NESDIS Snowfall Rate Product and its Applications
Part 1: Huan Meng1, Jun Dong2, Cezar Kongoli2, Ralph Ferraro1 , Banghua Yan1
1NOAA/NESDIS/Center for Satellite Applications and Research
2ESSIC/Cooperative Institute for Climate and Satellites – Maryland
Part 2: Kris White3, Emily Berndt3, Bradley Zavodsky3
3NASA/MSFC/Short-term Prediction Research and Transition Center
JPSS Seminar, November 20, 2017
Project Overview
• A project supported by the JPSS PGRR
Program;
• Project period: 2015-2017
• Objective:
Enhancement of the ATMS and AMSU/MHS
land Snowfall Rate (SFR) product
Expansion of the SFR suite to support
applications
• Algorithm enhancement
Snowfall detection
Snowfall rate
• SFR product from new sensors/satellites
SSMIS aboard DMSP F16, F17, and F18
GMI aboard NASA GPM
Composite NEXRAD
Reflectivity
Retrieved Snowfall
Rate
2
ATMS and MHS SFR
• ATMS: Advanced Technology Microwave
Sounder
AMSU: Advanced Microwave Sounding Unit
MHS: Microwave Humidity Sounder
ATMS is the successor to AMSU/MHS
• AMSU/MHS SFR is a NOAA operational
product, four satellites: NOAA-18,-19 and
Metop-A,-B (AMSU/MHS)
• Compared to AMSU/MHS, ATMS has improved
sampling configuration and more channels,
especially two more water vapor channels
suitable for precipitation retrieval.
• S-NPP ATMS SFR was developed with the
support of the JPSS PGRR Program
• ATMS SFR outperforms AMSU/MHS SFR
AMSU/MHS ATMS Ch GHz Pol Ch GHz Pol
1 23.8 QV 1 23.8 QV
2 31.399 QV 2 31.4 QV
3 50.299 QV 3 50.3 QH
4 51.76 QH
4 52.8 QV 5 52.8 QH
5 53.595 ± 0.115 QH 6 53.596 ± 0.115 QH
6 54.4 QH 7 54.4 QH
7 54.94 QV 8 54.94 QH
8 55.5 QH 9 55.5 QH
9 fo = 57.29 QH 10 fo = 57.29 QH
10 fo ± 0.217 QH 11 fo±0.3222±0.217 QH
11 fo±0.3222±0.048 QH 12 fo±
0.3222±0.048
QH
12 fo
±0.3222±0.022
QH 13 fo±0.3222±0.022 QH
13 fo±
0.3222±0.010
QH 14 fo±0.3222
±0.010
QH
14 fo±0.3222±0.004
5
QH 15 fo±
0.3222±0.0045
QH
15 89.0 QV
16 89.0 QV 16 88.2 QV
17 157.0 QV 17 165.5 QH
18 183.31 ± 1 QH 18 183.31 ± 7 QH
19 183.31 ± 3 QH 19 183.31 ± 4.5 QH
20 191.31 QV 20 183.31 ± 3 QH
21 183.31 ± 1.8 QH
22 183.31 ± 1 QH
(Table by Bill Blackwell)
3
Algorithm Methodology
Two major components in the SFR algorithm
• Snowfall Detection (SD)
Statistical approach (Kongoli et al., 2015, 2017)
Coupled principal component and logistic regression model
• Snowfall Rate
Physical model (Meng et al., 2017; Ferraro et al., 2017)
1DVAR-based retrieval
• Issues
Weakened signal from shallow-cloud snowfall and snowfall with
supercooled cloud liquid water due to emission effect
Missed snowfall
Underestimated snowfall rate
4
SD Enhancement (1/3)
Previous SD model - satellite module
Predictors: High frequency channels above 89 (MHS)/88.2
GHz (ATMS), 5 from MHS and 7 from ATMS; 3 principal
components
Two temperature regimes: warm and cold, defined by
53.6 GHz
Model output is the probability of snowfall
Training data sets are composed of matching satellite data
and ground snowfall observations, i.e. ‘truth’ data
5
Environment-based weather module
Logistic regression model
Input data: RH, V-Vel, CThick
Blended model: The SD algorithm is an optimal combination of
the satellite module and weather module
Additional model-based screening to improve the accuracy of the
SD algorithm
SD Enhancement (2/3)
Example of Shallow-Cloud Snowfall
6
The combined SD improves statistics for
both shallow and thick-cloud snowfall
ATMS SD statistics against gauge
snowfall observations
