Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
FIDUCEO has received funding from the
European Union’s Horizon 2020 Programme
for Research and Innovation, under Grant
Agreement no. 638822
Principles of Metrology and their applicability to
Earth Observation
Nigel Fox
Magna Carta - 1215
One of the oldest documents formalising measurement in the UK
“There is to be one measure of wine and ale
and corn within the realm, namely the
London quarter, and one breadth of cloth,
and it is to be the same with weights.” ‘measurements’ of the Earth if they are to be trusted,
meaningful and interoperable should be treated in the
same way traceable to international agreed standards
Documented methods, estimated uncertainties,
supporting evidence
For EO and Climate ECVs needs some translation &
adaptation of standards and methods:
Organisation of World
Metrology
The Convention of the Metre
(Convention du Mètre)
International System of Units (SI)
(Système International d'Unités)
Mutual Recognition Arrangement
(CIPM-MRA)
1875
1960
1999
• Identical worldwide
• Century-long stability
• Absolute accuracy
• How do we make sure a wing
built in one country fits a
fuselage built in another?
• How do we make sure the SI
units are stable over
centuries?
• How do we improve SI over
time without losing
interoperability and stability?
Three principles
Traceability
Uncertainty Analysis
Comparison
SI
Primary standard
Secondary standard
Laboratory calibration
Industrial / field measurement
Traceability
Unit definition
At BIPM and NMIs
Users
Incre
asin
g u
ncert
ain
ty
Traceability:
An unbroken chain
SI
Documented
procedures
Rigorous
uncertainty
analysis
Audits
Transfer
standards
Rigorous Uncertainty
Analysis
The Guide to the expression of Uncertainty in Measurement (GUM)
• The foremost authority and guide to the expression and calculation of uncertainty in measurement science
• Written by the BIPM, ISO, etc.
• Covers a wide number of applications
• Also a set of supplements
http://www.bipm.org/en/publications/guides/gum.html
Accredited Calibration
Laboratories
auditing procedures
transfer
standards
calibration
INDUSTRY
EURAMET
Regional
comparisons
CONVENTION OF THE METRE
Key comparison of primary unit
National Metrology Institutes SIM APMP
Mutual Recognition
Arrangement
Lab-to-lab
(results of a scientific
comparison)
Protocol
Measurements
Pre-Draft A
Draft A
Draft B
Final report
• Written by working group
• Approved by participants
• Reviewed then approved by CCPR
• Results sent to pilot only
• Often star-form
• Relative results (intra-lab consistency)
• Review of uncertainty statements
• No results shown
• Discussion on dealing with outliers (blind)
• First time participants see results
• Review by participants
• Review by experts
• Approval by CCPR
• Published
Used to:
• Validate CMCs
• Inform customers
MRA Formal comparison
Luminous Intensity key
comparison
Ongoing research:
Outliers
-2.0%
-1.5%
-1.0%
-0.5%
0.0%
0.5%
1.0%
1.5%
2.0%
0 2 4 6 8 10
Participants
De
via
tio
n f
rom
KC
RV
(all)
KCRV(subset) - KCRV(all) X_i - KCRV(all) X_i(excluded) - KCRV(all)
c2obs 20.5
c20.05(n) 15.507
Error bars: expanded unc.(k = 2)
If you exclude 5 or 7 the others
(including 7 or 5) become consistent
CCPR Guidelines on
comparisons
CMC Database
https://kcdb.bipm.org/
Evidence: Formal peer review or audit of procedures,
participation in a relevant key comparison (within 10
years) with declared uncertainties defended, review
within region and between regions
• Identical worldwide
• Century-long stability
• Absolute accuracy
Achieved through:
• Traceability
• Uncertainty Analysis
• Comparison
The traceability
chain is broken
No reference in
space …
The Quality Assurance framework for Earth Observation (QA4EO)
Looks to make the GUM accessible to the EO community
QA4EO Principle:
‘All data and derived products shall have associated
with them a fully traceable indicator of their quality’,
documented and quantitatively tied to an
international standard ideally SI
Operational framework: Principles and scope (space example)
All activities which contribute to the
delivery of an end product derived
from an input measurand
Pre-Flight
- Requirement/Design Specification
- Instrument build: characterisation/calibration
- Data processing: algorithms, ref/support data,
Post-Launch
- Instrument performance
- Output data quality characteristics:
- accuracy
- equivalence to others (sensors/in-situ)
- Processing – high level products
- Data distribution/archive …
Collection – Processing – Validation - Delivery
Archive
Reprocessed
+QI
+QI
Fiducial Reference
measurments (FRMs)
23
& have Uc levels fit for the application they are used for
Uc for Validation
measurements MUST
also be evaluated and
compared to assess
consistency with that
derived by sensor
FRM comparisons
24 www.frm4sts.org
FRM4STS lab
comparison
PHASE 1: PREPARATION
Invitation to participate October 2015
Preparation and formal agreement of the protocols Jan - March 2016
PHASE 2: MEASUREMENTS
Comparison of participants’ radiometers and blackbodies June 2016
Field comparison of participants’ radiometers at NPL June/July 2016
Participants send all data and reports to pilot July 2016
PHASE 3: ANALYSIS AND REPORT WRITING
Pre-draft A: Participants send preliminary report describing
their measurement system and uncertainties to the pilot. This
will be circulated to all participants.
