end-to-end simulation of sentinel-2 data with emphasis on...
TRANSCRIPT
Luis Guanter1, Karl Segl2, Hermann Kaufmann2
(1) Institute for Space Sciences, Freie Universität Berlin, Germany
(2) GFZ German Research Centre for Geosciences, Germany
End-to-End Simulation of Sentinel-2 Data with
Emphasis on Atmospheric Correction Methods
SENTINEL-2 for Science Workshop, ESA-ESRIN, 20-22 May 2014
Background activities:
1. Consolidate requirements for Sentinel-2/MSI.
2. Develop and test methods for atmospheric correction (AC) of S-2 data.
3. Explore synergy between Sentinel-2 and the EnMAP imaging spectroscopymission (VNIR-SWIR, GSD~30m).
We have adapted the end-to-end scene simulator of EnMAP to S-2.
This talk: overview of methodology and some tests.
“Raw data” (digital numbers) Calibrated Radiance Orthorectification Surface reflectance
Motivation
SENTINEL-2 for Science Workshop, ESA-ESRIN, 20-22 May 2014
Calibration
Non-linearity
Dark Signal
Absolute Calibration
Output Data
(Reflectance)
Input Data(Reflectance)
DEM, CWV Map
Sentinel-2 Scene
Simulator
Spectral Module
Spatial Module
Atmospheric
Module
Radiometric Module
Sensor Data (DN)
Bac
kw
ard
Sim
ula
tion
Fo
rwa
rd S
imu
lati
on
L1 Processors
Radiometric
Correction
Orthorectification
L2A Processor
Atmospheric
Correction
Sentinel-2 End-to-End Simulation
Applications:
• Sensitivity analysis• Development of pre-processing algorithms• Cal/Val
Level 0/1a Digital Numbers, satellite projection
Level 1b TOA radiances, satellite projection
Level 1c TOA reflectance, geographical
projection, ortho-rec.
Level 2a BOA reflectance, geographical
projection, ortho-rec.
SENTINEL-2 for Science Workshop, ESA-ESRIN, 20-22 May 2014
Forward simulation: reflectance + atmospheric radiative transfer
Input:
1) Reflectance: (1) hyperspectral reflectance data
covering the VNIR-SWIR or (2) multi/hyperspectral
VNIR reflectance data + spectral library.
2) Other: DEM, water vapour map, AOT550 (average).
Reflectance propagated to TOA radiance
(Lambertian) in high spectral resolution (and high
spatial resolution for the spatial simulator).
Synthetic DEM Synthetic WV
SENTINEL-2 for Science Workshop, ESA-ESRIN, 20-22 May 2014
Forward simulation: instrument module
Hyperspectral output convolved with the MSI spectral response functions MSI-
like TOA radiance data, Level 1b
Radiometric module: linearity, striping, noise, 12-bit quantisation - according to
missions specifications DNs, Level 1a
(Optional) Spatial module:
- pixel-wise instrument MTF, real acquisition geometry, band-specific GSD (10m & bin.)
complex, time consuming, requires input data with high spatial resolution only for
particular tests
SENTINEL-2 for Science Workshop, ESA-ESRIN, 20-22 May 2014
Forward simulation: atmospheric effects on TOA radiance data
Red-edge
Ref: CWV=2 gcm-2, AOT550=0.2
Aerosol
optical
thickness
(AOT)
Water
vapour
column
(CWV)
SENTINEL-2 for Science Workshop, ESA-ESRIN, 20-22 May 2014
L1a
L1b
B8 (10m) B8a (20m) B9 (20x60m)
Level 1 processing: radiometric calibration & spatial resampling
B9 (60x60m)
The stripes result from different gains, digital offsets and non-linearity characteristics of each pixel.
Rad
iom
etr
ic c
ali
bra
tio
n
SENTINEL-2 for Science Workshop, ESA-ESRIN, 20-22 May 2014
Atmospheric correction (Level 2a) – Aerosol optical thickness retrieval
Level 1c, starting point: top-of-atmosphere reflectance
AOT retrieval from MERIS (SCAPE-M, Guanter et al, 2008):
- Retrieval over 10km cells AOT550 mosaic as an output
- Maximum AOT threshold calculated from darkest pixels in the image
- Refinement through the inversion of AOT + vegetation/soil endmembers
- Aerosol type fixed (rural).
- VNIR so far: MODIS DDV/2.2 um approach can also be implemented.
