias
Seems to perform best at intermediate
humid atmospheres…
IASI IST – in situ IST
counts

16. IASI IST performance on land ice – level 2 elevation
STD and bias
The mean error is not affected by
Elevation.
Bias is zero at about 500 m
IASI IST – in situ IST
counts

17. IASI IST performance on land ice – level 2 OmC (cloud indicator)
STD and bias
Bias drops at positive OmC
Values – i.e. IASI IST gets cold
IASI IST – in situ IST
counts

18. IASI IST performance on land ice – level 2 Distance to observation
STD and bias
The error increases with increasing
distance to observation
IASI IST – in situ IST
counts

19. IASI IST performance on land ice – level 2 Sun-Zenith angle; day/night
STD and bias
Error increases
- going from “Day” to “Night” .
IASI IST – in situ IST
counts

20. IASI IST performance on land ice – level 2 - annual variability
IASI performance on Greenland Ice Sheet, by the month:
IASI IST; perform best during summer/daytime
IASI IST – Surface temperature estimated obs.
(many samples)
IASI IST – AIR temperature, obs.
(many samples)

21. IASI IST performance on land ice – level 2 Time sampling
IASI IST – in situ IST
No performance effect from
time sampling of plus/minus
50 minutes??!
STD and bias
counts

22. Sea Ice Validation and sensitivity analysis

23. IASI IST performance on sea ice – level 2
Parameter
Bias
Standard deviation
Correla­tion
Root mean square diff.
Number of matchups
Northern hemisphere
Sat IST - buoy air temp
0.4 K
7.5 K
0.75
26268
Sat IST - buoy surf temp
-7.7 K
8.9 K
0.53
11.8 K
12252
Sat IST - flight surf temp
3.6 K
3.3 K
0.82
4.8 K 16
Sat IST - NWP air temp
-0.6 K
2.8 K
0.96
2.9 K
34468
Sat IST - NWP skin temp
-0.3 K
Southern hemisphere
-2.3 K
0.76
722
6.2 K
9.0 K
0.15
10.8 K 39
-0.9 K
2.5 K
0.73
2.6 K
-1.5 K
2.3 K
2.7 K
General comparison of satellite data with in situ observations and NWP data
The IASI IST is apparently very cold compared to Arctic buoy surface temperatures!
We suspect that this can be an in situ observation arctifact, caused by snow/ice covered surface sensors.
Metop-AVHRR
Bias
STD r
Counts
IST, buoys Arctic
-2.76
3.69
0.89
7809
Comparison with ”cloud-free” Metop AVHRR IST, 2011 Dybkjaer et al. (2012)

24. Performance on sea ice – level 2 IASI IST – buoy air temperature
Northern Hemisphere
Southern Hemisphere
STD, Bias and distribution for IASI IST – Buoy air Temperature, as a function of IASI temperature indicator.

25. Performance on sea ice – level 2 IASI IST quality indicator
STD and bias
LARGE bias IASI IST – Obs IST:
A cluster of cold biased IASI
IST emerges at larger temperature
Indicators.
IASI IST – in situ IST
counts

26. Performance on sea ice – level 2 Ice concentration
Large STD at high IC… in particular for in situ IST
May be caused by unreliable surface temperature (vertical) positions, i.e. buried in snow/ice.
IASI IST – in situ air Temp.
IASI IST – in situ IST

27. Performance on sea ice – level 2 Water vapour Total column
Generally large errors in dry atmosphere conditions.
Very cold IASI IST at extremely dry atmosphere giver large errors compared with in situ IST
…possible a warm bias in in situ IST data, as above
IASI IST – in situ air Temp.
IASI IST – in situ IST

28. Performance on sea ice – level 2 Sun Zenith angle
IASI IST performs best during Day time
IASI IST – in situ air Temp.

29. IASI IST performance on sea ice – level 2 distribution of buoy air T vs IASI IST, daytime/summer
Histograms of associated variables in
each data group of the figure are included
in the report:
Example Ice Concentration
All
Cat.-1
Cat.-2
Cat.-5
Cat.-3
Cat.-4
Group
Description
Bias
Standard deviation
Correla­tion
Root mean square diff.
% of all data selected
All data – Day+night
0.4 K
7.5 K
0.75
100 1
Day, ≥ -5C
10/37 2
Day, good data < -5C
12/37 3
Day, buoy close to melt
1/37 4
Day, sat - buoy < -5C
8/37 5
Day, sat - buoy > 5C
6/37
Marginal ice zone
3.7 K
6.1 K
0.32
7.1 K 6
Ice covered region
0.1 K
0.73 94
First year ice
-5.2 K
8.3 K
0.63
9.8 K 8
Multi-year ice
1.2 K
6.9 K
0.61
7.0 K 33