April 2016
Receipt of comments from participants May 2016
Draft A (results circulated to participants) July 2016
Final draft report circulated to participants August 2016
Draft B submitted to CEOS WGCV September 2016
Final Report published October 2016
Difference between the mean of the values reported by
participating blackbodies from the values measured by
AMBER (shown in blue) and PTB (shown in red) for a
nominal blackbody temperature of 25 oC.
Traceability
Delivered through common processing chain evaluated for Uc
Site 1
RadCalNet portal
Calibration & QC
& Processing
Raw measurements
Surface reflectance and atmosphere products (RadCalNet specific)
FTP FTP
RadCalNet Processing
& QC
Hyperspectral TOA
reflectance @ 30 mn
interval for nadir view
Site 2
Calibration & QC
& Processing
Raw measurements
Surface reflectance and atmosphere products (RadCalNet specific)
QA site
owner,
NPL
support
on Uc
QA site
owner,
NPL
support
on Uc
FIDUCEO has received funding from the
European Union’s Horizon 2020 Programme
for Research and Innovation, under Grant
Agreement no. 638822
Uncertainty Analysis applied to Earth
Observation
Sam Hunt
X1
X2
X3
Y
Error effects Input quantities
Measurement model
Output quantity
Principle of
Uncertainty Analysis
1 2, , , 0NY f X X X
Measurement function
The Measurement
Equation
Uncertainty analysis
Uncertainty analysis
Uncertainty analysis
Uncertainty analysis
Uncertainty analysis
some
uncertainty in
the +0 too!
X1
X2
X3
Y
Error effects Input quantities
Locally linearised model
Output quantity
1u X
2,au X
2,bu X
2,cu X
3u X
u Y
Propagation of
Uncertainty 1 2
yn
f f f
x x x
C
1 2, , , 0NY f X X X
GUM: Law of Propagation
of Uncertainties
2 1
2 2c
1 1 1
2 ,
n n n
i i ji i ji i j i
f f fu y u x u x x
x x x
1 2y
n
f f f
x x x
C
21 1 2 1
22 1 2 2
21 2
, ,
, ,
, ,
1 2
1
2
n
E n
n n n
u E u E E u E E
u E E u E u E E
u E E u E E un E
n
U
Algebraic form
Matrix form
For each effect you need
to know
Size of uncertainty
Sensitivity coefficient
Uncertainty probability distribution
Form and scale of error correlation
• Spectrally
• Spatially
• Temporally
Error correlation form --
