Transcript pres_PLUMMER-CLOUDATSR-MER_AATSR - Envisat

Establishing the GLOBCARBON
cloud detection system over land
for the ATSR sensor series
Stephen Plummer (IGBP@ESA)
MERIS-AATSR
26-30 Sept 2005
Carbon Data Assimilation
Geo-referenced
Geo-referenced
emissions
emissionsinventories
inventories
Climate and weather
fields
Ocean time series
Biogeochemical
pCO2
Surface
observation
pCO2
nutrients
Water column
inventories
Atmospheric
Atmospheric
measurements
measurements
Remote
Remote sensing
sensing of
of
atmospheric
atmospheric CO
CO22
Atmospheric
Atmospheric Transport
Transport
Model
Model
Ocean
Ocean Carbon
Carbon
Model
Model
Coastal
Coastal
studies
studies
Optimised
Optimised
fluxes
fluxes
Terrestrial
Terrestrial
Carbon
Carbon Model
Model
rivers
Lateral fluxes
Data
assimilation
link
Optimised
Optimised
model
model
parameters
parameters
Eddy-covariance
flux towers
Biomass soil
carbon
inventories
Ecological
studies
Ocean remote sensing
Ocean colour
Altimetry
Winds
SST
SSS
Remote sensing of
vegetation properties
Growth cycle
Fires
Biomass
Radiation
Land cover/use
To feed in to this Earth observation must deliver long time series estimates of global
vegetation behaviour.
MERIS-AATSR
26-30 Sept 2005
GLOBCARBON Objectives
•
develop a service quasi-independent of the original Earth
Observation source.
•
focus on a system to estimate:
 Burned area
 fAPAR and LAI
 Vegetation growth cycle
•
cover six complete years: 1998 to 2003 (now up to 2007)
•
cover VEGETATION, ATSR-2, ENVISAT (AATSR, MERIS)
•
be applicable to existing archives and future satellite systems
•
be available at resolutions of ¼, ½ degree and 10km with
statistics
•
build on the existing research experience
MERIS-AATSR
26-30 Sept 2005
Requirements – ATSR Series
•
Processing of only those pixels not affected by cloud,
snow, cloud shadow or atmosphere
•
This requires processing of approximately 500,000 ATSR2 scenes and 25,000 AATSR striplines
•
All processing must be automatic
•
GLOBCARBON requires the implementation of a effective
cloud detection system over land but the existing system
was designed for oceans
MERIS-AATSR
26-30 Sept 2005
ATSR-2 Cloud system
Original RGB (1.6, 0.87, 0.67)
ATSR-2 Cloud masked RGB
[Remaining cloud has same
cloud flag as clear land (1027)]
MERIS-AATSR
26-30 Sept 2005
ATSR Cloud Mask
• Cloud Mask = 9 tests
• Implemented on Land = 4 tests
– Thin cirrus 11/12μm
– Medium/high level 3.7/12μm
(not daytime)
– Fog/low status 3.7/11μm
(not daytime)
– 11μm spatial coherence 11μm
RESULT = NEED A NEW CLOUD DETECTION SYSTEM
MERIS-AATSR
26-30 Sept 2005
A ‘new’ Cloud Detection System
SNOW
• GLOBCARBON
tight
schedule – adopt existing
methods
• GLOBCARBON
high
processing throughput –
simple
methods,
low
computer cost
• Tested CLAVR, APOLLO
(2003), GLOBSCAR
• APOLLO (2003) chosen
with added ‘bells and
whistles’
• Added pre-APOLLO snow
detection
APOLLO 2003
Thermal
Gross
Cloud
Prob 1
Thin
Cloud
Dynamic
Vegetation
Test
Prob 2
Prob 3
Merge
Probs
Prob APOLLO
Cloud
Mask
MERIS-AATSR
Thermal-SWIR
Histogram
26-30 Sept 2005
Snow Detection
•
Implementation of MODIS
method (Hall et al. 2001)
•
Requirement: GREEN, RED,
NIR, SWIR, 11μm
•
Pre-screening of cloud
•
Based on Normalised
Difference Snow Index:
NDSI  555  1600  / 555  1600 
Basic
Snow in Forest
BT11i , j  277 K
BT11i , j  277 K
NDSI i , j  0.4
0.1  NDSI i , j  0.4
 555i , j  0.11
NDVI i , j  MSNOW 1i , j
865i , j  0.1
NDVI i , j  MSNOW 2 i , j
 670i , j  0.1
 865i , j  0.1
MERIS-AATSR
26-30 Sept 2005
Thermal Gross Cloud Test
•
As with ATSR-2 cloud but
implemented over land
•
Requirement: RED, NIR,
11μm, 12μm
•
NIR/RED used to mask off
pixels not cloud (NIR/RED
less than 1.6).
