ACIX-III Land

At m osp h e r i c Cor re c t i on
In t e r- c om p ar i son e X e rc i se
2nd Workshop on International Cooperation in
Spaceborne Imaging Spectroscopy
19 – 21 October 2022, Frascati, Italy

Georgia Doxani I Serco for ESA-ESRIN

Ferran Gascon I ESA-ESRIN
Philip Townsend I JPL, University of Wisconsin
David Thompson I JPL, California Institute of Technology
Philip Brodrick I JPL, California Institute of Technology
Adam Chlus I JPL, California Institute of Technology
WHY?
Free and open access policy to Earth Observation
imagery has stimulated the development and operational
OPEN use of AC processors for generating Bottom-of-
ACCESS
Atmosphere (BOA) products

The objective is to point out:

S W
• Strengths & Weaknesses
C D • Commonalities & Differences

2
HOW?

Definition of the
Application of the AC Analysis of the results
inter-comparison
processors
protocol

Coordinators & Participants Coordinators

Participants applied their AC schemes processed the AC results
discussed all the major on a set of test sites keeping and assessed the inter-
the processing parameters comparison metrics. The
points and defined the inter-
results presented and
comparison procedure. constant. The results were discussed with the
submitted for analysis to participants.
ACIX coordinators.

3
 WHEN?
CMIX ACIX-II Land ACIX-II Aqua
10 processors 12 processors 8 processors
2016 5 validation datasets 120 AERONET sites 20 AERONET OC sites 2022
- Publication link Publication (soon) Publication link -
2018 Website link Website link Website link TBD

ACIX
2018
11 processors -
2022 CMIX-II ACIX-III Land ACIX-III Aqua
19 AERONET sites
13 processors 13 processors 14 processors
Publication link
Website link
4
 With the support of:

HOW?

ACIX-III CMIX-II

Atmospheric Correction
Inter-comparison
Cloud Masking
Inter-comparison
Processors over Processors over
LAND sites AQUATIC sites

Sentinel-2, Landsat-8 Sentinel-2, Landsat-8

PRISMA & PRISMA & PRISMA (TBD)

5
 With the support of:

HOW?
ACIX-III

Atmospheric Correction
Inter-comparison

Processors over
LAND sites
Jet Propulsion Laboratory
California Institute of Technology

Ferran Gascon Georgia Doxani

Phil Townsend David Thompson Philip Brodrick Adam Chlus
6
WHO? # Name Affiliation Processor’s Name

1 Raquel de los Reyes DLR PACO

2 Feng Yin University College London SIAC

3 Stefan Adriaensen VITO iCOR
4 Béatrice Berthelot MAGELLIUM MAGAC
5 QUAC
Tim Perkins Spectral Sciences, Inc.
6 FLAASH
7 Ian Brosnan NASA Ames Research Center HECC
8 Philip Brodrick JPL isofit

9 Weile Wang NASA Ames Research Center GeoNEX-AC

10 Laurent Poutier ONERA COCHISE

11 Yaokai Liu CAS.CHINA Hikerliu

12 Quinten Vanhellemont RBINS ACOLITE/DSF

Angelo Palombo
CNR Institute of Methodologies for
13 ImaACor
Environmental Analysis (IMAA)
Federico Santini

7
 With the support of:

HOW?
START
END

ACIX coordinators
distribute to the
participants the
image data
ACIX coordinators
acquired over the
and participants test sites
prepare a synthesis of ACIX participants
all the experimental access and
results to be published download the data
in the scientific
literature

ACIX coordinators ACIX participants

analyse the results

ACIX run their atmospheric
and report the web and FTP correction processors
themselves, keeping
inter-comparison interface the processing
outputs parameters constant

ACIX participants
ACIX coordinators
submit their results
retrieve the and information
submitted results about the
and reports implementation

END Of Test and

Submission Phase
8
Input Data [ASI]

With the support of:

PRISMA
All PRISMA Products are in HDF5-EOS format and include HYP data cube + PAN
image + metadata

