11

ABI Ice Thickness and Age Algorithm (AITA)

June 15, 2011

Presented by: Xuanji Wang1

Other team members: Jeff Key2, Yinghui Liu1

1UW-Madison/CIMSS2NOAA/NESDIS/STAR

2011 GOES-R AWG Annual Meeting, June 13-16, Fort Collins, CO

GOES-R AWG Cryosphere Team

• Executive Summary

• Algorithm Description• Algorithm changes from 80% to 100%

• Example AITA output

• ADEB and IV&V Responses Summary

• Requirements Specification Evolution

• Validation Strategy

• Validation Results

• Summary

Outline

2

3

The GOES-R Sea and Lake Ice Thickness and Age is an Option 2 The GOES-R Sea and Lake Ice Thickness and Age is an Option 2 product. product.

Software Version 5 was delivered in May 2011. ATBD (100%) is on Software Version 5 was delivered in May 2011. ATBD (100%) is on track for a August delivery.track for a August delivery.

The algorithm has been developed, tested, and validated using The algorithm has been developed, tested, and validated using AVHRR, MODIS, SEVIRI data and in situ observed data from AVHRR, MODIS, SEVIRI data and in situ observed data from submarine, mooring sites, and stations as well as numerical model submarine, mooring sites, and stations as well as numerical model simulations.simulations.

Validation studies indicate spec compliance for the product.Validation studies indicate spec compliance for the product.

Executive Summary

3

44

Algorithm DescriptionOTIM – Physics and Applicability

One-dimensional Thermodynamic Ice Model (OTIM)

(1-αs)(1-i0)Fr – Flup + Fl

dn + Fs + Fe + Fc = Fa(αs, Ts, U, hi, C, hs, …)

Snow layer

Ice layerhs

hi

Ts

T0

Tf

ZFcs = Fci

Fr(Sz)αsFr

i0(1-αs)Fr

(1-αs)(1-i0)Fr

Flup Fl

dn Fs Fe Fc Fa

Based on the surface energy budget at thermo-equilibrium state, the fundamental equation is

After parameterizations of thermal radiation (Fr, Flup, Fl

dn) and turbulent (sensible &

latent) heat (Fs, Fe), ice thickness hi becomes a function of 11 model controlling variables:

hi = f(αs, i0, Sz, Ts, Ti, Ta, Pa, hw, U, C, hs, Fa)

Ta

Ti

U Chw Pa

Cloud

Algorithm Description OTIM – Structure and Components

5

66

Algorithm DescriptionOTIM - High Level Flowchart

OTIM and Ice age classification begin

Surface air temperature, pressure, wind, and moisture

Cloud mask/amount, ice and snow physical properties, snow

cover/depth, sea/lake mask

For each pixel

Surface albedo

Surface albedo

Surface Temperature

Surface Temperature

Surface radiative fluxes

Surface radiative fluxes

Surface heat fluxes

Surface heat fluxes

One-dimensional Thermodynamic Ice Model - OTIM

All-sky sea/lake ice thickness

Sea/lake ice age classification

All-sky sea/lake ice age

OTIM and Ice age Classification end

ABI Channel Used: Radiances and retrieved products, e.g., cloud mask.

Ancillary Data:NWP/NCEP/ECMWF, etc.

Products Generated: Ice thickness, ice age, and their associated quality flags.

ABI Channel Used: Radiances and retrieved products, e.g., cloud mask.

Ancillary Data:NWP/NCEP/ECMWF, etc.

Products Generated: Ice thickness, ice age, and their associated quality flags.

Ice Age Categories:

Ice free: Directly from ice identification/concentration algorithm when ice concentration is less than 15%.

First-year ice: Ice thickness < 1.80m. First-year ice includes New Ice (<5cm), Nilas Ice (5~10cm), Grey Ice (10~15cm), Grey-white Ice (15~30cm), Thin First-year Ice (30~70cm), Medium First-year Ice (70~120cm), Thick First-year Ice (120~180cm).

Older ice: Ice thickness >= 1.80m.

Algorithm DescriptionOTIM - Ice Age Classification

7

88

Algorithm Changes from 80% to 100%

A new lookup table for estimating residual heat flux is added in the version 5 code.

Two independent subroutines have been added to the version 5 code for the model variable initialization and metadata analysis in order to avoid machine type issue.

Metadata output added. Quality flag added.

99

NASA MODIS Data: Great Lakes

Ice Thickness (m) over Great Lakes area, February 24, 2008.

Ice Age derived from Ice Thickness over Great Lakes area, February 24, 2008.

Ice Age Classification:

1: Free of ice (white)

2: New ice

3: Grey ice

4: Grey-white ice

5: Thin first-year ice

6: Median first-year ice

7: Thick first-year ice

8: Old ice

Clear sky condition

Example AITA Output

9

1010

MSG SEVIRI Data: Caspian Sea

Ice Thickness (m) over Caspian Sea, February 24, 2008.

Ice Age derived from Ice Thickness over Caspian Sea, February 24, 2008.

