reflectance intercomparisons at a radcalnet site for ... · 9 intercomparison results 12 runs over...
TRANSCRIPT
1
GODDARD SPACE FLIGHT CENTER
Reflectance intercomparisons at a RadCalNetsite for training and traceability
K. Thome, C. Ong, C. Barrientos Gajardo, I. Lau, J. Czapla-Myers, B. Wenny
JACIE Workshop
College Park, MD 17-19 September 2018
2
Training course background
Geoscience Spaceborne Imaging Spectroscopy
Technical Committee (GSIS TC) within IEEE
conducted a training activity in July 2017 Follow-on to a successful course at the 2016 IGARSS
meeting in Beijing
Outcome of the original course was that practical
experience with field measurements would be beneficial
Decision made to leverage the 2017 IGARSS locale of Ft.
Worth, Texas to do the practicum at Railroad Valley Playa
in Nevada
3
Goals
Training for personnel working across various
earth observation areas including sensor
engineers to calibration scientists
Visit a CEOS WGCV RadCalNet test site and
the associated calibration laboratory for that
site
Demonstrate measurement protocols for
surface and atmospheric parameters
Concentrate on vicarious calibration of
imaging spectrometers
Minimize uncertainties
Perform a reflectance-based calibration of
an on-orbit satellite sensor
Expose the students to >30 hours of old guy
road trip music
9/12/18, 3)34 PMwestern us map - Bing Maps
Page 1 of 2ht tps://www.bing.com/maps?q=western+us+map&FORM=HDRSC4
1600 Elm St, Fort Worth, TX 76102
US-6, Tonopah, NV 89049
26 hr 53 min, 2522.2 km
Moderate traffic (22 hr 38 min without traffic)
Via I-20 W, I-10 W
A
B
© 2018 HERE© 2018 HERE
250 miles250 miles 250 km250 km
Railroad Valley Playa, USA
• >20 years working experience on site
• 4 radiometers, sun photometer, met
station
• Dry lakebed
4
Course provided an opportunity
Number of people and availability of equipment allowed for
studies of reflectance measurement uncertainties
Surface reflectance inter-comparison similar to past campaigns
including those from Lunar Lake 2000 JACIE efforts
Impacts of spatial and spectral homogeneity of the site being
measured
Approaches to separate instrumental variability from test site variability
Methods to improve site sampling
Wavelength (nm)
400 700 1000 1300 1600 1900 2200
0.25
0.35
0.45
0.55
JAC UA
JPL SDSU
5
Laboratory visit
Exercise included a tour of the optical calibration lab operated
by University of Arizona Remote Sensing Group (RSG)
S. Biggar of RSG described the facilities and led the tour of the group’s
radiometric calibration facility
Tutorial put the laboratory measurements into context of the field data
Also provided a chance to describe the measurement campaign
plans
6
Measurement protocols
Underlying theme of the training course is related to SI-
traceable uncertainties within reflectance-based calibration
Lectures emphasize the development of error budgets and
approaches to limit uncertainties
2017 Training Course examined uncertainties for reflectance-based
vicarious calibration
Leveling reference
Alignment to reference
Reference measurement
and note taking
7
Intercomparison studyTwo areas were used to evaluate the consistency
between retrievals of surface reflectance
Results would indicate how well the training went
Also provide insight into uncertainties of the
vaalidation of atmospherically-corrected, analysis
ready surface reflectance products
Help evaluate error budgets for Railroad Valley
RadCalNet TOA reflectances
Instrumentation and references were from RSG,
thus evaluation of absolute uncertainty was more
difficult
Did collect data to evaluate two different SI-
traceability paths to surface reflectance
On-site characterization of RadCalNet
radiometers
8
Repeatability
80 m linear transect collected multiple times
by each group over multiple times Three teams of two people
Spectroradiometer operator
Note taker/spotter/reference panel operator
Teams consisted of a mixture of levels of field spectroradiometer
experience
Reference followed by 80-m transect and finish with reference
Spectra of site collected while walking
80 m East-West
9
Intercomparison Results
12 runs over 3 hours on the first day by 7 operators using 2
spectroradiometers and 2 panels
Clear skies over sun with some patchy clouds as day progressed and
temperatures >37 oC
~4% variability in the average reflectance of all runs from July 31
Instrument dependant variations were seen between the
spectroradiometers
-8
-6
-4
-2
0
2
4
350 650 950 1250 1550 1850 2150 2450
Perc
ent
Dif
fere
nce
Wavelength (nm)User 1 to User 2 Morning to afternoon Day 1 to Day 2 Continuous to stop/measure-8
-6
-4
-2
0
2
4
350 650 950 1250 1550 1850 2150 2450
Pe
rce
nt
Dif
fere
nce
Wavelength (nm)
10
Measurement method comparisonComparison of the moving collection (continuous)
method versus a stop and measure method Collect data at 8 spots on 80-m transect
Operator collects at while standing stationary
Panel reference measurement at start and end of each
surface measurement
Minimize impacts of atmosphere, instrument variations
Drawback is collection samples less area and can take
longer to cover site
11
Stop-Measure Results
Retrieved reflectance from two different collection approaches
within variability of a single collection approach
Noticeable difference between results of the two sampling strategies
Standard deviations were typically larger for the stop-measure
approach (overall, per sample, averaged per sample)
-4
-2
0
2
4
6
8
350 650 950 1250 1550 1850 2150 2450
Co
nti
nu
ou
s to
sto
p-m
easu
re
Perc
ent
Dif
fere
ce
Wavelength (nm)
-10
-8
-6
-4
-2
0
2
4
6
8
10
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
350 650 950 1250 1550 1850 2150 2450
Perc
ent
Dif
fere
nce
Ref
flec
tan
ce
Wavelength
Continuous Stop-measure Percent Difference Continuous Std Dev. Stop-measure Std Dev
0
2
4
6
8
350 850 1350 1850 2350
Perc
en
t St
and
ard
D
evia
tio
n
Wavelength (nm)
0
1
2
3
4
5
6
350 650 950 1250 1550 1850 2150 2450Perc
ent
Stan
dar
d D
evia
tio
n
Wavelength
Continuous
Stop-measure
12
Reflectance-based test site
Sentinel-2a overpass offered an opportunity to evaluate large-
site retrievals
Used continuous sampling and stop-measure sampling approaches
Transects are 20 m apart over a 240-m x 80-m area
Reference panel measurement, relocated at the southern end, at the
start and every second transect
Stop-measure relied on 12 random transect locations from the 48
possible
13
Reflectance-based site results
Cloudiness prevented all operators from collecting and from
determining the Sentinel-2A comparison
Results match with 80-m transect
collections
Results are within typical results of
reflectance-based collections
-6
-4
-2
0
2
4
6
350 650 950 1250 1550 1850 2150 2450
Pe
rce
nt
Dif
fere
nce
fro
m 1
90
0 U
TC
colle
ctio
n
Wavelength
14
Reflectance- versus radiance-based retrieval
Compared surface reflectance reflectance ratio to absolute radiance measurements
Near simultaneous collection
Field spectrometer data converted to
reflectance using diffuser
Radiance data converted to reflectance
via radiative transfer code calculations
based on measured atmospheric conditions
15
Was the training successful?
