Clouds/haze detection methodology

Cloud detection and mapping in Data
Warehouse Phase 2
Objectives
1. Introduce standard cloud classification schemes of
CCMs
2. Show special approach with CCMs for CORE datasets
a) Cloud classification scheme suggested by EC
b) Cloud handling in VHR_IMAGE_2015
c) Cloud handling in HR_IMAGE_2015
d) Clouds in the master shapefile
3. Clarify Sentinel-2 future cloud masking capacity
1)
2a) Cloud classification scheme suggested by EC
1) Almost transparent haze
2) Semi-transparent haze/clouds
3) Clouds
4) Shadows
5) Snow
2b) Cloud handling in VHR_IMAGE_2015
• The objective is to have a 100% cloud-free coverage of Europe
• Europe is sub-divided into 136 large regions assigned to different CCMs, plus 5 DOM regions
• There is a nominal max threshold of 5/10/20% cloud per production unit that is accepted per
Zone; such production units can vary in size and be adapted/cut in order to match the max
cloud notation:
• CSPs can a-priori not reject production units that fall below those cloud coverage max values,
however could reject based on haze interpretation deviating from the native haze encoding
of CCMs; will the master shapefile declare the cloud notation of the original data strip or of
the production unit?
• CCMs might propose additional production units that are useful for completion of coverage
but have higher CC: those could be rejected by CSPs
• Any cloudy km2 will be deducted from the respective price pro-rated
2c) Cloud handling in HR_IMAGE_2015
• The objective is to have at least one 100% cloud-free coverage
of Europe in ZONE A, and 2 100% cloud-free coverages in
zones B and C, and possibly one 100% cloud free coverage in
Zone D (French DOMs)
• Europe is sub-divided into 39 countries plus 5DOM regions +Faroe Island
• There is a nominal max threshold of 5-20% cloud per product that is accepted dependent on
the Zone;
• Standard scenes sizes will be delivered (quarter scenes in case of GAF?), having native cloud
cover notation (unshifted in case of GAF)
• GAF: the CCME will pre-select all products with a cloud cover ≤ 50%. The CCME will
furthermore pre-assign a sub-selection of those products to COV#1 and COV#2, to indicate
an optimised coverage composition. Note: this will be solely based on Resourcesat-2 data,
and not considering Spot-5 data.
• Airbus: CCM will deliver ANY Spot-5 data acquired in 2014, for information to CSPs and
selection of GAF data; Hazes are not encoded in Spot5 processing lines, only the presence of
clouds will be reported
• CSPs can a-priori not reject products that fall below the defined cloud coverage max values,
however could reject CC >20% and rejections based on haze interpretation deviating from
the native haze encoding of CCMs;
• Any cloudy km2 will be deducted from the respective price pro-rated
2d) Cloud handling in HR/VHR_IMAGE_2015 Master shape-file
workflow monitoring tool, for:
 CCMEs to provide synopsis of acquisition achievements ;
 ESA to propose them to EEA for selection of best scenes;
 Tracking status of each strip or scene, from proposal/delivery through a set of key attributes.
MST content will be updated in turns by CCME for data proposal, EEA (CSP) for data selection ,
ESA for order confirmation, CCME for data delivery.
Clouds will be classified based on the
native information of each individual
CCM
Cloud info will be inserted into the
master shapefile as part of the
procedure [COPE-GSOP-EOPG-TN-150008 , V1.1, issued 9 March]
 CC
Percentage of cloud cover,
according to the native cloud cover
notation of each CCM, per original
strip or production unit?
 Haze Encoding of haze (on a scene
basis) as defined in DAP-S ; Manual
classification, where available values
are: “no haze”, “almost transparent
haze”, “semi-transparent haze,“cloud”
Sentinel-2 Cloud Detection Overview
•
•
•
Sentinel-2 has cloud detection algorithms at two levels:
•
Level-1C Product
•
Level-2A Product
Level-1C cloud detection algorithm is based on a simple algorithm
identifying the following cloud classes:
•
Opaque clouds
•
Cirrus clouds
Level-2A cloud mask is based on a more refined algorithm
identifying the following cloud classes:
•
Thin Cirrus
•
Clouds High Probability
•
Clouds Medium Probability
•
Clouds Low Probability
•
Cloud Shadows
Sentinel-2 Level-1C Cloud Detection
•
The calculation is performed at a spatial resolution of 60 m.
•
Opaque clouds detection algorithm:
•
•
B1/B2 (blue) and B11/B12 (SWIR) used to discriminate snow
and opaque clouds.
•
B10 (cirrus) to detect the ice high-altitude/cirrus clouds
(applied after first criteria based on B1/B2).
Cirrus clouds detection algorithm:
•
High reflectance in band B10.
•
Low reflectance in band B1/B2.
Sentinel-2 Level-2A Cloud Detection
•
The Level-2A Scene Classification algorithm uses as input the topof-atmosphere reflectances from Level-1C product.
•
The Scene Classification algorithm identifies 10 scene classes:
dark areas, cloud shadows, vegetation, bare soils, water, cloud low
probability, cloud medium probability, cloud high probability, thin
cirrus and snow.
•
Thresholds are applied on band ratios and indexes like the
Normalized Difference Vegetation (NDVI) and Normalized
Difference Snow Index (NDSI).
•
For each of these thresholds tests, a level of confidence is
associated.
•
At the end of the processing chain a probabilistic cloud mask
quality map and a snow mask quality map is produced.
Use of Level-2A clouds information
TOA reflectance
(RGB composite = bands at
665, 560 and 443 nm)
Cirrus band image
(1375 nm)
Data simulated using AVIRIS provided by NASA
BOA reflectance
(After cirrus detection and
atmospheric correction)
Future evolution of CCMs towards
harmonsation
Harmonisation of cloud cover delineation and cloud shadow
detection among CCMs is on the wishlist of EEA, however the
CCMs have their ground segments that have partly not much
flexibility, and the number of spectral bands is different among
CCMs, and absence of relevant cloud bands is an issue that
can not easily be overcome.
A good 1st step could be to align the nomenclature and use
e.g. cloud shadow, opaque clouds and cloud shadow
ESA UNCLASSIFIED – For Internal Use