example 3 thermal channels, common reference blackbody (ICT),
uncertain temperature (same for all channels)
1 1 11 1 11 1 1
Correlation coefficient – channel to channel
Temperature covariance– channel to channel
𝑢 𝑇 0 00 𝑢 𝑇 00 0 𝑢 𝑇
1 1 11 1 11 1 1
𝑢 𝑇 0 00 𝑢 𝑇 00 0 𝑢 𝑇
Error correlation form
3 spectral channels, common reference blackbody (ICT),
uncertain temperature (same for all channels)
Temperature covariance– channel to channel
𝑢 𝑇 0 00 𝑢 𝑇 00 0 𝑢 𝑇
1 1 11 1 11 1 1
𝑢 𝑇 0 00 𝑢 𝑇 00 0 𝑢 𝑇
Earth radiance covariance – channel to channel
E, 1 E, 1
E, 2 E, 2
E, 3 E, 3
0 0 0 0
0 0 1 1 1 0 0
0 0 0 0 1 1 1 0 0 0 0
0 0 1 1 1 0 0
0 0 0 0
T T T T T
L L
T Tu T u T
L Lu T u T
T Tu T u T
L L
T T
C V R V C
Using this…
E, 1 E, 1
E, 2 E, 2
E, 3 E, 3
0 0 0 0
0 0 1 1 1 0 0
0 0 0 0 1 1 1 0 0 0 0
0 0 1 1 1 0 0
0 0 0 0
T T T T T
L L
T Tu T u T
L Lu T u T
T Tu T u T
L L
T T
C V R V C
Error covariance between measured radiance in different
channels due to uncertainty associated with ICT temperature
So, when you have a retrieval combining these three channels,
the error correlation affects the uncertainty via off-diagonals
E, 1 E, 2 E, 1, ,y f L L L
Earth count noise
(independent channel to
channel… )
E, 1 E, 1
E, 2 E, 2
E, 3 E, 3
E, 1 E, 1
E E
E, 2 E, 2
E E
E, 3 E, 3
E E
0 0 0 0
0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0
0 0 0
1 0 0
0 0
0 1
0
1
0
C C
CE CE CE CE CE C C
C C
L L
C Cu u
L Lu u
C Cu u
L L
C C
C V R V C
… everything stays diagonal
Y
Error effects Input quantities
Measurement model
Output quantity
Monte Carlo Approach
1 2, , , 0NY f X X X
Both Earth count and
temperature and …
ch
effects,
i i i i i
i
U C V RV C
E, 1 E, 1
E, 2 E, 2
E, 3 E, 3
1 1 1
1 1
0 0 0 0
0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0
0 0 0 0
1
1 1 1
T T T T T
L L
T Tu T u T
L Lu T u T
T Tu T u T
L L
T T
C V R V C
E, 1 E, 1
E, 2 E, 2
E, 3 E, 3
E, 1 E, 1
E E
E, 2 E, 2
E E
E, 3 E, 3
E E
0 0 0 0
0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0
0 0 0
1 0 0
0 0
0 1
0
1
0
C C
CE CE CE CE CE C C
C C
L L
C Cu u
L Lu u
C Cu u
L L
C C
C V R V C
+
FIDUCEO has received funding from the
