Soil Moisture During the Other Half of the Year--Snow and Freezing Soil Considerations on Liquid Soil Moisture Retrieval Accuracy

1
Soil Moisture During the Other Half of the
Year--Snow and Freezing Soil Considerations on
Liquid Soil Moisture Retrieval Accuracy

• Edward Kim
• NASA/GSFC
• NAFE Workshop 22-23 September, 2008
• U. Melbourne

2
Outline

• What happens to SM in the other half of the
  year?
• Why is that important?
• Why are frozen ground snow important for SM
  remote sensing? (what is their effect on liquid
  SM retrieval accuracy?)
• How can SMOS SMAP address snow frozen ground?
• Future research needs

3
Why are snow frozen ground important to SM RS?

• Snow frozen ground directly affect the accuracy
  of liquid SM retrievals via RS, and since
  snow/frozen ground occur over large areas for
  long periods, they directly affect where when a
  SM satellite can or cannot observe SM.
• Melting snow is a major source of SM (and runoff)
  in the spring.
• For large parts of the Earths land areas, it is
  the major source.
• Seasonal snow water has significant intrinsic
  value.
• When snow melts in large volume and/or suddenly,
  SM soil freeze/thaw state directly determine
  whether flooding will occur.

4
Soil moisture
5
SMOSREX
SGP97, SGP99
Regions of the Earths land surface where water
is frozen (seasonally or permanently),
constituted by snow, ice, frozen soils and
vegetation.
Q What do we need to know about snow fg in the
context of SM? A (snow) where when, how
much, how wet A (frozen ground) frozen or
thawed state
3
6
Soil moisture in the other half of the year/ the
other half of the planet
Q how to sample frozen soil moisture? Note
Geologists classify ice as a rock
7
Why is seasonal snow important?

• Largest areal extent of any component of the
  cryosphere.
• Over 30 of Earths total land surface has
  seasonal snow, and 10 is permanently covered by
  snow and ice.
• Forms a significant of the Earths surface for
  a significant part of the year. Over 60 of the
  northern hemisphere land surface has snow cover
  in midwinter.

4
8
Why are snow frozen ground important?

• And then it all melts away? Largest annual change
  (e.g., albedo)
• Seasonal snow ( permanent snow and ice) affects
  the planetary albedo.
• Snow cover frozen ground (extent, duration,
  water equivalent, melt onset) are sensitive to
  climate change.

Snow modulates how many W/m2 of solar radiation
are reflected or absorbed by the surface.
Recall the solar constant Is 1366 W/m2.
9
Why is frozen ground important?

• Seasonally and permanently frozen soils occur
  across 35 of the Earths land surface.
• Permafrost underlies 26 of the Earths land
  surface. Permafrost areas are natural indicators
  of climate change.
• Snow is an excellent insulator. Its presence or
  absence, depth, duration, etc. all affect the
  location timing of soil freezing.

7
10
Why is seasonal snow important?

• Over much of the industrialized world (1/6 of
  worlds population) 50-100 of runoff results
  from snowmelt1, affecting about a quarter of the
  global gross domestic product.

• Scientists, managers and planners need to know
  how much water is in the snowpack.2

Runoff Dominated by Snow1

• Snow arrival is out of phase with demand. Water
  resources are often concerned with predicting and
  impounding snowmelt to distribute throughout the
  dry period.

Red outline Snowmelt dominated and lacks
sufficient storage capacity to buffer shifts in
seasonal hydrograph.
1Barnett et al, Nature (17), Nov 2005 2Anthes et
al, p 4-39 3Anthes et al, p 4-38
11
Why is seasonal snow important?
Where will my water come from?
The western U.S. is critically dependent on
natural storage of water in seasonal snowpacks
80-90 of total annual streamflow originates as
snow. There are no significant opportunities for
additional capacity from man-made reservoirs.
On one day in Feb 2004, NWS model analyses
indicated the volume of water stored in snow
across the CONUS was 11 of the U.S. total annual
renewable fresh water resources (258 km3 or 59
of estimated U.S. total annual freshwater
withdrawal). Yet, the accuracy of total snow
water storage estimates remains unverified.
Models and state-of-the-art satellite retrievals
differ by 30! And, this is just for the U.S.
12
Snow, frozen soil and natural hazards
13
Why are snow frozen ground important to SM RS?

• Snow frozen ground directly affect the accuracy
  of liquid SM retrievals via RS, and since
  snow/frozen ground occur over large areas for
  long periods, they directly affect where when a
  SM satellite can or cannot observe SM.
• Melting snow is a major source of SM (and runoff)
  in the spring.
• For large parts of the Earths land areas, it is
  the major source.
• Seasonal snow water has significant intrinsic
  value.
• When snow melts in large volume and/or suddenly,
  SM soil freeze/thaw state directly determine
  whether flooding will occur.

