AIRS Cloud Fraction Trends from a PDF-basedApproach Compared to PATMOS

AIRS Virtual Science Team Meeting

L. Larrabee Strow1,2, Andy Tangborn 2, and Howard Motteler2

May 12, 2019

1UMBC Physics Dept.

2UMBC JCET

Motivation

• Climate trending with hyperspectral IR will require

• Frequent reprocessing

• Simple algorithms

• Ability to switch instruments transparently

• Using less $

• UMBC Goal: Gridded T/H2O/O3 etc. Anomaly Products• Requires "transpose" of data (a big deal)

• Quick access to L1c/CHIRP for all times for each grid point• We were ready to do this, but our new RAID caught COVID-19 in lat Feb.

• So, we transposed a few channels instead

• Which, we were going to use to break up our gridded radiances in

"mostly clear", "less clear", and "very cloudy" for separate processing.

• PDFs were made for a 64 x 120 lat/lon grid (equal area) spanning 17years every 16-days (rough metric for cloud amount)

• OISST was matched to ocean obs

• (Hope to match ERA-I or (ERA5/MERRA2) to these in the future

• Here we use these PDFs to produce statistical measurements of Cloud

Fraction in 16-day increments

Cloud fraction is a simple parameter for comparisons between

observations and climate models

1

Validate versus Patmosx

Patmosx

• Heidinger A. et. al., The Pathfinder Atmospheres–Extended

AVHRR Climate Dataset, V 95 BAMS, 2014

• Starts in 1983

• AVHRR based, very high spatial resolution

• Many cloud products, not just cloud fraction

• Fairly heavily used, long-term NOAA support

Patmosx has been used for climate model studies, more below.

Validation versus AIRS L3 and/or MODIS could follow. . .

2

PDF-based Cloud Fraction using AIRS

• Need surface temperature: for now OISST (big lien, ocean only

for now)• Compute cloud forcing per observation for a window channel

• Forcing = BTobs - BTcalc clear. Using 1231 cm−1 channel.

• BTcalc clear not done yet. Approximate with 2-channel regression

that converts Tsurf to BTcalc clear for the 1231 cm−1 channel

(introduced by H. Aumann).

• Minimal accuracy needed for BTcalc,surf, but do require some

stability

• Use PDF bins = -140:2:70 K (Quite coarse for now.)

• Compute PDFs of cloud forcing for every 16-day time period

for 64 x 120 lat/lon grid

• This produces small files that are quickly processed• Cloud Fraction (CF)

• CF =∑α

-140K PDFcloud forcing

• α is a "clear threshold" to separate clear from cloudy

• Generally use -(5-10)K for α. 3

Sample Cloud Forcing PDF

Grid point in Atlantic Ocean south of northern Africa: at (-5,0)°

lat/lon, (1.8,3.0)°

4

Mean Cloud Fraction: AIRS versus Patmosx

Threshold = 5K Threshold = 10K

5

Cloud Fraction Variability

AIRS versus Patmosx Standard Deviation (over time)

6

Mean Statistical Differences

0 0.2 0.4 0.6 0.8 1

Patmos CF

0

0.2

0.4

0.6

0.8

1

AIR

S C

F

Cloud Fraction

• Correlation Coefficient: 0.98

• Mean (Patmos - AIRS): 0.37% ± 2.5% (std)

• Mean (Patmos - AIRS): ±60° lat= 0.03% ± 3%(std)

Cloud Trends

• Mean (Patmos - AIRS): -0.05 ±0.12 %/yr(std)

• Mean (Patmos - AIRS): ±60° lat = = -0.066±0.015 %/yr (std)

Estimated AIRS trend uncertainty due to SST trend errors: ~0.06%/yr 2σ 7

Cloud Fraction Trends (%/year)

• Many similarities, but note Equatorial Atlantic

• Note SAO issue for Patmosx

• Note scale is cloud fraction in %

8

Statistical Trend Uncertainties: 2σ values

9

AIRS Cloud Fraction Trends above Noise

• Trends in gray regions less than confidence interval

• Note 90% label on LHS should be 68%, not 90%

10

Do Trends Vary with Season?

More similar with season than not.