European Union’s Horizon 2020 Programme
for Research and Innovation, under Grant
Agreement no. 638822
FIDUCEO developments for EO radiance
uncertainty
Chris Merchant
EO Radiance
Uncertainty Analysis
Understand the measurement equation for
radiance
Identify all known sources of error (effects)
Quantify their error correlations and distributions
(uncertainty)
Propagate to get radiance uncertainty
Structured approach centred on measurement
equation
The equation used to calculated “calibrated radiance” in the FCDR
Measurement equation
Gain
TOA Earth Radiance
True signal
GT
Measured Signal
C
E
M = GTRE
T +dCE
Measured gain
GM = GT +dGT
Measured Earth Radiance
Should respect the laws of physics (doesn’t always!) Should reflect the instrument
𝐶𝐸𝑀 = 𝐶𝐸
𝑇 + 𝛿𝐶𝐸
𝐿𝐸𝑇
𝐿𝐸𝑀 =𝐶𝐸𝑀
𝐺𝑀
𝐿 = 𝑓 𝐶𝐸; 𝐶𝐶𝐶𝑇, 𝐶𝑊𝐶𝑇; 𝑎1, 𝑎2, … + 0
𝐿 = 𝑓 𝐶𝐸; 𝐶𝐶𝐶𝑇, 𝐶𝑊𝐶𝑇; 𝑎1, 𝑎2, … + 0
𝜕𝐿
𝜕𝐶𝐸
𝐿 = 𝑓 𝐶𝐸; 𝐶𝐶𝐶𝑇, 𝐶𝑊𝐶𝑇; 𝑎1, 𝑎2, … + 0
𝜕𝐿
𝜕𝐶𝐸
𝑢(𝐶𝐸) digitisation 𝑢~1
2√3
𝑟~0
𝐿 = 𝑓 𝐶𝐸; 𝐶𝐶𝐶𝑇, 𝐶𝑊𝐶𝑇; 𝑎1, 𝑎2, … + 0
𝜕𝐿
𝜕𝐶𝐸
𝑢(𝐶𝐸) digitisation 𝑢~1
2√3
𝑟~0
𝑢(𝐿)~∆𝐿
2√3
𝑟~0
𝐿 = 𝑓 𝐶𝐸; 𝐶𝐶𝐶𝑇, 𝐶𝑊𝐶𝑇; 𝑎1, 𝑎2, … + 0
𝜕𝐿
𝜕𝐶𝐸
𝑢(𝐶𝐸) digitisation 𝑢~∆𝐿
2√3 detector &
amplifier noise
NOAA 16 HIRS Ch 5 (2001-2015)
Nois
e /
counts
June 1 June 30
Empirical assessment using
Allan deviation of the counts
timeseries viewing cal. targets
Plots: Gerrit Holl
𝐿 = 𝑓 𝐶𝐸; 𝐶𝐶𝐶𝑇, 𝐶𝑊𝐶𝑇; 𝑎1, 𝑎2, … + 0
𝜕𝐿
𝜕𝐶𝐸
𝑢(𝐶𝐸) digitisation 𝑢~∆𝐿
2√3
𝑟~0
detector &
amplifier noise
See poster by Gerrit Holl
𝐿 = 𝑓 𝐶𝐸; 𝐶𝐶𝐶𝑇, 𝐶𝑊𝐶𝑇; 𝑎1, 𝑎2, … + 0
𝜕𝐿
𝜕𝐶𝐸
𝑢 𝐶𝐸
𝑟~𝑒𝑚𝑝𝑖𝑟𝑖𝑐𝑎𝑙
digitisation 𝑢~∆𝐿
2√3
𝑟~0
detector &
amplifier noise
See poster by Gerrit Holl
𝜕𝐿
𝜕𝐶𝐶𝐶𝑇
𝐿 = 𝑓 𝐶𝐸; 𝐶𝐶𝐶𝑇, 𝐶𝑊𝐶𝑇; 𝑎1, 𝑎2, … + 0
𝜕𝐿
𝜕𝐶𝐸
𝑢 𝐶𝐸
𝑟~𝑒𝑚𝑝𝑖𝑟𝑖𝑐𝑎𝑙
digitisation 𝑢~∆𝐿
2√3
𝑟~0
detector &
amplifier noise
𝜕𝐿
𝜕𝐶𝐶𝐶𝑇
𝐶𝐶𝐶𝑇 = 𝑤𝑠𝐶𝐶𝐶𝑇𝑝,𝑠
𝑠𝑐𝑎𝑛𝑠𝑝𝑖𝑥𝑒𝑙𝑠
𝑢(𝐶𝐶𝐶𝑇𝑝,𝑠)
𝜕𝐶𝐶𝐶𝑇
𝜕𝐶𝐶𝐶𝑇𝑝,𝑠
𝐿 = 𝑓 𝐶𝐸; 𝐶𝐶𝐶𝑇, 𝐶𝑊𝐶𝑇; 𝑎1, 𝑎2, … + 0
𝜕𝐿
𝜕𝐶𝐸
𝑢 𝐶𝐸
𝑟~𝑒𝑚𝑝𝑖𝑟𝑖𝑐𝑎𝑙
digitisation detector &
amplifier noise
𝜕𝐿
𝜕𝐶𝐶𝐶𝑇
𝐶𝐶𝐶𝑇 = 𝑤𝑠𝐶𝐶𝐶𝑇𝑝,𝑠
𝑠𝑐𝑎𝑛𝑠𝑝𝑖𝑥𝑒𝑙𝑠
𝑢(𝐶𝐶𝐶𝑇𝑝,𝑠)
𝜕𝐶𝐶𝐶𝑇
𝜕𝐶𝐶𝐶𝑇𝑝,𝑠
Form of f.
Interpolation /
extrapolation
assumption.
𝐿 = 𝑓 𝐶𝐸; 𝐶𝐶𝐶𝑇, 𝐶𝑊𝐶𝑇; 𝑎1, 𝑎2, … + 0
𝜕𝐿
𝜕𝐶𝐸
𝑢 𝐶𝐸
𝑟~𝑒𝑚𝑝𝑖𝑟𝑖𝑐𝑎𝑙
digitisation detector &
amplifier noise
𝜕𝐿
𝜕𝐶𝐶𝐶𝑇
𝐶𝐶𝐶𝑇 = 𝑤𝑠𝐶𝐶𝐶𝑇𝑝,𝑠
𝑠𝑐𝑎𝑛𝑠𝑝𝑖𝑥𝑒𝑙𝑠
𝑢(𝐶𝐶𝐶𝑇𝑝,𝑠)
𝜕𝐶𝐶𝐶𝑇
𝜕𝐶𝐶𝐶𝑇𝑝,𝑠
Form of f.
