Transcript SO 2 Detection

Product Overview
(Fog/Low Cloud Detection)
1
Example Product Output
2
Arctic Ocean
Areas of interest
Barrow
Deadhorse
MVFR Probability
Kaktovik
Arctic Ocean
Barrow
Surface observation at Barrow (in
middle of an FLS deck) shows VFR
conditions, while further east along
the Arctic Ocean coast LIFR
conditions are being reported
Kaktovik
Deadhorse
Barrow
Notice how the
traditional BTD FLS
product would show the
same signal (color) for
both Barrow, Deadhorse,
and Kaktovik
Kaktovik
Deadhorse
MVFR Probability
Barrow
The GOES-R MVFR probability product
indicates a < 50% probability of MVFR at
Barrow and a > 50% probability of MVFR at Deadhorse
Deadhorse and Kaktovik. In general, the
GOES-R product is more sensitive than the
BTD to localized changes in ceiling.
Kaktovik
Barrow
Kaktovik
Deadhorse
The GOES-R FLS depth product shows that
there is some spatial variability in cloud
depth.
FLS Depth
Cloud Type
MVFR Probability
IFR Probability
Requirements
Product
Extent
Qualifier
Cloud
Cover
Conditions
Qualifier
Product
Statistics
Qualifier
Clear conditions down
to feature of interest (no
high clouds obscuring
fog) associated with
threshold accuracy
Over low cloud
and fog cases with
at least 42%
occurrence in the
region
M - Mesoscale
TBD
Product
Measureme
nt Precision
159 sec
LongTerm
Stability
FD – Full Disk
5 min
Data
Latency
15 min
Refresh
Rate Option
(Mode 4)
C – CONUS
70%
Correct
Detection
Quantitative out
to at least 70
degrees LZA
and qualitative
beyond
Refresh
Rate/Cover
age Time
Option
(Mode 3)
Fog/No Fog
Msmnt.
Accuracy
1 km
Day and night
Msmnt.
Range
Mapping
Accuracy
2 km
FD
Temporal
Coverage
Qualifiers
GOES-R
Horiz.
Res.
Vertical
Res.
0.5 km
(depth)
Geographi
c
Coverage
(G, H, C,
M)
User &
Priority
Name
Low Cloud
and Fog
Undefined
for binary
mask
9
Requirements
Product
Extent
Qualifier
Cloud
Cover
Conditions
Qualifier
Product
Statistics
Qualifier
Clear conditions down
to feature of interest (no
high clouds obscuring
fog) associated with
threshold accuracy
Over low cloud
and fog cases with
at least 42%
occurrence in the
region
M - Mesoscale
TBD
Product
Measureme
nt Precision
159 sec
LongTerm
Stability
FD – Full Disk
5 min
Data
Latency
15 min
Refresh
Rate Option
(Mode 4)
C – CONUS
70%
Correct
Detection
Quantitative out
to at least 70
degrees LZA
and qualitative
beyond
Refresh
Rate/Cover
age Time
Option
(Mode 3)
Fog/No Fog
Msmnt.
Accuracy
1 km
Day and night
Msmnt.
Range
Mapping
Accuracy
2 km
FD
Temporal
Coverage
Qualifiers
GOES-R
Horiz.
Res.
Vertical
Res.
0.5 km
(depth)
Geographi
c
Coverage
(G, H, C,
M)
User &
Priority
Name
Low Cloud
and Fog
Undefined
for binary
mask
10
Validation Approach
11
Validation Approach
Validation sources:
1). Ceiling height at
standard surface
stations
2). CALIOP cloud
boundaries
3). Special SODAR
equipped stations
4). Fog focused
field experiments
Validation method: Determine fog
detection accuracy as a function of
MVFR probability; Directly validate
fog depth
12
Validation Results
13
Surface Observation-Based
Fog Detection Validation
Comparisons to surface observations indicate that the 70% accuracy
specification is being satisfied
Accuracy
Peirces’s Skill Score
Day (0.51)
Night (0.59)
Combined (0.59)
Day (75%)
Night (83%)
Combined (81%)
14
CALIOP-Based Fog
Detection Validation
Accuracy
Comparisons to CALIOP
indicate that the 70%
accuracy specification is
being satisfied
Peirces’s Skill Score
Daytime Accuracy: 90%
Nighttime Accuracy: 91%
Overall Accuracy 91%
Overall skill: 0.61
15
SODAR-based Fog Depth
Validation
Comparisons with SODAR/ceilometer derived fog depth
indicate a bias of ~30 m. More data points will be added to
this analysis in the future.
Daytime
Nighttime
16
CALIOP-based Fog Depth
Validation
The GOES-R fog depth product was also compared to cloud depth
information derived from CALIOP. A bias of -403 m was found.
17
FRAM-ICE RPOJECT SITE
Yellowknife, NWT, Canada
INSTRUMENTS
Tower 4
FD12P
and
Sentry
Vis
Tower 3
Tower 2
Tower 1
GOES-R Fog/Low Stratus Detection Over Yellowknife
• This is a daytime
scene covering the
same area around
Yellowknife, NWT
and Great Slave
Lake
Yellowknife
• The SW corner of
this scene is
