Comparison of Measured and Modeled Snow Brightness

Comparison of Measured and Modeled
Snow Brightness Temperature
Using Various Field Techniques
for Grain Size Measurement
Edward KIM
Michael DURAND
NASA Goddard Space Flight Center
Ohio State University
University of Colorado, Boulder
Jet Propulsion Laboratory
University of California Los Angeles
Cold Regions Research and Engineering Laboratory
• Problem statement
• Our approach
• Description of snow field
• EM models
• Microwave radiometer
• Tb comparison results
• Summary & Conclusions
The Problem with Snow Remote
Sensing and “Grain Size”
• For microwave remote sensing of snow, accurate grain size
numbers are extremely important to have because…
• the microwave signature of snow is highly sensitive to grain size
• Unfortunately “grain size” is not easy to quantify accurately,
especially in the field
• So what is the “best” way to measure grain size in the field for
use in microwave snow emission models?
How to determine a “best” grain size
field measurement technique?
• What do we really need?
• What do we already have?
• How do we compare (and will
anyone else believe our results)?
– What are we going to use as “truth”?
• What is practical?
Our Approach
• Field Measurements
Small campaign: 6 days, 1 snow pit per day
Limited set of techniques; side-by-side
Stereology chosen as “truth”
Location: Storm Peak Lab, Steamboat, Colorado, USA
• Brightness Temperature Comparison
Try 2 EM models, both multi-layer
Grain size & stratigraphy info fed into EM models
Models output brightness temperatures (Tb)
Compare model Tb vs. observed Tb
Grain Size Measurement Techniques
• Hand lens
• NIR photography
• Spectroscopy
– Contact
– New probe
• Stereology
Hand lens grain size measurements
Pit A
Pit B
Pex & Dmax related as in
Durand, Kim, Margulis, 2008
• Plenty of precedent (e.g., CLPX)
• Non-repeatable (e.g., two pits one meter apart; different days,
plus user variations; ‘B’ pit was in the radiometer FOV)
NIR Camera grain size
• Repeatable (in theory)
• Empirically-based (transferability?)
Matzl and Schneebeli, 2006
Mätzler, 2002
Contact Spectroscopy
• Measures reflectance across
entire VIS/IR range
• Reflectance varies with
grain size (see plot at right)
• Vertical resolution ~2cm
• Requires commercial
spectrometer ($$)
• Requires dark tarp to block
unwanted background light
Spectral Profiler Probe Prototype
• Sends an optical package
into a slotted sleeve
inserted into the snowpack
to perform contact
spectroscopy in-situ, w/o
• Black tarp not shown to
block light at base.
drive tube
fiber optic to spectrometer
probe carrier body
optical camera
reflectance probe
aluminum sleeve
nylon brush
Spectral Profiler
• Send optics down hole
• Lateral reflectance spectra
• Fiber optic sends signal to
spectrometer on surface
• No snowpit needed!
• Prototype unit; 1st field
trial, so analysis still in
Stereology grain size (1/4)
3D cast of actual snow grain structure made with dimethyl phthalate
Frozen in field with dry ice (stop metamorphism)
Shipped to CRREL for processing in cold room
Relatively well-known technique, but logistically intensive
Perla & Davis, 1980’s & earlier references
Matzl and Schneebeli, 2010
Red line = 1mm
Stereology grain size (2/4)
Time consuming laboratory work to cut and
photograph: 20 slices / sample
Stereology grain size (3/4)
Red line
= 1mm
L=total length of cycloid lines
I=# of intersections crossed
v=ice volume fraction
Do= optical equiv. grain size
• Cycloids overlaid on image for estimating surface area
• Time-consuming manual counting work
• Yields SSA directly, but then need to convert to pex for MEMLS
Matzl and Schneebeli, 2010
Stereology grain size (4/4)
• Image is classified as air or ice
• Draw line through image, compute autocorrelation, do for each vertical line
• Exponential is fit to true autocorrelation function
Wiesmann et al., 1998
EM models
• Multi-layer HUT
MEMLS sensitivity of Tb to grain size
• Vertically averaged
optical equivalent grain
size from pit
• Sensitivity is the tangent
to the curve
• For this grain size
(vertical line), sensitivity
is similar for 19 & 37 GHz
• To achieve 5K Tb
accuracy, need 10% grain
size accuracy, so for
typical grain size, this
Radiometric measurements
• Brightness measured daily
for three days at 19 and 37
GHz, v-pol
Results & Discussion
• Grain size & stratigraphy info fed into EM
• Models output brightness temperatures (Tb)
• Compare model Tb vs. observed Tb
MEMLS (pex) vs. observations
• Bias and mean absolute error < 5 K
• No empirical tuning factors required!
• Based on laboratory work
Multi-layer HUT results
19v observed: 249 K
19v modeled: 230 K
(similar for SPL5 and
37v observed: 230 K
37v modeled: 153
K (similar for SPL5 and
The 19v is off by 20 K, and
the 37v is off by 80 K.
Averaged together, and
you have 50 K.
MEMLS results
Multi-layer HUT results
Note: HUT is set up to use hand lens Dmax measurements, so we’ve compared
that to MEMLS hand lens, and then to MEMLS stereology as a reference
Summary & Conclusions
• Used hand lens, spectroscopy, NIR, & stereology methods to
measure grain size; stereology was our ‘truth’
• Used the grain size/correlation length to drive EM models (MEMLS,
multi-layer HUT)
• Compared EM model Tb’s vs. observed Tb’s (‘truth’)
• Using lab-based methods, Tb errors are 5-8 K
• Using field-based methods, Tb errors are ~10 K
• For error <=5K, need grain size accurate to ~10% ==> 20-100um
• Using pex directly from stereology, no tuning empirical factor is
• Using SSA, an empirical factor (0.74) is required to get the right Tb –
attested to in literature (Mätzler, 2002)
• Need further examination of multi-layer HUT to understand
apparent cold bias

similar documents