QWG7 iCOR atmospheric correction evaluation QWG7... · remotesensing.vito.be iCOR atmospheric correction MODTRAN-5 based LUTs AOT retrieval based on Guanter et al. (2008) Superpixels

QWG7 – iCOR atmospheric correction evaluation

E. Wolters, S. Sterckx, S. Adriaensen

03/05/2018 VITO Remote sensing

remotesensing.vito.be

Contents

• iCOR atmospheric correction

• iCOR processing PROBA-V L2A

• Evaluation setup

• Results

• Summary & recommendation


iCOR atmospheric correction MODTRAN-5 based LUTs

AOT retrieval based on Guanter et al. (2008)

Superpixels of 30 x 30 km2 atmosphere invariant within area

Dark target retrieval (max. AOT boundary), refinement by end-member inversion

Implementation for PROBA-V:

AOT LUTs computed for all cameras

altitude dependent LUTs

Rural aerosol model

Application to Level 2A segments (TOA reflectance)


iCOR processing

• 18 AERONET stations selected

• Time period 2014 – 2015

• iCOR LUTs for all cameras

• ~3,500 1 km segments processed

• Workflow adapted for MEP processing

• Parallel processing, ~300 executors

• Run time < 4 h


iCOR data processing

• No iCOR AOT fall-back implemented yet

• Outside of scope current research

• Spectrally homogeneous areas: AOT retrieval extrapolated from distant macro-pixels


AERONET stations

AERONET station Coordinates [olon, olat] Altitude [m] Site information Category

West Africa Banizoumbou (Niger) 2.665 E, 13.541 N 250 Cultivated sandy area. Bare soil

IER Cinzana (Mali) 5.934 W, 13.278 N 285 Cultivated area. Bare soil

Zinder Airport (Niger) 8.990 E, 13.777 N 456 Semi-arid cultivated area (from June –

October) just south of Zinder. Bare soil

East Asia Beijing (China) 116.381 E, 39.977 N 92 Urban area; instruments located at

research institute’s rooftop. Urban

Western Europe

Aubière LAMP (France) 3.111 E, 45.761 N 423 Located near Clermond Ferrand, urban

area. Urban

Bure OPE (France) 5.505 E, 48.562 N 393 Open grassland area near

Hourdelaincourt. Vegetated

Carpentras (France) 5.058 E, 44.083 N 100 Cultivated area about 5 km north of Carpentras.

Vegetated

Mainz (Germany) 8.300 E, 49.999 N 150 Mixture of urban and rural area, within 1

km from the Main River, moderate to

highly polluted region.

Urban

Murcia (Spain) 1.171 W, 38.001 N 69 Located in rural and dry area, few km

west of the city. Vegetated

Toulouse Météo France (France) 1.374 E, 43.575 N 150 Located in rural/urban area southwest of Toulouse.

Urban

Davos (Switzerland) 9.844 E, 46.813 N 1596 Located at the northern edge of Davos,

close to Davosersee. Clean air. Vegetated

Hohenpeißenberg DWD (Germany) 11.012 E, 47.802 N 990 Located at a mountain top just outside

the village. Vegetated

South America

Alta Floresta (Brazil) 56.104 W, 9.871 S 277 Located near an airport, west of Alta

Floresta. Vegetated

Ji Parana SE (Brazil) 61.852 W, 10.934 S 218 In pasture, surrounded by trees. Vegetated

La Paz (Bolivia) 68.066 W, 16.539 S 3439 Located at the edge of La Paz, air

contains mixture of urban and biomass

burning.

Bare soil

North America

Bonanza Creek (AK, USA) 148.316 W, 64.743 N 150 Field site surrounded by conifer trees. Vegetated

Goldstone (NV, USA) 116.972 W, 35.233 N 1100 Location in Mojave desert, within Deep

Space Network Telescope area. Bare soil

Monterey (CA, USA) 121.855 W, 36.593 N 50 Located near Monterey Airport, ~2 km

from a bay. Urban

vegetated

urban

Bare soil


Performance metrics • Accuracy (A):

𝐴 =1

𝑛 𝜀𝑖

𝑛

𝑖=1

• Precision (P):

𝑃 =1

𝑛 − 1 (𝜀𝑖 − 𝐴)

2

𝑛

𝑖=1

• Uncertainty (U):

𝑈 =1

𝑛 𝜀𝑖

2

𝑛

𝑖=1


iCOR AOT evaluation • AERONET Level 2.0 data

𝐴𝑂𝑇0.55 = 𝐴𝑂𝑇0.44𝜆

𝜆𝑜

−𝛼

• AERONET median over +/- 30 min from overpass

• Cloud cover < 60% over 50 × 50 pixels around station

• iCOR median over 9 × 9 pixels


iCOR TOC reflectance evaluation • Rayference reference dataset (single-pixel)

• iCOR single pixel values

• Cloud cover < 60% over 50 × 50 pixels

• Current operational (‘OP’) TOC from S1 TOC data

• Temporal collocation using S1 TIME grid


Results: iCOR AOT


Results: current AOT retrieval AOT retrieval AOT latitudinal fall-back


Results: iCOR TOC BLUE

iCOR

OP S1


Results: iCOR TOC RED

iCOR

OP S1


Results: iCOR TOC NIR

iCOR

OP S1


Results: iCOR TOC SWIR

iCOR

OP S1


Results: time series examples Aubière (France, urban) Cinzana (Mali, BS)


Results: time series examples Monterey (USA, urban) Carpentras (France, veg)


Summary AOT

• iCOR compares well against AERONET over vegetated and urban areas

• Low agreement over bare soils lack of spectral variability, fall-back necessary

• Results conform ACIX evaluation (LS 8 & S-2A)


Summary TOC reflectance

• Performance differences iCOR vs OP S1 TOC small

• Slight improvement for BLUE, RED over vegetation

• iCOR performance slightly lower than OP S1 for NIR and SWIR

• Performance iCOR AOT iCOR TOC, aerosol models MODTRAN?


iCOR vs SMAC aerosol models

Aerosol type relative contribution

Water soluble Dust-like Soot ωo @ 0.55 µm [-]

iCOR (MODTRAN5) 70% 30% 0% 0.943

SMAC (6S) 29% 70% 1% 0.893

• Different contributions of water soluble and dust-like aerosols • Small soot contribution in 6S more absorption


Options for C2 AC baseline 1. Maintain SMAC with BLUE/SWIR AOT retrieval

Provide additional SMAC accuracy information to users (e.g. good, acceptable, unreliable, bad)

Replace latitudinal fall-back with CAMS NRT/climatology

Effort 30 – 50 WD

2. Implement iCOR (given the good AOT retrievals) Additional research for NIR and SWIR TOC required

Impact of different aerosol microphysical properties 6S and MODTRAN?

Larger effort: 70 – 80 WD


SMAC: accuracy vs SZA+VZA, WV Proud et al. (2010), SMAC on SEVIRI RED – SWIR, Africa:

• SMAC accuracy decreases for a.o. joint SZA+VZA > ~100o, WV > 2.0 g cm-2

Joint SZA+VZA SZA

AOT WV

THANK YOU


Sentinel-2 image Copernicus Sentinel data (2016)

QWG7 iCOR atmospheric correction evaluation QWG7... · remotesensing.vito.be iCOR atmospheric correction MODTRAN-5 based LUTs AOT retrieval based on Guanter et al. (2008) Superpixels

Documents