Ice Age Classification:

1: Free of ice (white)

2: New ice

3: Grey ice

4: Grey-white ice

5: Thin first-year ice

6: Median first-year ice

7: Thick first-year ice

8: Old ice

Clear sky condition 10

Example AITA Output

11

ADEB and IV&V Response Summary

All comments on ATBD clarification have been addressed. Suggestions on algorithm improvement have been considered

and implemented. Suggestions on validation work are taken and more extensive

validation work is undergoing.

Critical Recommendations and Responses Recommendations: Continue to pursue more complete data sets. Measurement validation

must be conducted with thoroughness and completeness within a sustained validation/verification framework and should consider using “human in the loop.” This recommendation was made previously but has rarely been adopted.

Response: Ice age/thickness “deep-dive” validation uses all types of available data: in situ measurements, submarine and moored sonar, numerical model simulations, and estimates from other sensors such as ICESat. We believe that there is little more that can be done for this products. We also have plans to use data from current and upcoming field campaigns, such as NASA’s IceBridge flights.

12

ADEB and IV&V Response Summary

Critical Recommendations and Responses Recommendations: Recommend teams continue work on “stretch goals” that approach the state-of-the-art.

Response: The current requirements are based on the original Threshold values rather than the Goal. Nevertheless, we have tightened the Threshold requirements somewhat, and continue to strive for meeting the Goal requirements. For example, our Sea and Lake Ice Age product far exceeds the Threshold requirement of two ice classes by providing an ice thickness estimate. The thickness estimate can be used to classify ice into the Goal categories, even though it is not a current requirement to do so.

Recommendations: plans to reach 100% should be clearly stated.

Response: Tasks for moving ice thickness/age products from 80% to 100% maturity include: improving the daytime ice age algorithm with an approach that parameterizes the residual heat fluxes, improving the parameterized relationship between ice thickness and ice age, performing additional deep-dive validation, adding metadata to the products, and updating the ATBDs.

Recommendations: OK for advance to 100%. The board also recommends further validation including in-situ sonar and RadarSat. Team should consider future implementation that uses GOES-R produced fluxes vs. independently calculating those fluxes within this algorithm.

Response: Additional sonar data has been acquired and used for validation. RADARSAT data are being investigated. If GOES-R flux products are of sufficiently high quality, and we expect that they will be, they will be used in the ice age algorithm.

Nam

e

User &

Priority

Geograp

hic C

overage

(G, H

, C, M

)

Vertical R

esolution

Horizon

tal Resolu

tion

Map

pin

g Accu

racy

Measu

remen

t Ran

ge

Measu

remen

t Accu

racy

Prod

uct R

efresh

Rate/C

overage Tim

e

Ven

dor A

llocated G

roun

d L

atency

Prod

uct M

easurem

ent P

recision

Sea

& Lake Ice: A

ge

GO

ES

-R

Full D

isk (FD

)

Ice S

urface

1 km3 km

Ice free are

as,

First year ice

Olde

r Ice

80

% co

rrect cla

ssificatio

n (c

ha

ng

ed

from

orig

ina

l 85

% c

orre

ctio

n c

las

sific

atio

n)

6 hr

3236 sec

1 category Requirements Specification Evolution

13

Nam

e

User &

Priority

Geograp

hic C

overage

(G, H

, C, M

)

Tem

poral C

overage Qu

alifiers

Prod

uct E

xtent Q

ualifier

Clou

d C

over Con

dition

s Q

ualifier

Prod

uct S

tatistics Qu

alifier

Sea

& Lake Ice: A

ge

GO

ES

-R

Full D

isk (FD

)

Sun

at less than 67 de

gree, solar zenith

angle

Qua

ntitative out to at least 67 de

grees LZA

and q

ualitative at large

r LZ

A

Clear conditions asso

ciated

with thresh

old accuracy

Over specified geogra

phic areaRequirements Specification Evolution

14

1515

Validation ‘Truth’ Data Station ice thickness measurements since 2002 from Canadian Ice Service (CIS).

Submarine cruise measurements of ice draft data using Upward Looking Sonar (ULS) from National Snow and Ice Data Center (NSIDC) and Unified Sea Ice Thickness climate Data Record (USITCDR) at UW-Seattle.

Moored Upward Looking Sonar (ULS) measured ice draft/thickness data from NSIDC and Beaufort Gyre Exploration Program (BGEP).

Numerical model simulated ice thickness from a Pan-Arctic Ice-Ocean Modeling and Assimilation System (PIOMAS) at UW-Seattle.

Ice thickness from past, current, and future field campaigns from RADARSAT-1, ICESat, and IceBridge.

Ice Age derived from SMMR and SSM/I passive microwave data with NASA

team algorithm since 1978.

Validation Strategy

1616

Validation Test Data

» Advanced Very High Resolution Radiometer (AVHRR) Polar Pathfinder (APP) Extended (APP-x) Data over the Arctic and Great Lakes.

» MODIS data over the Arctic and the Great Lakes.

» Meteosat SEVIRI data over the Caspian Sea.

Validation Strategy

1717

Validation Method» Direct match-up and comparison in ice thickness between

OTIM retrievals and ‘truth’ data.