*The statistics for the satellite-only SD algorithm
62% of ground truth data is ‘trace’, i.e. very light
snowfall – very challenging to detect from satellite
observations
Previous SD
Radar Reflectivity
Combined SD
SD Enhancement (3/3)
Warm Regime Cold Regime
POD (%) 51 (41*) 52 (30)
FAR (%) 5 (4) 9 (6)
HSS 0.5 (0.43) 0.42 (0.28)
7
Snowfall Rate Algorithm (1/2)
1D variational method
Forward simulation of Tb’s with a radiative transfer model (RTM)
(Yan et al., 2008)
Iteration scheme with ΔTBi thresholds
IWP and De are retrieved when iteration stops
Ic: ice water path
De: ice particle effective diameter
i: emissivity at 23.8, 31.4, 89(MHS)/88.2(ATMS), 157/165.5, and 190.31/183±7 GHz
TBi: brightness temperature at 23.8, 31.4, 89/88.2, 157/165.5, and 190.31/183±7 GHz
A: Jacobian matrix, derivatives of TBi over IWP, De, and i
E: error matrix
176/190
165/157
88/89
31
23
1
176/190
165/157
88/89
31
23
)(
B
B
B
B
B
TT
e
c
T
T
T
T
T
AEAA
D
I
8
Terminal velocity is a function of atmospheric conditions and
ice particle properties, Heymsfield and Westbrook (2010):
Snowfall rate model (Meng et al., 2017):
,
An adjusting factor, a, to compensate for non-uniform ice
water content distribution in cloud column; derived from
collocated satellite and StageIV radar and gauge combined
hourly precipitation data
Snowfall Rate Algorithm (2/2)
9
Snowfall Rate Enhancement
Ice only: r = 0.44
MRMS
Recalibration using NSSL Radar Multi-
Sensor (MRMS) snowfall data as ‘truth’
Better spatial (via integration) and
temporal compatibility than StageIV
The radiative transfer model (RTM) in the
current SFR algorithm does not include
the effect of cloud liquid water
The RTM has been modified to include
CLW
Leads to increased SFR in most cases –
mitigate the dry bias in SFR
Developing more robust initialization of
cloud properties
Ice + liquid: r = 0.50
10
• A major nor’easter swept through the Mid-
Atlantic and the Northeast on March 14-15,
2017
• The ATMS and MHS SFR products captured the evolution of the snowstorm with five satellites including S-NPP, POES and Metop.
March 14-15, 2017 Nor’easter 24-hour snowfall accumulation
ending March 15, 2017 12 UTC
NOHRSC
SFR
(Courtesy of Patrick Meyers)
Corr Coeff
Bias (mm/hr)
RMS (mm/hr)
ATMS 0.68 0.05 0.71
MHS 0.65 -0.13 1.01
11
Courtesy of Patrick Meyers
ESSIC/CICS-MD
12
Expand SFR to using DMSP SSMIS and NASA GMI sensors
Snowfall is highly dynamic
Essential to utilize all available passive microwave sensors with
high frequencies to improve SFR temporal resolution
SFR Expansion
GPM
MOA
MOB
N18
F18
S-NPP/JPSS-1
{
F16 N19
F17
13
SSMIS SD (1/2)
• SSMIS is aboard three DMSP satellites:
F16, F17, and F18
• Conical scanning imager/sounder; different from
ATMS and MHS which are cross scanning
sounders
• SD model for individual satellite Channel failure
Different sensor characteristics
• Similar algorithm framework as ATMS SD
Satellite-based module
Blended algorithm: Satellite module + weather
module
Additional screenings
• Ground observations as training data set
Channel Freq (GHz) Pol FOV
1 19.35 H 45x74
2 19.35 V 45x74
3 22.235 V 45x74
4 37 H 28x45
5 37 V 28x45
6 50.3 H 37.5
7 52.8 H 37.5
8 53.596 H 37.5
9 54.4 H 37.5
10 55.5 H 37.5
11 57.29 RC 37.5
12 59.4 RC 37.5
13 63.283248 ±
0.285271 RC 75
14 60.792668 ±
0.357892 RC 75
15
60.792668 ±
0.357892 ±
0.002
RC 75
16
60.792668 ±
0.357892 ±
0.0055
RC 75
17
60.792668 ±
0.357892 ±
0.016
RC 75
18
60.792668 ±
0.357892 ±
0.050
RC 75
