1. Retrieval of microwave effective grain size for seasonally and spatially varying snow cover
Juha Lemmetyinen 1), Chris Derksen 2), Peter Toose 2), Martin Proksch 3), Jouni Pulliainen 1), Anna Kontu 1), Kimmo Rautiainen 1), Jaakko Seppänen 4), and Martti Hallikainen 4)
1) Arctic Research, Finnish Meteorological Institute, Finland
2) Environment Canada, Climate Research Division, Canada.
3) WSL Institute for Snow and Avalanche Research SLF, Switzerland
4) Aalto University, Department of Radio Science and Engineering, Finland
Microsnow2014

2. Overview Aim: Data: Results:
Study performance of HUT snow emission model over a whole winter season & over varying landscapes
Variability of effective grain size
Data:
Multi-scale radiometer observations from same test site (tower; mobile ground-based; airborne; satellite)
in situ information on snow structure (snow pits) and distribution (SWE/depth transects)
Results:
Analysis of forward model simulations at multiple scales, using in situ snow information for input
Retrieval of effective grain size at varying scales; seasonal changes and land cover effects.

3. ESA GlobSnow SWE Based on variational data assimilation
‘Calibration’ of forward model by calculation of effective grain size deff,r
compensation for incomplete input information (snow stratigraphy, land cover effects)
mitigation of forward model errors & deficiencies
Satellite scale retrieval: is there any physical connection left between deff,r and actual snow cover characteristics?

4. Grain size metrics Metrics of snow grain size used in this study:
’classical’ Grain size (Dmax)
Maximum extent of a snow grain, averaged for a layer
Determined visually from snow pits
Retrieved effective rain size (deff,r)
Precise retrieved grain size matching emission model output to brightness temperature observations.
All retrievals performed for GHz V-pol
Empirical effective grain size (Dmax,eff)
Empirical modification to Dmax matching HUT model to observations. Ideally, Dmax,eff = deff,r

5. Radiometry – multi-scale sensors
Fixed ground-based (tower)
Mobile ground based (sled)
Airborne
TB temporal evolution through whole season
Spatial variability of TB under forest canopy
Spatial variability of TB including forest canopy

6. Tower based observations
Sodankylä
Site of Nordic Snow Radar Experiment (NoSREx)
Focus on winter

7. Test site Dominant land cover in test area: Forest 70.9 % Bog 23.3 %
Lakes & rivers 3.7 %
Open (”barren”) 1.2 %
Other 0.8 %

8. Airborne TB at 18.7 GHz V-pol
Anna Kontu IGARSS 2014, Quebec, Canada

9. Microwave radiometer system
L, X-, Ku-, Ka- and W- band dual pol radiometers (Elbara-II & RPG-8CH-DP ‘SodRad’)
30 min measurement every 4 h
Elevations 30 – 70 , azimuth scanning possible (except L-band)
Sky tip calibration every 12 h
Sky zenith used to verify stability between snow observations

10. Mobile radiometer observations
Environment Canada Radiometers
Frequency (GHz)
19.0
37.0
Bandwidth [MHz]
1000
2000
Sensitivity [K]
0.04
0.03
Accuracy [K]
<2
<1
3dB [°] 6
i [°] 53
Spatial Footprint (m)
0.6 x 0.6
Radiometer calibration uncertainty
Frequency
19V
19H
37V
37H
RMSE
1.3
1.2
1.1

11. Airborne observations
March 17th 2011
HUTRAD multi-frequency radiometer system
6.8 GHz, GHz, 18.7 GHz, 36.5 GHz (dual pol)
15 x 7 km transects over study area
Two-point antenna calibration before and after flight operations
SC-7 Skyvan research aircraft

12. Snow cover observations
Weekly manual snow pit measurements
Snow stratigraphy
Density profile (snow fork and snow scale)
Grain size profile
Temperature profile
Moisture profile (snow fork)
Bulk values for SD, SWE, density
Additional snowpits and sampling during mobile radiometer and flight campaigns
Automated observations
AWS
Several snow depth sensors
Snow and soil temperature profile
Soil moisture profile
SWE from gamma-ray extinction