11

Previous Cloud Fraction Studies

Climate Model Comparisons

• Norris, J. et. al., Empirical Removal of Artifacts from the ISCCP and

PATMOS-x Satellite Cloud Records, Journal of Atmospheric and

Oceanic Technology, V 32, 2015

• Norris J. et. al., Evidence for climate change in the satellite cloud

record, Nature, Nautre, V. 536, 2016

• Trenberth, K. et. al., Global warming due to increasing absorbed

solar radiation, GRL, V 36, 2009

Issues

• Climate literature indicates that accurate cloud fraction observations

are difficults to find

• Norris used patterns of cloud fraction, not absolute trends!

Cloud Fraction Datasets

• ISSCP• PATMOSx pretty much succeeded ISSCP

• Heidinger A. et. al., The Pathfinder Atmospheres–Extended AVHRR

Climate Dataset, V 95 BAMS, 2014

• Starts in 1983, AVHRR

• MODIS/VIIRS: not examined, but there appear to be MODSIS to VIIRS

difficulties (maybe not with gridded like here??)

• AIRS Level 3 (need to examine!)

12

Comparison of AIRS vs Patmosx Means

• PATMOS transitions from NOAA Series to METOP1 during this time period.

Cloud Fraction Anomalies

2004 2006 2008 2010 2012 2014 2016 2018-2.5

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

2.5

Clo

ud F

rac A

nom

aly

(%

)

Patmosx

AIRS

Cloud Fraction Trends vs Latitude

-0.2 -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2

% Change in CF / Year

-80

-60

-40

-20

0

20

40

60

80

La

titu

de

AIRS

Patmos

• AIRS product very simple, more stable and insensitive to SAO

• Transition to CrIS/CHIRP should be simple (maybe none)13

Norris: 25-Year CF Trends (1983-2008)

4 A U G U S T 2 0 1 6 | V O L 5 3 6 | N A T U R E | 7 3

LETTER RESEARCH

albedo and the 1985–1989 ERBS albedo. Every observational record exhibits a decline in cloud amount or albedo at mid-latitudes in both hemispheres that is nearly always statistically significant. The ocean-only MAC-LWP dataset also reports less liquid water path around 40° N and 40° S (Extended Data Fig. 1b). Previous research found evi-dence for tropical expansion in recent decades19. Reduced cloudiness around 40° N and 40° S is consistent with a poleward expansion of

the subtropical dry zone cloud minimum and poleward retreat of the storm-track cloud maximum.

Figure 2d displays trends in zonal mean total cloud amount dur-ing the period 1983–2009 from the ALL simulations. Most individual simulations exhibit reduced cloud amount in the mid-latitudes of both hemispheres, and the ensemble mean trends are statistically significant (P < 0.05 two-sided). Furthermore, the majority of simulations repro-duce the observed increase in cloud amount and albedo occurring in the northern tropics. The spatial correlation between observed and simulated zonal cloud trends is highly significant (Table 1 and Extended Data Fig. 3).

Since the correction procedures applied to the satellite datasets removed any real global mean change that might be present, for maxi-mum comparability we subtracted the 60° S–60° N average change in total cloud amount from the model output before creating Figs 1c and d, and 2d. Without this adjustment, the ALL ensemble mean cloud amount averaged over 60° S–60° N decreases by 0.13% over 25 years. Although highly statistically significant (P < 0.0001 two-sided), the modelled reduction in 60° S–60° N average cloud amount during the period 1983–2009 is far smaller than what is detectable by our obser-vational systems. Extended Data Fig. 4a and b shows ALL cloud trends without the subtraction of the 60° S–60° N average change. They exhibit patterns similar to those seen in Figs 1c and 2d.

ISC

CP

–PA

TMO

S-x

cloud trend (percentage am

ount per 25-year period)

CER

ES–ER

BS

albedo change (percentage albedo

per 25-year period)

CM

IP5 A

LL cloud trend (percentage am

ount per 25-year period)

Majority of observations and CMIP5 simulations agree

3.0 1.5 0.5 0.0 –0.5 –1.5 –3.0

1.8 0.9 0.3 0.0 –0.3 –0.9 –1.8

0.6 0.3 0.1 0.0 –0.1 –0.3 –0.6

a

b

c

d

Positive

Negative

Figure 1 | Change in observed and simulated cloud amount and albedo between the 1980s and 2000s. a, Trend in average of PATMOS-x and ISCCP total cloud amount 1983–2009. b, Change in albedo from January 1985–December 1989 (ERBS) to July 2002–June 2014 (CERES). c, Trend in ensemble mean total cloud amount 1983–2009 from CMIP5 historical simulations with all radiative forcings (ALL). d, Locations where majority of observations and majority of simulations show increases (blue) or decreases (orange). Black dots indicate agreement among all three satellite records on sign of change in a and b and trend statistical significance (P < 0.05 two-sided) in c. All trends and changes are relative to the 60° S–60° N mean change.