Groups demonstrated that they
could retrieve surface
reflectance within the
uncertainties of the approach
Results raised further questions
Variability of the stop and measure approach was larger than
continuous sampling
Cause is unclear whether it was surface, instrumental, or
atmospheric
Points to the need for a better sample for Type A uncertainty
assessment
Difference between the continuous results and stop and measure
results indicate the need for better site sampling techniques
16
2018 Joint campaign sought to answer questions from 2017 training
Refine collection approaches and develop a set of guidelines for
future intercomparisons and surface reflectance validation Four basic collections
Reflectance-based calibration of Terra platform sensors
80-m playa transect
60 cm x 60 cm gray tarp sample
60 cm x 60 cm barium sulfate painted panel
Sky conditions were not the most cooperative
Intercomparison results still being evaluated
17
2018 NASA GSFC results
Ratio of retrieved reflectance from NASA GSFC data sets relative to
clear-sky, optimal collection for that target
0.88
0.9
0.92
0.94
0.96
0.98
1
1.02
1.04
1.06
1.08
350 850 1350 1850 2350
Rat
io t
o M
ay 3
Cle
ar S
ky
Wavelength (nm)
0.9
0.95
1
1.05
1.1
350 650 950 1250 1550 1850 2150 2450
Rat
io t
o M
ay 3
Cle
ar S
ky C
ase
Wavelength (nm)
Barium sulfate 48% tarp
GSFC results indicate reflectance can be retrieved with better than 5% repeatability regardless of sky conditions
Key is representative reference sample
Result of near-lambertian samples
Will impact experiment designs for surface reflectance validation campaigns
0.98
0.99
1
1.01
1.02
1.03
1.04
1.05
1.06
1.07
1.08
1.09
1.1
350 650 950 1250 1550 1850 2150 2450
Rat
io t
o M
ay 3
Cle
ar S
ky C
ase
Wavelength (nm)
80 m by 80 m area
18
Added measurement from 2017 - Gray tarp
Limiting reflectance collection to a small sample area leads to
teams measuring the same area
Tarp sample measurements will isolate differences
caused by the instruments or the reference
Effects from from changing solar irradiance are
limited because of short time between
references and the gray sample
Short time between groups limits the BRDF
effects of the sample
Tarp is being evaluated as a traveling standard
that would allow intercomparisons without the
need for groups to gather in a single location
One round of collections had all groups using the
same reference standard
19
Added measurement from 2017 - Barium sulfate panel
Barium sulfate has a reflectance similar to those of the
reference standard
Using BaSO4 panel reduces instrument effects such as those caused by
non-linearity, lower SNR at lower reflectance, etc.
Both BaSO4 and gray tarp retrievals can be compared to laboratory-
based predictions of reflectance allowing a check on the absolute
uncertainty
20
What we hope to learn from 2017 and 2018
The ensemble of results should help indicate what the limits
could be to absolute and relative uncertainties
Best agreement should be for retrieval of reflectance of BaSO4 panel
Highest SNR
Uniform sample
Rapid data collection
Minimal BRDF effects
Gray sample should give next best results since it is still of a very limited
spatial area
Lower reflectance could point out SNR or linearity effects
Can still be some BRDF and spatial heterogeneity effects
80-m transect gives a realistic measurement while limiting area
Full site measurements evaluated against calibration of an imaging
sensor viewing all of the characterized areas
21
Summary
Training of students during the practicum shows that following
protocols leads to repeatability similar to experienced groups
Previous work shows absolute uncertainties are dominated by
reflectance standard uncertainty as show
Still unclear what dominates the lack of repeatability
Methods to isolate Type A uncertainties need development
Small, stable samples
Stable mounting of spectrometer relative to small sample
Stop and measure approach should limit uncertainties from variability
in spectrometer and atmosphere
Need to implement more rigorous evaluations of temporal and spatial
sampling approaches
22
Acknowledgements Thank you to IEEE Geoscience & Remote Sensing Society and
the Geoscience Spaceborne Imaging Spectroscopy (GSIS)
Technical Committee
Thank you to the University of Arizona for access to their
laboratory and test site
http://www.grss-ieee.org/hyperspectral-calval-workshop-2/
http://www.grss-ieee.org/hyperspectral-calval-field-practicum/