PRISMA L1 data: Top-of-Atmosphere Radiance radiometrically corrected and calibrated in

physical units (incl. Cloud mask; Sun-glint Mask; Classification Mask; Calibration and
characterization data)

• Absolute HYP radiometric accuracy < 5% (TOA or BOA)

• SNR 160:1 in VNIR and 100:1 in SWIR (240:1 in PAN)
• MTF (@Nyquist) 0.3 for HYP and 0.2 for PAN
• Geometric localization errors (CE90) < 200m (15m with GCPs, available starting in Q1/2023)
• Smile effect is well below 0.1 pix (1 nm) for current products, following the product
requirements. The results are better for SWIR than the VNIR cube.

9
Input Data [Users’ feedback: Geometry]

With the support of:

PRISMA

Users’ feedback report geometric errors > 200m depending on the viewing
on the scene topography and viewing angles.

Action in ACIX

JPL team will implement their imaging matching code to improve the
coordinate accuracies using Landsat scenes as a reference.

10
Input Data [Users’ feedback: Smile distortion]

With the support of:

PRISMA

Users indicated that there is an important smile effect and calibration error of
up to 3 nm in the VNIR

Action in ACIX

A smile correction approach is being investigated and discussed together with

the participants

11
Input Data [Users’ feedback: Metadata]

With the support of:

PRISMA

Ancillary information, i.e., View Zenith Angle, Solar Zenith Angle, etc., is
missing from PRISMA L1 data

Action in ACIX
L1 and L2C products will be downloaded and provided to participants

12
Quality Layers

With the support of:

Input
Quality flags provided as part of the products defining clearly the
appropriate flags to be included in the inter-comparison process.

Ideally, a simple mask to be provided defining only valid and invalid

pixels for the AC inter-comparison.

Analysis
The analysis will be made initially for the pixels that are considered of
good quality by all AC processors (masks’ union).

(?) Additional inter-comparison investigation may be performed using the

corresponding individual quality mask per processor. 13
Quality Layers

With the support of:

The quality flags can be different for AOD, WV and BOA.

Different layers can be provided and they will be combined correspondingly.

Each participant should indicate the flags to be involved in the analysis, eg. no
clouds, no cloud shadows, no snow, no water for WV, no high aerosol for BOA, etc.

The coordinators will not question the flags, not any quality flags validation

14
HOW?

Aerosol
Surface
Optical Water
Reflectance
Depth Vapour

15
Aerosol
Optical
Depth

Aerosol
Water
Optical
Vapour
Depth

Statistics
AERONET No. of samples
R2 (Coefficient of determination)
RMSE
bias
Estimated AOD (/WV) & compared to Level 1.5
(cloud screened) AERONET data

1. Interpolate AERONET values @ λ=550 nm using

Angstrom Exponent APU analysis
2. Average AERONET values over time period within !
Accuracy (A): ! = (∑ #$%! %!"# ) (1)
±15 min from AOD retrieved values (PRISMA "

overpass)
! (2)
Precision (P): ' = ) ∑ # (% − !))
("'!) *%! !"#
3. Average estimated AOD values over an image
subset of 9 km x 9 km centred on the AERONET (3)
!
Sunphotometer station Uncertainty (U): + = ) ∑#$%! %!"# ))
"

16
AERONET sites with PRISMA acquisitions

129 AERONET sites are covered by

PRISMA

Sites mainly in Europe and North America

AERONET sites with valid measurements

to be investigated

17
HOW?

Aerosol
Surface
Optical Water
Reflectance
Depth Vapour

18
 HOW?
02. Campaign Data
Surface
01. Ground based validation Reference BOA retrieved by
AVIRIS NG for ESA CHIME & SBG
Reflectance RadCalNet [La Crau] 2021 campaign and NEON Airborne
(France), Gobabeb (Namibia)], Observation Platform (AOP)
SR will be provided by CNES in
the same angular conditions as
Surface
Surface
Reflectance
Reflectance
PRISMA

05. SR inter-comparison
Plotting the SR time
03. AERONET corrected data (TBD)
series per date, band and AC data generated by 6SV
AC approach. radiative transfer model using
AERONET data. AOT, aerosol
model and column water vapour
will be derived from AERONET
sunphotometer measurements and
will be used in the radiative
transfer model in order to perform
the AC of TOA reflectance.