Table 1 | Correlation between observed and modelled cloud trend patterns

Forcing type

Spatial pattern ALL GHG AA OZ NAT

Grid box total cloud amount

0.39 (0.0001) [0.003]

0.21 (0.05) [0.08]

0.00 0.00 0.26 (0.03) [0.04]

Zonal mean total cloud amount

0.80 (0.002) [0.009]

0.62 (0.008) [0.06]

−0.35 0.27 0.69 (0.03) [0.03]

Zonal mean cloud amount in the 50–180 hPa and 180–320 hPa intervals

0.76 (0.003) [0.03]

0.73 (0.004) [0.04]

−0.62 0.73 (0.003) [0.04]

Parentheses and square brackets indicate one-sided P values obtained from the preindustrial simulations shown in Extended Data Fig. 3 and from formal significance tests, respectively.

ISC

CP

clo

ud tr

end

(per

cent

age

amou

nt

per 2

5-ye

ar p

erio

d) 2.0 100

1.0 85

0.0 70

–1.0 55

–2.0 40

PA

TMO

S-x

clo

ud tr

end

(per

cent

age

amou

nt

per 2

5-ye

ar p

erio

d) 2.0 100

1.0 85

0.0 70

–1.0 55

–2.0 40

CER

ES –

ER

BS

al

bedo

cha

nge

(per

cent

age

albe

dope

r 25-

year

per

iod) 1.2 50

0.6 40

30 0.0

–0.6 20

–1.2 10

CM

IP5

ALL

clo

ud tr

end

(per

cent

age

amou

nt

per 2

5-ye

ar p

erio

d)

Clim

atol

ogy

(per

cent

age

amou

nt)

Clim

atol

ogy

(per

cent

age

amou

nt)

Clim

atol

ogy

(per

cent

age

amou

nt)

Clim

atol

ogy

(per

cent

age

albe

do)

0.8 100

0.4 85

0.0 70

–0.4 55

–0.8 40 60° S 40° S 20° S Eq 20° N 40° N 60° N

a

b

c

d

Figure 2 | Zonal mean change in observed and simulated cloud amount and albedo between the 1980s and the 2000s. a, Trend in ISCCP total cloud amount 1983–2009. b, Trend in PATMOS-x total cloud amount 1983–2009. c, Change in albedo from January 1985–December 1989 (ERBS) to July 2002–June 2014 (CERES). d, Trend in ensemble mean total cloud amount 1983–2009 from CMIP5 historical simulations with all radiative forcings (ALL). Zonal mean climatology is dotted, linear trend or change is solid, circles indicate trend statistical significance (P < 0.05 two-sided), and bars indicate the interquartile range of individual simulations. All trends and changes are relative to the 60° S–60° N mean change.

© 2016 Macmillan Publishers Limited, part of Springer Nature. All rights reserved.

• Norris only showing trend versus ±60° mean value!

• Don’t expect 1983 - 2008 trends toagree with AIRS, but "reasonable"

-60 -40 -20 Eq 20 40 60

Latitude

-4

-2

0

2

% O

ve

r 2

5-Y

ea

rs

AIRS

Patmosx

• AIRS and Patmosx trends in theabove graph are 17 years(2002-2019)

• AIRS trends appear more realistic

• High Patmosx value near -20 islikely Southern Atlantic Anomalyproblems

14

Summary

• Extremely simple cloud fraction algorithm competitive or

better than Patmosx (one channel)

• Magnitudes as expected, in ballpark of 30-year

• Amenable to rigorous error analysis

• Simple to compute allowing easy reprocessing• Many ways to extend:

• Roughly assign to high/middle/low clouds although mixed

scenes will confuse this metric

• Use more optically thick channels for middle versus high cloud

differentiation: clear RTA calcs are very accurate

• Use alternative data for cloud classification

• Use multiple window channels for thin cirrus classification

• Very difficult to produce if radiances are not binned by grid

point (transpose)

• Next: add land and use BTcalc intead of Tsurf with regression

for forcing15