Interpolation /
extrapolation
assumption.
Real example of measurement
equation analysis diagram
Example of AVHRR
Illustrates
• non-linear measurement
equation
• branching structure
(secondary and tertiary
measurement equations)
• uncertainty in calibration
parameters
(harmonisation
uncertainty)
For each effect you need
to know
Size of uncertainty
Sensitivity coefficient
Uncertainty probability distribution
Form and scale of error correlation
• Spectrally
• Spatially
• Temporally
Challenge of complexity...
Investigating each source
of uncertainty
Error correlation between
measured values (Type B)
Earth view Calibration target view Space view
Error correlation between
measured values
Earth view Calibration target view Space view
Error correlation between
measured values
Earth view Calibration target view Space view
Error correlation between
measured values
Earth view Calibration target view Space view
Error correlation
dimensions
LEO:
Pixel-to-pixel
Scanline-to-scanline
Orbit-to-orbit
Temporal
Spectral
GEO:
Pixel-to-pixel
Scanline-to-scanline
Image-to-image
Temporal
Spectral
Codify the uncertainty analysis
Name of effect Earth Count Noise
Averaged Space Count Noise
Averaged IWCT Count Noise
Affected term in measurement
function
CE CS Ct
Instruments in the series affected All All All
Correlation
type and form
Pixel-to-pixel [pixels] Random* Rectangular Absolute
Rectangular Absolute
from scanline to scanline
[scanlines]
Random* Triangular Triangular
between images
[images]
N/A N/A N/A
Between orbits [orbit] Random Random Random
Over time [time] Random Random Random
Correlation
scale
Pixel-to-pixel [pixels] [0]
from scanline to scanline
[scanlines]
[0] n = 51 n = 51
between images
[images]
N/A N/A N/A
Between orbits [orbit] [0] [0] [0]
Over time [time] [0] [0] [0]
Channels/
bands
List of channels / bands
affected
All All All
Correlation coefficient matrix Identity
matrix (1s
down
diagonal only)*
Identity matrix (1s
down diagonal only)*
Identity matrix (1s
down diagonal only)*
Uncertainty PDF shape
Digitised Gaussian
Digitised Gaussian Digitised Gaussian
units Counts Counts Counts
magnitude Provided per
pixel
Provided per scanline
Provided per scanline
Sensitivity coefficient , Eq 4-1 , Eq 4-2 , Eq 4-3
For each effect (“end twig” of
tree), we formalise the
conclusions of the uncertainty
analysis.
For documentation, the analysis is
documented and summarised in
an effects table.
The set of codified effects table
+ satellite data + harmonisation
results comprise the “full FCDR”.
Error correlation forms
Rectangle Absolute
• Fully systematic or systematic within a calibration
period
Triangle Relative
• Rolling averages
Bell-shaped Relative
• Weighted rolling averages, splines, smoothing,
other
Repeating
• E.g. once per orbit, diurnal or seasonal cycles
Mixed
Table descriptor Comments Example
Name of effect A unique name Internal calibration target count
noise
Affected term in measurement function Name and standard symbol
Instruments in the series affected Identifier All instruments all satellites
Correlation type
and form
Pixel-to-pixel [pixels] One of the types Rectangular absolute from scanline to scanline
[scanlines] Triangular relative
between images
[images] N/A for orbiting satellite
Between orbits [orbit] Random Over time [time] Random
Correlation scale Pixel-to-pixel [pixels] As needed to define type [-∞,∞] (fully correlated across
scan) from scanline to scanline
[scanlines] n = 51 (51 scanlines averaged
in rolling average) between images
[images] N/A for orbiting satellite
Between orbits [orbit] 0 Over time [time] 0
Channels/bands List of channels / bands
affected Channel names All channels
Error correlation coefficient
matrix A matrix Identity matrix (diagonal).
Uncertainty PDF shape
Functional form Gaussian
units Units Counts
magnitude Given once per orbit file
Sensitivity coefficient Value, equation or
parameterisation of sensitivity
of measurand to term
ICTC
E
ICT
L
C
E
ICT
L
C
Why a metrological approach?
Why consider all sources of uncertainty?
See blog article http://www.fiduceo.eu/node/237
If you compare two measurements on different space-time scales the dominant sources of uncertainty in that difference change.
