MERIS: The GLOBCOVER land cover processing lines

1
MERIS The GLOBCOVER land cover processing lines

• POSTEL Service Center P. Bicheron, M. Leroy
• GLOBCOVER Team F. Niño, P.Defourny, C.Vancutsem,
  C.Brockmann, U. Krämer, B. Berthelot, B. Miras,
  M. Huc, H. Nguyen
• ESA O. Arino, F. Ranera, H. Laur

2
Objective

• The GLOBCOVER project started in April 2005 and
  will end in June 2007
• Develop and demonstrate a service of production
  of a global land cover map for 2005
• at 300 m resolution using MERIS
• with FAO Land Cover Classification System
• Two deliverables
• A hardware software automatised system
• Two validated products
• Time composited surface reflectances over Dec.
  2004 and June 2006
• Thematic classes map

3
Processing line scheme
4
Overview

• The GLOBCOVER lines represents a major effort and
  improvement in the processing of MERIS FR
  products
• Spectral Surface Reflectances are generated with
  a
• better geometric accuracy than current MERIS
  products,
• better cloud/shadow detection,
• a new compositing method

5
Geometry requirement

• Geometric properties requirements
• Co-registration of MERIS products RMSE error lt
  150m
• Absolute geo-location RMSE error lt 150m
• Tools
• Geo-location AMORGOS 2.0
• Quality assessment MEDICIS

Correlation points
Local translation that maximize similarity
measurements between images
6
Geometric performance

• Relative validation
• Sites Madagascar, Spain, Tunisia, Romania,
  Poland Season all
• 146 couples MERIS
• Absolute validation
• Site Madagascar, Spain
  Season summer
• First results on 10 Landsat images
• Conlusions
• RMSE in lat and lon are similar gt no specific
  problem
• Standard deviation of RMSE is satifsfactory (i.e.
  small)
• Relative and absolute validation results meet
  well Globcover specification (lt150m)

7
Atmospherical correction

• A Neural Network relying on MOMO method has been
  selected because of good validation results
  (ESA-Albedo map project)
• The atmospheric correction is a complex process
• L1b radiance to reflectance conversion
• Aerosol correction requires aerosol optical
  depths from MODIS-8 days
• Gaseous absorption correction
• Ozone from ECMWF auxiliary data present at MERIS
  L1b tie-points
• O2 from b11/b10 ratio, H2O from B15/b14 ratio
• LUTs for transmission assessment

8
Atmospherical correction
Atm. Corrected RGB
L1B RGB
9
Cloud detection

• Two methods are used
• 1. Cloud Probability Method atmospherical scheme
  MOMO (Preusker, Freie Universität Berlin) through
  a neural network
• CLEAR_1, CLOUD_1
• 2. Blue Bands Cloud Screening global thresholds
  using 443, 753, 760, 865 nm
• 4 states CLEAR_2, THIN CLOUD_2, DENSE_CLOUD_2 and
  SNOW_2
• Final cloud detection combines both excluding
  snow
• Update to come for the end of 2006
• Pressure threshold depending on altitude
• Cloud shadow
• Using CTP instead of constant height
• Bright sand test
• Potential use of climatology (TBC)
• Spectral climatology, MERIS albedo map

10
BRDF Correction and Compositing

• For BRDF correction, two methods are used
• The Mean Composite (Vancutsem et al., 2002) is
  processed as a first reference Surface Spectral
  Reflectance on a compositing period equal to 51
  days
• The CYCLOPES (Hagolle et al., 2004) method is
  then applied to detect valid Surface Spectral
  Reflectance
• Removal of Outliers (residual thin clouds,
  aerosols, shadows) with an iterative procedure
• Compositing
• A Mean is applied over the valid Surface
  Spectral Reflectance for 3 temporal windows
• Bi-Monthly (1st-15th and 16th to 30th)
• Monthly (centered on the 15th)
• Seasonal (4 seasons of 3 months)

11
First GlobCover 300 m Mosaic
Colored composition using 670nm, 865 nm, 550 nm
12
Turkey Bosphore Strait Colored composition (865
nm, 560 nm, 412 nm)
13
National focusNile Valley
Monthly composite (May) over a 2.5 x 2.5 area
14
Regional compositeFrance

• Seasonal composite (Sep-Nov 2005) over a 10 x 10
  area

15
Regional composite Spain/Morocco

• Seasonal composite (Sep-Nov 2005) over a 10 x 10
  area

16
Next milestones

• September 2006
• Monthly mosaics over Europe (1 year) and at
  global scale (1 month)
• GLOBCOVER landcover map over Europe (draft)
• December 2006
• Monthly mosaics global full year 2005
• GLOBCOVER landcover map global (draft)
• June 2007
• GLOBCOVER V1 landcover map global