19
01. Ground based validation RadCalNet

Gobabeb [Namibia]: 37 PRISMA scenes [CC<10%] La Crau [France ]: 12 PRISMA scenes [CC<10%]

The measurements will be processed to PRISMA observation geometry and

reflectance spectrum by CNES

The area of interest will be proposed by the site PIs (CNES) regarding the
homogeneity of the sites for the study period to be selected

The study period is proposed to be the full PRISMA archive over these sites
20
02. Campaign Data

AVIRIS NG for ESA CHIME & SBG 2021

Country Site Latitude Longitude PRISMA date AVIRIS-NG date(s) PRISMA scene

Germany Demmin 53.7723 13.089 20210518 20210530 20210518102102_20210518102107_0001

Spain Camarena 39.9703 -4.1459 20210630 20210630 20210630110522_20210630110526_0001

Braccagni 42.8374 11.0709 20210604 20210604 20210604101721_20210604101725_0001

Italy Rio Tinto 37.7791 -6.5747 20210625 20210625 20210625111917_20210625111921_0001

Jolanda 44.8905 11.957 20210621 20210625 20210621101020_20210621101024_0001

Great Britain FlowCountry 58.3652 -3.9573 20210603 20210615 20210603113329_20210603113333_0001

NEON Airborne Observation Platform (AOP)

Site Latitude Longitude PRISMA date NEON AOP date(s) PRISMA scene
MOAB 38.248283 -109.38827 20200702 20200705 20200702181741_20200702181745_0001
WREF 45.82049 -121.95191 20210729 20210718 20210729190927_20210729190932_0001
SRER 31.91068 -110.83549 20210829 20210823-20210902 20210829180958_20210829181002_0001
NIWO 40.05425 -105.58237 20200822 20200801 - 20200807 20200822175652_20200822175657_0001

22
03. AERONET corrected data (TBD)

Eric Vermote, 2016. MODIS/VIIRS Surface Reflectance

26
03. AERONET corrected data (TBD)

Example from ACIX-II

!
Accuracy (A): ! = ∑"#$!
! ()
$%%,'
"!

! + () /
Precision (P): & = ∑.$!
"
$%%,' −!
(+" ,!)

! " () /
Uncertainty (U): ( = ∑#$!
!
($%%,' )
"!

27
03. AERONET corrected data (TBD)

28
Lessons Learnt

Respect the protocol and follow the filenames, file formats, etc. as agreed
More time to read data and investigate the mismatching with the protocol than the
validation analysis itself

Respect the deadlines

Requests for including results after the result submission deadline is not fair and
respectful to the rest of the group

29
Timeline

Report Release
to the participants

Results Submission
Start
Invitation to 1st workshop
developers ESA/ESRIN
(Frascati, Italy)

20-21
Dec 2021 June TBD TBD TBD TBD TBD
2022

Input Data Inter-comparison Report

Distribution Results Submission to
Scientific Journal
to Participants
Presentation
2nd workshop
30
Timeline

Report Release
to the participants

Results Submission
Start
Invitation to 1st workshop
developers ESA/ESRIN
(Frascati, Italy)

20-21
June Oct TBD TBD
Dec 2021 TBD TBD
2022 2022

Data Downloading Inter-comparison Report

and Results Submission to
Scientific Journal
Preprocessing Presentation
Approaches 2nd workshop
31
Thank you for your attention!

1st WS of ACIX-III Land, -Aqua and CMIX-II:

https://earth.esa.int/eogateway/events/1st-workshop-of-acix-iii-land-aqua-and-cmix-ii/agenda

32