19 91.665 H 13x16
20 91.665 V 13x16
21 150 H 13x16
22 183.311 ± 1 H 13x16
23 183.311 ± 3 H 13x16
24 183.311 ± 6.6 H 13x16
14
SSMIS SD (2/2)
• Performance against ground observations
POD (%) FAR (%) HSS
F16 57 17 0.41
F17 48 15 0.34
F18 55 16 0.38
• Algorithm difference from ATMS SD
Satellite-based module: Logistic regression model; no principle
analysis due to less water vapor sounding channels
52.8 and 53.6 GHz significantly improve F16 and F18 model
performance
One temperature regime
15
SSMIS SFR
Physically-based SFR algorithm, similar to ATMS
SFR
1DVAR to retrieve cloud properties
Modified RTM: bias correction, scattering property LUT
Frequencies: 22, 37V, 91V, 150 (F16 and F17), 183±6.6 GHz
Algorithm for individual satellite
Algorithm calibration
NSSL MRMS
SFR and MRMS spatial collocation through convolution
Time lag by 30 min based on statistical analysis
Histogram matching
Corr Coeff Bias
mm/hr
RMS
mm/hr
F16 0.49 -0.10 0.93
F17 0.57 -0.07 0.85
F18 0.43 0.00 0.99
16
GMI SD
• GMI is the radiometer aboard NASA
Global Precipitation Mission (GPM)
core satellite
• Conical scanning radiometer with high
spatial resolution
• Similar algorithm framework as ATMS
SD model
Blended algorithm
Two temperature regimes
Channel Freq (GHz) Pol IFOV
(km x km)
1 10.65 V 19x32
2 10.65 H 19x32
3 18.7 V 11x18
4 18.7 H 11x18
5 23.8 V 10x16
6 36.5 V 9x16
7 36.5 H 9x16
8 89 V 4x7
9 89 H 4x7
10 166 V 4x6
11 166 H 4x6
12 183.31±3 V 4x6
13 183.31±7 V 4x6 Warm Regime Cold Regime
POD (%) 68 56
FAR (%) 15 14
HSS 0.55 0.45
17
GMI SFR
Physically-based SFR algorithm
1DVAR to retrieve cloud properties
Modified RTM: bias correction, scattering property LUT
Frequencies: 23, 36V, 89V, 166V, 183±7 GHz
Algorithm calibration
NSSL MRMS
SFR and MRMS spatial collocation through convolution
Time lag by 30 min based on statistical analysis
Histogram matching
Corr Coeff Bias
(mm/hr)
RMS
(mm/hr)
0.52 -0.10 0.84
MRMS
GMI SFR
18
Towards a LEO+GEO SFR
Fusing LEO MW and GEO IR data
Take advantage of the accuracy of MW
product and the temporal resolution of
IR data
Track snowstorm movement between
LEO overpasses using GEO data
Identify trend in snowfall intensity
between LEO overpasses using GEO
data
GOES-R3 project
Part of the CICS GOES-R3 Total Water
project
Performance period: FY17-FY18
Explore the fusion of SFR and GOES-
16 products and imagery
SFR overlaid on GOES IR with METAR reports (Courtesy of Bill Line)
17:05Z
19:40Z
19
Radar and Satellite Merged SFR (mSFR)
• Merging MRMS instantaneous snowfall product and SFR provide better
spatial and temporal (10-min) coverage and ability to loop the data (mSFR)
• Developed with the support of NASA/SPoRT
20
Radar and Satellite Merged (mSFR) m
m/h
r (liquid
equiv
ale
nt)
21
• Hydrology
Most blended satellite precipitation datasets do
not include satellite snowfall rate product – use
other data sources (model, ground
observations, etc.)
CMORPH is a NOAA global blended
precipitation analysis product with wide-ranging
applications
The first generation CMORPH only has rain
rate. The SFR product is integrated in the
second generation pole-to-pole CMORPH
• Weather forecasting
Fill observational gaps in mountains and
remote regions where radar and weather
stations are sparse or radar blockage and
overshooting are common
Provide quantitative snowfall information to
complement snowfall observations or
estimations from other sources (stations, radar,
GOES imagery data, etc.)