Why a metrological approach?
Why consider all sources of uncertainty?
Example
of SST CDR
Specifying an “accuracy” target here …
... only weakly constrains
the uncertainty here
FIDUCEO has received funding from the
European Union’s Horizon 2020 Programme
for Research and Innovation, under Grant
Agreement no. 638822
Applying FIDUCEO thinking to AVHRR
Jon Mittaz
CURRENT STATUS QUO:
Without a FIDUCEO
approach
No pixel level uncertainty estimates with data – just the design specification
AVHRR IR Channels over whole series life (1978-present) have had 4 different operational calibration algorithms
• Linear calibration (even though the 11/12 micron channels are non-linear): TIROS-N to NOAA-8
• Non-linear correction from lookup table: NOAA-9 to NOAA-12
• Non-linear correction from lookup tables including a “negative radiance of space” term: NOAA-14
• Walton et al. (1998) calibration: NOAA-15 to present(may apply to some NOAA-14 as well)
• Coefficients for older AVHRRs are available from Walton et al. for AVHRR/2 and AVHRR/3 instruments (not for AVHRR/1 – TIROS-N, NOAA-06, NOAA-8, NOAA-10)
Difficult to make a consistent time series from the data
Many different source of error are still present in the operational data
If a metrological approach had been taken from the beginning these
problems would have been significantly reduced
AVHRR problems start
with pre-launch
Not done well… Very simple test environment
• Goal to meet design specifications only
• AVHRR/1 and AVHRR/2 IR channels good to 1K only
• AVHRR/3 IR channels good to 0.5K
Calibration Targets -
ECT (180->320K) &
Space Target @ 70K
Run at 5 instrument temperatures
of 10, 15, 20, 25, 30°C (15,20,25°C
for early sensors + NOAA-16)
Not exactly a high tech setup…
AVHRR operational
calibration (Walton et al.)
Derived on the basis of the pre-launch data
For 11&12 µm channels (non-linear) a linear estimate of the radiance id first derived and
the radiance is then correct for the non-linearity
2
210 LinLinEarthS
BBS
SBBSLin NaNaaRadianceCC
CC
NNNN
NS is the ‘Negative Radiance of Space’ and a0, a1 and a2 are coefficients derived from pre-launch
calibration data
It looks like it
works well
(standard
deviation from
pre-launch ~
0.08K 11µm
channel)
BUT Taken from Walton et al. (1998)
Applying the operational calibration to pre-launch data (AVHRR/3,
Walton et al. 1998)
(which it was derived from in the first place…)
You’d hope you’d get the
‘right’ answer
• Actually see biases of
up to 1K…
Problems with pre-launch
data
Non-physical aspects of
Walton et al.
• Just at face value there are a number of problems
with the calibration equations
• Uses a “Negative Radiance of Space” whatever that
is…
• Total equation has variable non-linear term determined
from the way algorithm works rather than being based
on any physical understanding of the detector system
itself
2
2
2
21
2
210
)(
)()21())1((
EarthS
BBS
SBB
EarthS
BBS
SBBSSS
Earth
CCCC
NRa
CCCC
NRNaaNaNaa
R
THE FIDUCEO method of trying to understand the instrument from using a
physics based approach would not had had these issues
AVHRR: Problems with pre-
launch measurements
1. Instrument temperature drifted by ~ 1K but assumed
constant in deriving calibration
2. The test chamber was not temperature controlled
ICT
Calibration model including extra
component
Calibration model including extra
component
0.1K
0.5K
0.3K
Mismatch between measured ICT
temperature (PRT) and inferred ICT
temperature (ECT Calibration) (PRT
coefficient correction proposed in
1991 which would have broken PRT
traceability)
AVHRR
Test
Chamber
Taking a metrological approach from the start would have spotted these
problems
AVHRR Operational
Algorithm: Impacts on in-
orbit calibration
AVHRR pre-launch data had a lot of problems which
were not noted when developing operational
calibrations including the current one
No account was taken of the possible change in
calibration between the pre-launch testing and the in-
orbit environment
• Pre-launch data was run so that the AVHRR was
always thermally stable
• This is not the case when in orbit…
Along with pre-launch a detailed analysis of in-orbit
data should be undertaken beyond checking the
design specifications
Things can (do) change when you launch an instrument into space and
need to check for such changes
• AVHRR comparisons against
IASI show strong trends in
operational calibration
• Refitting calibration gives
significant improvements
• Demonstrates that you must
monitor the instrument in-
orbit for changes from pre-
launch data
• Thermal environment will
be different
AVHRR: Problems with in-
orbit data
This bias is still present in the operational AVHRR Level 1B
AVHRR: FIDUCEO, a
metrological approach
At Level 1 we start with the “Traceability Tree”
• Starts with the measurement equation
• Looks at each term and breaks it down into however many underlying processes are needed to get back to root process
• Links lowest level processes to their impact and associated uncertainty on the observed Earth radiance
Correlated error terms are also considered
Note that the process of obtaining metrologically traceable uncertainties also means removing all systematic error sources as far as possible
• Guide to the expression of Uncertainty in Measurement (GUM) (2008) Section 3.2.4
• It is assumed that the result of a measurement has been corrected for all recognized significant systematic effects and that every effort has been made to identify such effects.