» Compare Ice Age derived from Ice Thickness (this algorithm) with independent Ice Age estimation from Nimbus-7 SMMR and DMSP SSM/I Passive Microwave Data (NASA team algorithm)

Validation Performance Metrics» Cumulative frequency, bias mean, and absolute bias mean,

accuracy, precision.

Validation Strategy

1818

Comparison of AVHRR Ice Thickness with submarine ULS measurements and numerical model simulations

Submarine trajectory, 1999

Ice thickness cumulative distribution retrieved by OTIM with APP-x data, submarine sonar data, and simulated thickness from the PIOMAS model. Submarine ice draft (mean and median only) was converted to ice thickness by a factor of 1.11.

Ice thickness values retrieved by OTIM with APP-x data, submarine sonar data, and simulated thickness from the PIOMAS model along the submarine track segments. Submarine ice draft (mean and median only) was converted to ice thickness by a factor of 1.11.

Validation Results

19

Comparison of AVHRR Ice Thickness with station measurements and numerical model simulations

Comparisons of ice thickness cumulative distribution retrieved by OTIM with APP-x data, simulated thickness from the PIOMAS model and station measurements at Alert.

Comparisons of ice thickness values retrieved by OTIM with APP-x data, measurements at Alert, and simulated thickness from the PIOMAS model.

Station location map. Totally 11 Canadian stations participate the New Arctic Program starting from 2002 for ice

thickness measurements.

Validation Results

19

20

Comparison of AVHRR Ice Thickness with Moored ULS measurements and numerical model simulations

Comparisons of ice thickness cumulative distribution retrieved by OTIM with APP-x data, simulated thickness from the PIOMAS model, and ULS measurements at the mooring site A.

Comparisons of ice thickness values retrieved by OTIM with APP-x data, simulated thickness from the PIOMAS model and ULS measurements at the mooring site A.

Mooring Location

Site A (75o0.499’N, 149o58.660’W)

Site B (78o1.490’N, 149o49.203’W)

Site C (76o59.232’N, 139o54.562’W)

The location information of the three mooring sites during the Beaufort Gyre Exploration Project starting from 2003.

Validation Results

20

21

Comparison of OTIM derived Ice Thickness with Submarine and Moored ULS measurements, and station measurements

In-situ MeasurementsOTIM

Thickness mean (m)

Bias mean (m)

Accuracy(%)

Submarine 1.80-0.07 96%

OTIM 1.73

Mooring Sites 1.29-0.09 93%

OTIM 1.20

Stations 1.31-0.11 91%

OTIM 1.20

All 1.47-0.09 94%

OTIM 1.38

Validation Results

21

2222

0:Ice free

1: First-year ice

2: Older ice

OTIM Ice Age from MODIS data vs Microwave (NASA) Ice Age

SMMR and SSM/I vs AITA

(2006)

MODIS TERRA & AQUA

(2412 swaths, Day & Night, March, 2006)

Validation Results

2323

OTIM Ice Age with MODIS data vs Microwave (NASA) Ice Age

Ice Age

(OTIM vs Microwave)

Statistics Accuracy

Ice Free D&N:93%, N:93%, D:~100%

First-year Ice D&N:92%, N:92%, D:~100%

Older Ice D&N:84%, N:84%, D:~100%

All D&N:89%, N:89%, D:~100%

Error Sources

1. Ice identification algorithm

2. Cloud mask/shadow detection

3. Relationship between thickness and age

4. Ice motion/Dynamic processes

D=day; N=night

Validation Results

2424

OTIM Ice Age with MODIS data vs microwave (NASA) Ice Age

Note: Number in each cell stands for the number of pixels that belong to the ice age category difference corresponding to NASA and OTIM ice age classifications used to do statistics, i.e., accuracy and precision in ice age classification.

Ice Age Difference(OTIM vs Microwave)

No Difference

1 Category Difference

2 Category Difference

(D&N:49820)

(N:49822)

(D:34774)

(D&N:7154)

(N:7146)

(D:30)

(D&N:52)

(N:52)

(D:0)

Precision

(D&N:0.34 Category)

(N:0.34 Category)

(D:0.03 Category)

D=day; N=night

Validation Results

24

25

The ABI Ice Thickness and Age Algorithm provides a unique solution that utilizes the new capabilities offered by the ABI and its derived products.

» ABI allows us to monitor ice conditions at high temporal and spatial resolution.

» The Sea & Lake Ice Age Product has been run offline and within the framework and the results are exactly the same.

This product meets the specifications and is proving useful to downstream applications.

» The Sea & Lake Ice Age Product uses MODIS and AVHRR data as proxy and is validated against Upward Looking Sonar (ULS) measurements from submarine and mooring sites, stations observations, and passive microwave data to demonstrate that the Ice Age algorithm meets product requirements of 80% correct accuracy and less than one category precision.

Summary

25

26

Version 5 100% code is delivered and the 100% ATBD is coming soon. » This algorithm will be further modified to improve the accuracy. Additional

validation data will be employed once they become available.

Summary

26