30. IASI IST performance on sea ice – level 2 distribution of data, nighttime + twilight
Histograms of all associated variables in
each data group of the figure are included
in the report
Group
Description
Bias
Standard deviation
Correla­tion
Root mean square diff.
% of all data selected 6
Night, good data
38/62 7
Night, sat - buoy < -5C
10/62 8
Night, sat - buoy > 5C
14/62
Marginal ice zone
3.7 K
6.1 K
0.32
7.1 K
Ice covered region
0.1 K
7.5 K
0.73 94
First year ice
-5.2 K
8.3 K
0.63
9.8 K
Multi-year ice
1.2 K
6.9 K
0.61
7.0 K 33

31. IASI IST vs other Satellite algorithms – level 3 - Setup
Daily level 3 aggregated satellite fields
from the level 2 satellite IST within 36 hours
from the central analysis time on a regular 0.05
degree latitude and longitude grid.
The aggregated level 3 satellite are further
averaged for spatial equal area regions
of ~110x110 km throughout the Arctic.

32. Clear sky bias
Seasonal dependent Bias between All-sky and Clear-sky surface temperatures:
All-Sky IST has warm or no Bias from April to September (0-3 K)
Clear-Sky IST has cold or no Bias from October to March (0-4, only few samples…)
PROMICE monthly mean
surface temperature
All observations (red)
Cloudfree (Metop_A PPS) (blue)

33. IASI IST vs other Satellite algorithms – level 3 - General performance of daily mean IST, within 110x110 km
Satellite IST generally agree in the dark during winter.
During summer IASI IST is warm, with very low variability.

34. IASI IST vs other Satellite IST - level 3
IASI IST show less variability than TIR algorithms for warm IST
Scatterplots of the L3 observations from IASI
against Modis Terra (left) and Metop AVHRR (right).

35. IASI IST vs other Satellite algorithms – level 3 - Latitude dependency
Averaged Surface temperature from selected subsets,
At 75 N (top) and 85 N (bottom)
IASI IST is warm Bias against TIR IST
increases polewards
STD seems to perform slightly better with latitude
Latitudinal bias (solid) and standard deviation (dashed) of the differences,
when L3 IASI is compared against the other satellite L3 IR products

36. IASI IST performance vs traditional surface temperature algorithms – level 3
Products
Latitude
North
Number of matches
Bias
Standard deviation
Correla­tion
Root mean square diff.
IASI - Metop AVHRR
All
1425
2.8K
2.4K
0.98
3.7K
IASI – Modis Aqua
1585
4.5K
3.1K
0.97
5.4K
IASI – Modis Terra
1540
4.3K
3.0K
5.2K 70
248
1.4K
2.1K
0.99
2.6K
257
3.3K
4.9K
246
3.6K 75
491
2.3K
549
3.9K
5.0K
533
4.0K
5.1K 80
459
521
5.7K
507
2.7K 85
227
258
5.8K
6.4K
254
6.0K
COMPARE TO TIR PRODUCTS:
Validation statistics from radiometer temperature:
FRM – Sea Ice Field Experiment:
Match-Up Data from NW Greenland IR120 Radiometer
Winter/Spring 2016.
Inter-comparison statistics of daily level 3 products for all averaging regions and for each latitude.

37. PROMICE Cloud information vs IASI cloud indicators
- Apparently, there is no correlation between PROMICE Cloud index and temperature error and
IASI temperature quality indicator.
Scatterplot of temperature error
IASI IST – PROMICE air temperature,
as a function of PROMICE cloud index [0:1].
Scatterplot of OmC cloud index as a function
of PROMICE cloud index

38. Summary Sea Ice: Land Ice: Bias increases Polewards from 70 N
IASI temperature quality indicator Works well!
Very cold IASI IST at extremely dry atmosphere giver large errors compared with in situ IST.
IASI IST performs best during Day time
Large STD at nearly 100% IC…
We might look at too warm in situ surface temperatures – possible due to snow cover.
IASI IST show very variance for temperatures warmer than ~10 C
Land Ice:
IASI IST performs best during summer
Performance in independent of time sampling (?!)
Seems to perform best at intermediate wet atmospheres
Bias drops at positive OmC Values – i.e. IASI IST gets cold
IASI temperature quality indicator Works well!
Error increases with increasing distance to observation
Going from Day to Night performance between 80 and 90 degree sun-elevation