13. Mobile radiometer observations
Basic snow pit information from all sites
Density, temperature profiles
Snow grain size/type profile
Snow characterization from selected sites
SSA from NIR photography
SSA & correlation length from CT analysis of snow samples
Snow pits made directly in instrument footprints

14. Mobile radiometer observations
Basic sampling of snow depth and density
SWE every 500 meters, SD every 100 meters
Provides reference for SWE spatial variability and connection to vegetation/land cover

15. Forward modeling experiments

16. Winter time series 1-layer Forward modeling of winter time series
Snowpits, made outside of instrument footprints, were aggregated to 1- and 2-layer representative pits (depth weight averaging for grain size)
3rd order fit applied to grain size data
Reduces random error

17. Winter time series 2-layer Forward modeling of winter time series
Snowpits, made outside of instrument footprints, were aggregated to 1- and 2-layer representative pits (depth weight averaging for grain size)
3rd order fit applied to grain size data
Reduces random error
Soil parameters optimized using GHz (model by Wegmüller & Mätzler)

18. Winter time series 1-layer simulations:
underestimation for both 19 and 37 GHz
Use of empirical Dmax,eff formulation improves results

19. Winter time series 1-layer simulations:
underestimation for both 19 and 37 GHz
Use of empirical Dmax,eff formulation improves results

20. Winter time series 2-layer simulations:
Underestimation improved (still apparent especially for 19 GHz)
Improvement largely due to better representativity of grain size value

21. Winter time series 2-layer simulations:
Underestimation improved (still apparent especially for 19 GHz)
Improvement largely due to better representativity of grain size value

22. Sled based 1- & n-layer simulations:
In situ measurements made directly in instrument footprints -> n-layer simulations applied
Vegetation canopy effects compensated for forested sites (downwelling Tb reflected from snow)
Results again improved using multi-layer simulation
Use of empirical Dmax,eff required to obtain reasonable results
Sled based
Model bias using in situ data

23. Airborne Airborne simulations:
Snow pit data used to assign typical characteristics for snow for varying land cover (1-layer sim)
Vegetation canopy effects compensated based on forest biomass in instrument FOV
Results improved using multi-layer simulation
Use of empirical Dmax,eff required to obtain reasonable results

24. Results: tower based Tb vs. SWE

25. Results: mobile radiometers vs. SWE

26. Results: airborne radiometers vs. SWE

27. Results: airborne radiometers vs. SWE

28. Retrieval of effective grain size

29. Effective grain size Retrieval from AMSR-E over whole season:
Deff,r retrieved for satellite scene over Sodankylä test site
Forward model
observation

30. Effective grain size
Trend & overall level corresponds well to tower based retrieval – forward model is able to compensate for land cover effects

31. Effective grain size
Trend & overall level corresponds well to tower based retrieval – forward model is able to compensate for land cover effects
Reasonable match even with ’classical’ Dmax, averaged over whole snowpack (RMSE < 0.3 mm)

32. Effective grain size airborne & sled

33. Applicability of single deff for varying land cover
Effective grain diameter retrieved from average of GHz V-pol for whole measured scene
Forward simulation of individual footprints using mean snow depth (+/- std) and common deff,r

34. Summary
For a well-controlled test setup (no vegetation, land cover effects) the retrieved deff,r for the HUT snow emission model can bear a relation to the measured Dmax
HUT model was empirically formulated using Dmax to calculate extinction coefficient
For varying land cover, deff,r reacts to
Lack of input information
Model deficiencies
Actual differences in snow structure
For deff,r to be applicable over versatile land cover, 1.,2. should be minimized
Error in retrieved SWE (GlobSnow algorithm) as function of error in effective grain size
deff,r underestimated
deff,r overestimated