14
How can SMOS SMAP address snow frozen ground?

• Ignore it gt errors in SM retrieval, where/when
  unknown
• Avoid it gt mask out areas w/snow and/or frozen
  ground
• Then you need to know where are those areas
• There are existing snow maps, FG maps are only
  from NWP models, climatologies, or crude
  algorithms (e.g., degree-days)
• Explicitly account for it gt
• Need to include snow and/or FG in retrieval
  algorithms
• Can take advantage of existing snow RS products
• With a column model DA, it is theoretically
  possible to
• retrieve SM underneath snow
• retrieve separate liquid water ice content in
  soils
• predict the timing of soil freezing thawing
• The major intl NWP centres are already pursuing
  3, using RA

15
Subpixel heterogeneity

• Error mechanism when snow or frozen ground is
  ignored

16
Brightness Temperaturesfrom snow frozen
ground(a qualitative example)

• Tb e Tphys
• Frozen soil looks dry
• emissivity increases
• big jump possible, especially at L-band
• Tphys is cold (e.g., 275 K ? 265 K, small neg
  change)
• penetration depth increases, so internal soil
  structure might become significant (volume
  scattering)
• Dry Snow attenuates emission from soil emits
  some itself. Snow would need to be fairly deep,
  have large grains, and/or have a large SWE to
  have a big impact at L-band. But, how big
  depends on the required SM retrieval accuracy.
• Wet Snow (if thick enough) is very absorbing
  can act like a blackbody at 273 K

17
Example SM bias error from mixed pixel w/snow
Colder snow Tb
Warmer ground Tb
Net Tb of whole pixel is biased cold ?
retrieved SM is wet biased unless snow is
accounted for.
Illustrative only, not actual L-band data
Box represents one satellite pixel, whose mean Tb
is used to retrieve SM
18
OK, I wont ignore snow frozen ground, Ill
avoid them

• What can I use to mask them out?

19
Existing state-of-the-art snow frozen ground
products

• Snow
• RS-based products (e.g., blended products)
• Snow cover maps (snow masks for SM)
• Snow water equivalent (SWE)
• Melting snow maps
• Model-based products
• e.g., SNODAS
• Frozen ground
• RS-based products none
• Model-based products
• Many global regional NWP models include FG, but
  only crudely

20
state-of-the-art snow cover maps IMS Interactive
Multisensor Snow and Ice Mapping System
Daily, February 1997 to present (24 km)Daily,
February 2004 to present (4 km) NH only
IMS from D.Robinson
RUGSL
21
NOTE the discrepancies occur along the snow
/ no-snow boundary
IMS from D.Robinson
22
Blended snow product prototypes, SWE (mm) from
AMSR-E with additional snow extent from MODIS in
red (October 24-31, 2003, max). Lower image
represents AMSR-E snow extent in grey, with
percent area of additional pixels that MODIS
classifies as snow in blues. (AMSR-E _at_ 25 km,
MODIS CMG _at_ 5 km)
NSIDC blended product Fall Season
Example (R.Armstrong)
23
Blended snow product prototypes, SWE (mm) from
AMSR-E with additional snow extent from MODIS in
red (Feb 26 Mar 5, 2003, max). Lower image
represents AMSR-E snow extent in grey, with
percent area of additional pixels that MODIS
classifies as snow in blues. (AMSR-E _at_ 25 km,
MODIS CMG _at_ 5 km)
NSIDC blended product Winter Season
Example (R.Armstrong)
24
ANSA snow product goals

• Project Primary Objective To develop a global,
  daily, 25-km resolution, blended-snow product
  that includes include snow extent, fractional
  snow cover snowpack ripening and melting snow
• The blended-snow product is an all-weather
  product with snow mapped by both visible and
  near-IR (MODIS) and microwave data (AMSR-E and
  QuikSCAT) for clear sky conditions. Only AMSR-E
  and QuikSCAT data will be employed when clouds
  obscure the surface. By fusing products we will
  complement their capabilities and aid in reducing
  the limitations and errors inherent in each
  separate product
• Melt onset at global scale will be derived by
  using QuikSCAT and AMSR-E data. AMSR-E data will
  be used to estimate the brightness temperatures
  when QuikSCAT detects melting snow
• Snow cover estimation SOA
• Modis, 500m, but not daily, clouds daylight
  limitations
• Passive Microwave (amsr-e) gt25KM, but daily,
  all-weather, day/night
• One of longest satellite data records