shown to contain a
large amount of
thin, overlaying
cirrus clouds
(circled areas)
Cirrus overlapping low clouds
Yellowknife
Yellowknife
• It should be noted
that all the clouds
in this scene were
classified by the
GOES-R cloud type
algorithm as either
mixed phase or ice
clouds, which
should be the case
during an ice fog
event
Validation Results Summary
Product
Measurement
Range
Product
Measurement
Accuracy
Fog Detection
Validation Results
Product
Vertical
Resolution
Fog Depth
Validation
Results
Binary Yes/No
70% correct
detection (100%
of spec.)
1). 81% correct
detection (using
surface
observations)
2). 91% using
CALIOP
0.5 km (fog
depth)
1). SODAR
bias:
31 m (day)
25 m (night)
2). CALIOP
bias:
403 m (overall)
20
Summary
 The GOES-R ABI fog/low cloud detection algorithm provides
a new capability for objective detection of hazardous aviation
conditions created by fog/low clouds
 The GOES-R AWG fog algorithm meets all performance and
latency requirements.
 Improved ABI spatial and temporal resolution will likely
improve detection capabilities further.
 Prospects for future improvement: 1). Incorporate additional
NWP fields (e.g. wind). 2). Incorporate LEO data. 3). Working
on incorporating “valleyness” metric derived from DEM 4).
Correct for thin cirrus clouds 5). Work towards an all weather
MVFR and IFR probability capability
21
Product Overview
(SO2 Detection)
22
Example Product Output
Cordon Caulle (Chile), June 06 2011 – 17:00
23
Example Product Output
Grimsvotn, May 22, 2010 - 13:00 UTC
OMI SO2 Product
GOES-R SO2 Product
Iceland
The GOES-R and OMI products are in
good agreement (more on this later)
Iceland
24
Requirements and Product Qualifiers
SO2 Detection
Product
Statistics
Qualifier
Cloud Cover
Conditions
Qualifier
Product
Extent
Qualifier
Temporal
Coverage
Qualifiers
Geographic
Coverage
(G, H, C, M)
User &
Priority
Name
25
Product
Measurement
Precision
Vendor Allocated
Ground Latency
Product Refresh
Rate/Coverage
Time (Mode 4)
Product Refresh
Rate/Coverage
Time (Mode 3)
Measurement
Accuracy
Measurement
Range
Mapping
Accuracy
Horizontal
Resolution
Vertical
Resolution
Geographic
Coverage
(G, H, C, M)
User &
Priority
Name
Over specified
geographic area
Clear conditions down to
feature of interest associated
with threshold accuracy
Quantitative out to at
least 70 degrees LZA
and qualitative at larger
LZA
Day and night
Full Disk
GOES-R
SO2 Detection
N/A
806 sec
Full disk: 5
min
Full disk: 60
min
70% correct
detection
Binary yes/no
detection
from 10 to
700 Dobson
Units (DU)
1 km
2 km
Total
Column
Full Disk
GOES-R
SO2
Detection
Requirements and Product Qualifiers
SO2 Detection
Product
Statistics
Qualifier
Cloud Cover
Conditions
Qualifier
Product
Extent
Qualifier
Temporal
Coverage
Qualifiers
Geographic
Coverage
(G, H, C, M)
User &
Priority
Name
26
Product
Measurement
Precision
Vendor Allocated
Ground Latency
Product Refresh
Rate/Coverage
Time (Mode 4)
Product Refresh
Rate/Coverage
Time (Mode 3)
Measurement
Accuracy
Measurement
Range
Mapping
Accuracy
Horizontal
Resolution
Vertical
Resolution
Geographic
Coverage
(G, H, C, M)
User &
Priority
Name
Over specified
geographic area
Clear conditions down to
feature of interest associated
with threshold accuracy
Quantitative out to at
least 70 degrees LZA
and qualitative at larger
LZA
Day and night
Full Disk
GOES-R
SO2 Detection
N/A
806 sec
Full disk: 5
min
Full disk: 60
min
70% correct
detection
Binary yes/no
detection
from 10 to
700 Dobson
Units (DU)
1 km
2 km
Total
Column
Full Disk
GOES-R
SO2
Detection
Validation Approach
27
Validation Approach
The Ozone Monitoring
Instrument (OMI) is used as
“truth”
•OMI is a Dutch-Finnish
instrument on board the Aura
satellite in NASA’s A-Train.
•OMI uses measurements of
backscattered solar UV radiation
to detect SO2. OMI can detect SO2
at levels less than 1 DU.
•Although OMI is very sensitive to
SO2, it only views a given area of
earth once daily.
The ABI SO2 mask is validated as a
function of OMI SO2 loading
•The MODIS instrument on the
Aqua satellite (also in the A-Train)
observes the same area as OMI,
which allows us to study any SO2
clouds observed by OMI.
28
Validation Approach