OTfCaCC
CaRaaR InstrEE
T
TTE
)(2
2
2
210
AVHRR: FIDUCEO, a
metrological approach
AVHRR: Noise Effects
Detector noise case has been shown earlier but note
that currently many people are using a constant
NeDT=0.12K
• This value is the design specification and has
very little to do with the true behaviour of the
instrument
NOAA-07
FIDUCEO has forced an investigation of all effects and highlights time
variable effects not highlighted in the operational calibration.
AVHRR: Solar
contamination
Solar contamination of the Internal Calibration Target
(ICT) is a big problem
• Direct solar radiance contaminates the calibration
system
• Impact of solar radiation puts complex gradients
across ICT impacting its accuracy
There is an operational correction available (from
1995 onward) which tries to correct the direct solar
radiance part only
• Simple detection algorithm only, not based on
physical processes on-board
Improved solar
contamination modelling
Operational detection
(can still fail) Improved Modelling Operational modelling No operational detection
pre late 1994
Improvements from FIDUCEO due to using a physics
based model of how the instrument behaves
FIDUCEO
Operational
Orbit Drift effects
As the satellite orbit drifts there is a change in its
thermal state
• Not considered in current operational calibration
Due to changes in stray light components as the
thermal structure of the instrument changes over time
impacting the calibration
• Leads to time dependent biases which will
significantly impact geophysical retrievals such as
SST
An SST example
From Pathfinder SST
(v6) which kept SST
retrieval coefficients
constant. Operational
calibration used.
Note strong bias as
a function of time
Related to instrument
temperature (a proxy
for thermal
environment)
Can be modelled
(and is in FIDUCEO)
FIDUCEO and the AVHRR
By applying FIDUCEO principles to the AVHRR we
have gained
• Justifiable and defensible uncertainties at the
pixel level and beyond including error correlations
• An understanding of the instrument and all
associated effects which include
• Significantly reduced scene temperature biases
• Identified and significantly reduced orbit drift effects
• Significantly reduced impact of solar contamination
• …
Leads to improved AVHRR data as well as
traceable multi-component uncertainties
FIDUCEO has received funding from the
European Union’s Horizon 2020 Programme
for Research and Innovation, under Grant
Agreement no. 638822
Building new uncertainty concepts into current
and future missions
Chris Merchant
Uncertainty for CDRs
Provide uncertainty estimates
Follow metrological conventions
Give u per datum if necessary
Uncertain ≠ Bad quality
Explain the uncertainty info
Give advice to users on usage
Validate the uncertainties
Error correlation matters
DOI 10.5194/essd-9-511-2017
Use of radiance
uncertainties
For model-observation comparisons in “observation space”
For data assimilation, helping to build first-principles error
covariance estimates to confront/improve estimates
inferred within DA system
For proper estimation of Climate Data Record
uncertainties across spatio-temporal scales
• FIDUCEO exemplars – coming next year
Raw satellite data (L0)
Calibrated radiances (L1)
Climate data record (L2)
Gridded CDR (L3)
Analysed / processed (L4+)
Climate index / information
• Decision
• Insurance
• Liability
Evolve good practice towards … Good practice Apply to
Level 1 radiance provided with uncertainty
estimates per datum.
Key heritage sensor series.
Planned missions.
Multi-mission series should be harmonised. Key heritage sensor series.
Planned missions.
Propagate radiance uncertainties to inform
level 2 (swath) and 3 (gridded) geophysical
data.
Climate data records (CDRs)
and environmental data
records.