25
ANSA snow map 15 December
2007
Blended Snow Grid Values
(575) MODIS snow 80-100 and SWE 2-480 mm
(550) MODIS snow 21-79 and SWE 2-480 mm
(450) MODIS snow 1-20 and SWE 2-480 mm
(390) MODIS snow 80-100 and SWE 0 mm
(370) MODIS snow 21-79 and SWE 0 mm
(360) MODIS snow 1-20 and SWE 0 mm
(375) MODIS snow 1- 100 and SWE water mask
(355) MODIS snow 0 and SWE 2-480 mm
(350) MODIS cloud and SWE 2-480 mm
(330) MODIS cloud and SWE 0 mm
(300) MODIS cloud in AMSR-E swath gap
(345) MODIS snow1-100 in AMSR-E swath gap
(305, 290) MODIS no data SWE 2-480 mm
(295) MODIS in darkness and SWE 2-480mm
(250) MODIS in darkness and SWE 0 mm
(253) AMSR-E Permanent Snow/Ice
(201) MODIS snow 1-100 and SWE land not processed
(200) MODIS snow 1-100 and SWE no data
(0) Land
(1508) Ocean
(1498) Fill
26
State-of-the-art snow water equivalent
mapping Remotely sensed or model-based?
Passive microwave radiometry is used to map SWE,
but current resolution is 25 km.
Comparison of SSM/I-derived SWE to analyses from
modeling and data assimilation show differences
in continental-scale SWE of 35.
19
27
Just how much SWE is there?
AMSR-E SWE vs. NOAA model SWE
SWE(cm)
- 200
- 150
- 100
Difference in SWE (AMSR-E NOAA) February 10,
2004
- 50
- 45
- 40
- 35
- 30
- 25
- 20
- 15
- 10
AMSR Low
NSA High
- 8
- 6
- 4
- 2
Neutral
Ground Observed - NOAA
0
2
U.S. Volume Difference -113 km3
4
5 of U.S. Annual Renewable Water Resources 1/5
of the total annual flow of the Mississippi
River Entire total annual flow of the Yukon
River 6 times the total annual flow of the
Colorado River 10 times the total annual flow of
the Hudson River
6
8
AMSR High
NSA Low
10
15
3528 Stations Reporting
20
21
28
how would I explicitly account for snow and/or
frozen ground?
29
Radiance-based Assimilation for snow

• Advantages
• Potential way around
• previous limitations
• (forest example)
• Fewer uncertainties
• in a DA framework
• Physically-based
• Potential additional
• retrieved variables

Observed radiances
Integrated Snow Modeling Framework
modeled radiances

Forward radiance model
radiance assimilation
GSRAP Goddard Snow Radiance Assimilation Project
Snowpack physical model
Multivariate retrievals
Land surface model

• Disadvantages
• Complex
• Requires more
• ancillary information

Weather forcing
modular framework
30
Summary

• Snow frozen ground directly affect the accuracy
  of liquid SM retrievals via RS, and since
  snow/frozen ground occur over large areas for
  long periods, they directly affect where when a
  SM satellite can or cannot observe SM.
• Melting snow is a major source of SM (and runoff)
  in the spring. For large parts of the Earths
  land areas, it is the major source.
• When snow melts in large volume and/or suddenly,
  SM soil freeze/thaw state directly determine
  whether flooding will occur.
• F/T is also a key driver for the Carbon Cycle.
• Both SMOS SMAP are aware of snow FG issues.
  SMAP will use its radar to map F/T. For now,
  SMOS plans to avoid snow. SMAPs snow strategy
  will probably evolve between now launch.
• There are existing snow products available for
  the avoid strategy.
• Explicitly accounting for snow within current
  SMOS/SMAP SM retrieval accuracy requirements will
  require further modeling work, and field
  measurements.

31
CLPX-1 IOP1 Feb 2002
Local Hazards
Iridium Satphones
Medium Depth Pit
CRREL FMCW radar
Grain Size Obs
Japanese Radiometer
32
CLPX-1 IOP1 Feb 2002
Soil Sampling
Wind Scouring
33
Backups
34
The economic value of snow
35
Snow as a dielectric medium 3

• Thus, RT in (dry) snow becomes a problem of
  modeling the scattering
• But dry snow scattering is a very strong function
  of grain size snow density
• Note how the plot of penetration depth for dry
  snow (attenuation due to scattering) at right
  varies by orders of magnitude with grain size for
  a given frequency

36
Snow as a dielectric medium 4

• Wet snowis not our focus, but
• For all but very slightly wet snow (lt 2
  wetness), absorption dominates vs. scattering
• As long as the wet layer is thick vs. the
  penetration depth at that frequency

37
A real snowpack-note layers
38
Model snowvs.Real snow
1.0 mm
all SEM pictures are same scale (from USDA EMU)

• Single-layer model
• representation using spheres
• from J.Barlow, 2003

Model snow
Real snow
1.0 mm
Soil
Soil
39

• AMSR-E snow algorithm

Slide adapted from 2005 AMSR-E Science Team
presentation by Richard Kelly, Univ. Waterloo,
Canada
40
AMSR-E available products

• Daily, Pentad Monthly Products
• EASE-Grid northern and southern hemisphere
  products 721x721 grid cells per hemisphere
  25x25 km cells
• Pentad and Monthly products are maximum SWE

Daily 19 January 2005 Monthly April 2005
Slide adapted from 2005 AMSR-E Science Team
presentation by Richard Kelly, Univ. Waterloo,
Canada
41
Brightness Temperaturesfrom mixed pixels w/snow
frozen ground(a qualitative example, not
L-band)
SSM/I AMSR-E
NORTH PARK MSA, CLPX-1 19 GHz, 25 x 25 km