Applications
(Xie and Joyce, NOAA/NCEP/CPC)
Stage IV
Radar Precip
2nd Generation
CMORPH
Snowfall
Rainfall
22 22
User Interaction and Support
Near real-time SFR and mSFR production at CICS-MD using
Direct Broadcast (DB) data from University of Wisconsin
Madison
SFR production within 30 min of observation
SFR and mSFR data provided to SPoRT where the products
are converted to AREA file and AWIPS and disseminated to
NWS WPC and some WFOs
Support SFR assessment
User training
Interaction with users during assessment
SFR webpages with near real-time images
DB-based: http://cics.umd.edu/sfr
Global: https://www.star.nesdis.noaa.gov/corp/scsb/mspps_backup
/sfr_realtime.html
23
Other Notes
ATMS SFR added to JPSS L1RD
Requirement for transition to operation
A year-long effort with approval from authorities such as SPSRB,
PCB, NOSC, and DUSO
S-NPP SFR will be integrated in the MiRS system and
transitioned to operation
Code modification to follow SPSRB standard
Pending PSDI review
Publications
Two published papers
Two conditionally accepted papers
SFR assessment in the coming winter
24
Summary
A blended snowfall detection algorithm was developed for ATMS and
MHS; it improves snowfall detection for both shallow- and deep-cloud
snowfall
The radiative transfer model has been modified to include cloud
liquid water; ongoing work on the initialization of cloud properties
Developed the SSMIS snowfall detection and snowfall rate
algorithms; adding three DMSP satellites
Developed the GMI snowfall detection and snowfall rate algorithms;
adding NASA GPM satellite
ATMS SFR has been added to JPSS L1RD; S-NPP SFR will be
transitioned to operation inside MiRS in the near future
Product assessment in winter 2017-2018 for the full-suite of SFR
from nine satellites; effort led by SPoRT with participation from NWS
WFOs
25
Application and Assessment of NESDIS Snowfall Rate
Kris White, Emily Berndt, Brad Zavodsky
NASA SPoRT
JPSS Science Seminar – Part 2
20 November 2017
End User Interactions
Winter 2014 assessment: January to mid April 2014
Goal: determine operational utility in the forecaster
environment as it relates to:
radar gaps
beam blockage/overshooting
tracking snowfall rate maxima (in combination with other satellite
imagery)
Participating Offices
Albuquerque, NM
Burlington, VT
Charleston, WV
Sterling, VA
Satellite Analysis Branch
27
Product in 2014
AMSU/MHS from NOAA-18, -19, MetOp-A, -B
Up to 8 SFR retrievals per day at mid-latitudes
Land only retrievals
Limited to regions with surface temperature > 22°F
30 to 90 minutes time lag between retrieved snow and snow reaching the ground
Detectable snowfall rates from 0.004 in/hr to 0.2 in/hr (2 in/hr if snow to liquid ratio is 10:1)
Processed near real-time at NOAA/NESDIS with a 30 min to 3 hour latency
SPoRT made data AWIPS/NAWIPS compatible for dissemination to NWS
Module and Quick Guide developed
28
Results
Heavy snowfall events in warmer temperature regimes were
captured well
13-14 April 2014 a fast-moving upper level trough and a
backdoor cold front moved south from the eastern plains of
Colorado to New Mexico
“The 0429Z SFR product
has the greatest values
observed in NM for this
event. Our Clayton observer
(CAO – orange oval) did call
in at 06Z with a report of 1.5
inches of snow. We didn’t
receive snowfall at the Las
Vegas ASOS (LVS – pink
oval). Another spotter call
from 05Z reported 2.5
inches of snow at a location
near the purple arrow.” –
Albuquerque, NM WFO.
29
Results
Inability to detect lighter snowfall amounts that were detected
by ASOS
“It looks like the SFR
product did not detect all of
the snow that was falling
around 11 UTC. But the
misses can either be
described as either (1) the
surface temperature being
too cold or (2) the
probabilistic model that is
part of the calculations,
indicating probabilities were
too low to determine if there
was snow” – Charleston,
WV WFO.