Propagate CDR uncertainty to higher-levels. Climate information derived
(in part) from CDRs
Decision makers and other users access and
trust information on uncertainty.
Presentation of climate
information in climate
services.
Not just for heritage
sensors
We think the principles and techniques we are learning
on the historical sensors have much wider applicability
Particularly, they can be embedded into space agency
practice for adding value by including per-datum
uncertainty in L1
• reprocessing of archive mission data
• specification of instruments and products for
future missions
An uncertainty/traceability focus in
Phase B-D Aspect Compliance focus Metrology focus Estimating the
magnitude of
pixel-level
uncertainty (e.g.,
in radiance)
Worst-case combination of
uncertainty from error sources
to compared against a
(generally) aggregated total
uncertainty requirement.
Deliberately pessimistic to
ensure compliance and
acceptance.
Individual
models/calculations of
uncertainty from error
sources, traceably
documented per error
source. Realistic combination
to inform expected in-flight
characteristics. Characterising
the error-
correlation
structure across
pixels and
channels
Only in response to specific
relevant requirements (e.g.
cross-talk limits). Not
considered for many error
sources.
Integral part of uncertainty
characterisation for all error
sources
An uncertainty/traceability focus in
Phase B-D Aspect Compliance focus Metrology focus Traceably documenting uncertainty information
Documentation focused on acceptance milestones. Results perhaps mixed with commercially sensitive and confidential material, usually not available in a form supporting traceability
Documentation freely available and organised such as to support systematic traceability
Dissemination of understanding of error sources to users
Not actively or systematically attempted -- generic information may be published. Not quantitatively integrated into satellite products
Understanding is embedded in product processing chain in order to include quantitative uncertainty information directly in satellite products at L1
http://www.fiduceo.eu/blogs
Beyond FIDUCEO – link to “Green paper”
Satellite missions: metrological upgrade
Harmonisation and Recalibration
Why worry about all sources of errors?
The National Physical Laboratory
is operated by NPL Management
Ltd, a wholly-owned company of
the Department for Business,
Energy and Industrial Strategy
(BEIS).
Thank you
MetEOC and MetEOC-2 were funded by EMRP MetEOC-3 is funded under EMPIR
FIDUCEO has received funding from the European
Union’s Horizon 2020 Programme for Research and
Innovation, under Grant Agreement no. 638822
Discussion 1
Each table please spend 10 mins on each Q. Identify
a rapporteur/notetaker to record main points and feed
back in plenary.
Q1 What degree of need for improved,
transparent uncertainty information is recognised
amongst users/product/service developers?
Q2 What are benefits and challenges to applying
EO-metrology principles to L1 and L2?
Q3 Is the current approach to instrument
uncertainty characterisation and pre-flight cal/val
adequate (from point of view of ultimate users of
L1 and derived data)? If no, what problems are
caused?
Discussion 2
Full set of discussion questions will be: Q1 What degree of need for improved, transparent uncertainty information
is recognised amongst users/product/service developers?
Q2 What are benefits and challenges to applying EO-metrology principles
to L1 and L2?
Q3 Is the current approach to instrument uncertainty characterisation and
pre-flight cal/val adequate (from point of view of ultimate users of L1 and
derived data)? If no, what problems are caused?
Q4 - What should next case studies be for L1?
Q5 - What priority case studies should we address next for L1 to L2+ ?
Q6 What additional information from instrument dev and pre-flight cal
should be made available to users and how?
Q7 How could we build the core principles of providing uncertainty into
the development of phase of new missions?
Q8 Are there additional steps that can be built into in-flight operational
missions to validate and test performance?
Q9 What activities/strategies do we need to consider to validate Uc of L1
and L2 products and ensure their interoperability?
Q10 Is targeted training on Uc analysis needed, and how to develop this?
Discussion 2:
Q4 - What should next case studies be for L1?
Q5 - What priority case studies should we
address next for L1 to L2+ ?
Discussion 3
Q6 What additional information from instrument
dev and pre-flight cal should be made available to
users and how?
Q7 How could we build the core principles of
providing uncertainty into the development of
phase of new missions?
Q8 Are there additional steps that can be built
into in-flight operational missions to validate and
test performance?
Discussion 4
Q9 What activities/strategies do we need to
consider to validate Uc of L1 and L2 products
and ensure their interoperability?
Q10 Is targeted training on Uc analysis needed,
and how to develop this?