- Atmospheric radiative transfer simulations from a look-up table (MODTRAN4)
SENTINEL-2 for Science Workshop, ESA-ESRIN, 20-22 May 2014
Differential absorption technique: pair reference + measurement channel
Exploits the ratio of bands B9 to B8a (broad B8 affected by WV, not a reference)
CWV retrieval per-pixel:
At this point AOT550 mosaic, pixel-wise CWV & elevation surface reflectance
retrieval (Lambertian surface assumed)
CWV=a+b*log(T)+c*log2(T)
T=L(B9)/L(B8a); a, b, c = f(air mass factor, B8 TOA refl.)
Atmospheric correction (Level 2a) – Water vapour & surf. reflec. retrieval
Guanter et al. RSE, 2008
B9
B8aB8
SENTINEL-2 for Science Workshop, ESA-ESRIN, 20-22 May 2014
Atmospheric correction – End-to-end simulation (noise free)
Test of consistency: end-to-end simulation run
for an agricultual site (Demmin, Germany)
- Tests for different AOT550 and CWV levels
- Ideal case: only unknowns AOT, CWV and
surface reflectance (many assumptions...)
- Offset in CWV retrieval, but very low impact
on reflectance retrieval.
- Almost perfect AOT retrieval, possibly due to
some very dark areas (and assumptions!).
Water vapour
Aerosol optical thickness
Surface reflectance
SENTINEL-2 for Science Workshop, ESA-ESRIN, 20-22 May 2014
Test: effect of atmospheric correction on S-2 vegetation indices
Motivation: Sentinel-2 to deliver an enormous amount of data. Question: when
working with vegetation indices, can we use TOA reflectance (L1c) data?
Test: calculated vegetation indices from TOA and BOA reflectance data and
analysed the differences
NDVI =(r(B7)-r(B4))/(r(B7)+r(B4)) (Tucker, 1979)
CI_red-edge =(r(B7)/r(B5) – 1 (Gitelson et al, 2003, 2006)
MTCI = r(B6)-r(B5))/(r(B5)-r(B4)) (Dash & Curran, 2004)
B5B4
B6
B7
SENTINEL-2 for Science Workshop, ESA-ESRIN, 20-22 May 2014
Test: effect of atmospheric correction on S-2 vegetation indices
NDVI=
(r(B7)-r(B4))/(r(B7)+r(B4))
CI_red-edge=
r(B7)/r(B5) – 1
SENTINEL-2 for Science Workshop, ESA-ESRIN, 20-22 May 2014
Test: effect of atmospheric correction on S-2 vegetation indices
NDVI=
(r(B7)-r(B4))/(r(B7)+r(B4))
CI_red-edge =
r(B7)/r(B5) – 1
SENTINEL-2 for Science Workshop, ESA-ESRIN, 20-22 May 2014
Test: effect of atmospheric correction on S-2 vegetation indices
MTCI=
(r(B6)-r(B5))/(r(B5)-r(B4))
NDVI=
(r(B7)-r(B4))/(r(B7)+r(B4))
SENTINEL-2 for Science Workshop, ESA-ESRIN, 20-22 May 2014
Synergy EnMAP + S-2
South Namibia
EnMAP L2 imagesR/G/B: 2.2/0.8/0.4 µm
GSD: 30 m
Bands: 242
Sentinel-2 L2A mosaicR/G/B: 2.2/0.8/0.5 µm
GSD: 20 m
Bands: 13
O-2 O-1 O+0 O+1 O+2
Spatial simulations & image mosaicing
SENTINEL-2 for Science Workshop, ESA-ESRIN, 20-22 May 2014
An end-to-end simulation tool for Sentinel-2 has beenimplemented building on previous EnMAP models:
• Allows forward simulation of S-2 L1a/b/c images for a range ofatmospheric, surface and instrumental configurations.
• The backward simulation is based on a fully automatic processing chain.
• Potential for the investigation of S-2/EnMAP synergy and Cal/Valactivities.
Sensitivity analysis indicates that climatology values could be usedfor water vapour if processing time is an issue (avoid pixel-wisewater vapour retrieval).
Test of vegetation indices: different robustness againstatmospheric effects, MTCI highly affected.
End-to-end simulator already used to produce S2-like +EnMAP data for a number of sites and users. Could also be runon the SPOT Take 5 data set if there is interest.
Summary
SENTINEL-2 for Science Workshop, ESA-ESRIN, 20-22 May 2014
Thank you for your attention!!
… and in particular to F. Gascon, P. Martimort and C. Isola for detailed
information on S-2/MSI & to the German Federal Ministry of Economic Affairs and Technology and the German Research Foundation (DFG) for
funding.