•
Threshold found as 2K less
than minimum BT at 11μm
•
Threshold applied to BT at
12μm
•
Probability range between
threshold and threshold
minus 20K
RED = CLOUD, GREEN = POSS CLOUD, BLUE = CLEAR
MERIS-AATSR
26-30 Sept 2005
Thin Cloud Test
•
As with ATSR-2 cloud but
implemented over land
•
Requirement: 11μm,
12μm, LUT, SAT ZEN
•
Threshold from LUT of
Thermal Brightness
Difference and secant of
SAT ZEN
•
Probability range between
threshold ± half min
distance between
minimum or maximum of
TBD for image
RED = CLOUD, GREEN = POSS CLOUD, BLUE = CLEAR
MERIS-AATSR
26-30 Sept 2005
Dynamic Vegetation Test
•
Requirement: RED, 11μm
•
Test 1: BT11 < Threshold of
274.5K (or if desert 290K)
and RED > 0.2
•
Test 2: RED >0.6
•
Probability range = 0.1 ±
threshold (RED) and 5K ±
threshold (BT11)
•
Probability Test 1 product of 2
parts
•
Final Probability maximum of
Test 1 and Test 2
RED = CLOUD, GREEN = POSS CLOUD, BLUE = CLEAR
MERIS-AATSR
26-30 Sept 2005
APOLLO Final Probability
•
Clear pixels: the probability is 0 in all three tests
•
Cloud pixels: the probability of 1 occurs in any of the tests
•
Suspect pixels: the probability is maximum probability for
values between 0 and 1
•
Pixels are masked where final probability > 0.5
RED =
CLOUD,
GREEN =
POSS
CLOUD,
BLUE =
CLEAR
MERIS-AATSR
26-30 Sept 2005
SWIR Thermal Test
• Performed on pixels not
flagged by APOLLO
• Requirement: SWIR,
11μm
• Number APOLLO pixels
> 30 Number clear
pixels > 2620 Minimum
BT11 < 280K.
• Thresholds based on
histogram maxima
• Probability is product of
probabilities for BT11
and SWIR
• Pixels masked if
probability > value
already present
MERIS-AATSR
26-30 Sept 2005
Results - Amazon
MERIS-AATSR
26-30 Sept 2005
Results - Kalahari
MERIS-AATSR
26-30 Sept 2005
Results - Siberia
MERIS-AATSR
26-30 Sept 2005
Conclusions
• ATSR/AATSR Cloud Detection system was developed to serve
GLOBCARBON based on APOLLO (2003) and MODIS snow detection
• The system proved effective in 3 different biomes and with highly
variable cloud when tested on 49 ATSR-2 images. In only one case did
was the system not sufficiently effective.
• Further testing is required since the examples are limited in time (1
month)
and
may
not
represent
all
cases.
• The coefficients used in the system are exactly as used in MODIS snow
detection and APOLLO (2003). These may need adjusting for spectral
characteristics of the ATSR series.
• The snow detection in particular misses far too much snow while also
detecting some cloud.
• The system has been implemented for processing 500,000 ATSR-2
scenes and 25,000 AATSR orbits.
MERIS-AATSR
26-30 Sept 2005
Acknowledgements
• Many thanks to:
– Walter Heyns at VITO for testing the IDL code and
pointing out errors prior to its implementation in
the GLOBCARBON processor.
– the developers of APOLLO – Karl Kriebel,
Gerhard Gesell, Martina Kästner and
Herman Mannstein.
– the MODIS snow team for providing clear details
for the implementation of the algorithm
– ESA, especially Olivier Arino, for supporting the
GLOBCARBON idea through thick and thin.
MERIS-AATSR
26-30 Sept 2005
Failures
Cloud dominates
the SWIR-BT11
histograms so the
determination of
the thresholds is
not effective
MERIS-AATSR
26-30 Sept 2005