OMI SO2 Quality Flag
(QF) from the OMI data
set is used to filter out
poor quality SO2 loading
retrievals
Spurious data
Before filtering
After filtering 29
Validation Approach
•The OMI SO2 loading product and
the GOES-R proxy data (MODIS or
SEVIRI) are matched up in time
and re-mapped to the same grid to
allow for quantitative
comparisons.
30
Validation Results
31
SO2 Validation
•The ABI SO2 mask
is validated as a
function of OMI
SO2 loading.
•Only scenes that
contained SO2
clouds were used
so this analysis
reflects how the
algorithm will
perform in relevant
situations
•The accuracy is
79% when the OMI
SO2 indicates 10
DU or more of SO2.
Accuracy
Accuracy Requirement
32
SO2 Validation
•While not required
the true skill score
was also
evaluated.
True Skill Score
•The true skill
score is 0.59 when
the OMI SO2
indicates 10 DU or
more of SO2.
•The true skill
score is 0.70 when
the OMI SO2
indicates 14 DU or
more of SO2.
33
Validation Results
Summary
F&PS
requirement:
Product
Measurement
Range
F&PS
requirement:
Product
Measurement
Accuracy
Validation results
Binary yes/no
detection from 10
to 700 Dobson
Units (DU)
70% correct
detection
79% correct detection
when loading is 10 DU
or greater (0.70 true
skill score when
loading is 14 DU or
greater)
34
Summary

The ABI SO2 Detection algorithm provides a new capability to objectively
detect SO2 clouds, which may be an aviation hazard and impact climate, at
high temporal resolution (this is the first quantitative geostationary capability).

The GOES-R AWG SO2 algorithm meets all performance and latency
requirements.

The improved spatial and temporal resolution of the ABI, along with a 7.3 um
band that is better suited for SO2 detection, will likely lead to improved SO2
detection capabilities (relative to SEVIRI and MODIS). ABI channel 10 was
specifically designed to help with SO2 detection (no SO2 product = wasted
instrument capability)

Prospects for future improvement: 1). Express results as a probability, not a
mask. 2). Correct for high level ice and ash clouds using bands not sensitive to
SO2 3). Improve quantitative estimates of SO2 loading and possibly height
35