30
Results
10:47 UTC, no snow at BHM (red oval)
1619 UTC, snowing at BHM (black circle)
• Under the right conditions, SFR can be used as a short term forecast
product
• In-cloud snow not reaching the surface can seed existing clouds to
increase the likelihood of snow reaching the surface
At 1047 UTC, SFR showed
snow in clouds but no snow
was observed at surface in
Birmingham; in-cloud seeding
was occurring
At 1553 UTC, snowfall was
reported at BHM and later
intensified; SFR product, IR,
and NEXRAD all reported
snow by 1619 UTC
More details:
https://nasasport.wordpress.co
m/2014/02/24/birmingham-
alabama-surprise-snow-of-
january-28-2014-or-was-it/
31
Feedback Summary
Product was limited at times by its latency but forecasters
found it valuable in a operational sense and to validate
snow reports
Overall feedback was positive with more than 75% of
responses indicating the product was useful to improve
data coverage in areas with radar gaps and in
combination with satellite observations to track snowfall
maxima
Limitations were uncovered
Lighter snowfall rates not detected
Sometimes missing snowfall captured by radar
32
Recommendations to Developers
Reduce the latency to < 60 minutes
Explore improving the low snowfall rate detection
efficiency (an increased false alarm rate may be
acceptable if more events are captured)
Investigate the ability to retrieve snowfall rates when
surface temperatures are colder. This would make it
easier for northern WFOs to use the product and enable
use in Alaska
33
End User Interactions
Winter 2016 assessment: January to February 2016
Goal: determine operational utility in the forecaster environment as it relates to: radar gaps
beam blockage/overshooting
tracking snowfall rate maxima (in combination with other satellite imagery)
Determine areas where cloud seeding may be occurring ahead of falling precipitation
Participating Offices Albuquerque, NM
Anchorage, AK
Juneau, AK
Boulder, CO
Charleston, WV
Sterling, VA
34
New Product Developments
Added ATMS
Alaska product developed
Generated from Direct Broadcast data to reduce latency to within 30 minutes
Snow to liquid ratio options
10:1
18:1
35:1
Merged snowfall rate product
Polar orbiter swath complimented with NSSL’s Multi-Radar/Multi-Sensor precipitation data
Product updated every 10 minutes
Put in image 1 animation from
the blog
https://nasasport.wordpress.co
m/category/nesdis-snowfall-
rate/page/2/
35
Results
Rain to snow transition
event 15 Feb. 2016
where rainfall mixed
with and then
transitioned to all snow
“Much of the precipitation across
West Virginia was still in the form
of rain…with an area of snow
extending from northwest PA
across Ohio into southwest
portions of the state. There
appears to be several
observations of rain across Ohio
with surface temperatures of 32
to 35 DegF where the SFR
product indicated snow in the
clouds” – Charleston, WV WFO.
36
Results
Enhanced snowfall
detection in areas with
lack of radar coverage
– case from NW New
Mexico, 4-5 Jan 2016
“The arrival of a SFR product at
0010 UTC 5 January
2016 showed the extent of the
precipitation was much greater
with the merged POES image
overlaid on the radar data.” –
Brian Guyer, WFO ABQ
37
Results
Although more the 75% of responses indicated low to
medium confidence in SFR values, 75% of responses
indicated the product was useful
Most responses indicated SFR was used to identify
snowfall in data-deprived regions
Reasons the SRF product was not useful:
Underestimated snowfall amount
Not available over water or coastline
Missed location of light/moderate snow detected by other sources
38
Results
More the 85% of responses indicated medium to high
confidence in merged SFR values, 100% of responses
indicated the product was useful
75% of forecasters indicated the ability to loop the product with
blended radar made the product more useful
Most responses indicated merged SFR was used to identify
snowfall in data-deprived regions and track the maxima
Reasons the merged SRF product was not useful:
Underestimated snowfall amount
Still too latent
Missed location of light/moderate/heavy snow detected by other
sources
39
Upcoming Assessment Winter 2017-18
Changes to product
Inclusion of SSMIS (DMSP: F16, F17, F18) and GMI (aboard NASA GPM)
Improved snowfall detection algorithm
Goal: Determine operational utility in the forecaster environment as it relates to:
Temporal resolution of data/imagery
Accuracy of snowfall detection and rates based on type of snowfall event
radar gaps
beam blockage/overshooting
tracking snowfall rate maxima (in combination with other satellite imagery)
Determine areas where cloud seeding may be occurring ahead of falling precipitation
Participating Offices
Albuquerque, NM
Anchorage, AK
Boulder, CO
Charleston, WV (limited participation)
Great Falls, MT (waiting to hear)
Juneau, AK (waiting to hear)
Fairbanks, AK (waiting to hear)
Sterling, VA (waiting to hear)
40
Summary
SPoRT and NESDIS have collaborated over the last 4
years to introduce the snowfall rate product to NWS
forecasters and assess the utility in the operational
environment
User feedback has let to product improvements including
a merged product, availability of liquid to snow ratio
displays, and inclusion of additional polar-orbiting data
Successful story of R2O and O2R with a period of
intensive interaction between product developers and
end-users.
41