proba-v snap toolbox s3 snow idepix cloud shadow · snap status and update serco business •icor...

20
Serco Business Proba-V QWG-08 Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow Carsten Brockmann, Jan Wevers 07.11.2018

Upload: others

Post on 22-Aug-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

Proba-V QWG-08

Proba-V SNAP ToolboxS3 Snow

IdePix Cloud Shadow

Carsten Brockmann, Jan Wevers

07.11.2018

Page 2: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

SNAP Status and Update

• iCOR incompatibility with SNAP resolved by VITO

• SNAP evolution KO 16.11.18• Time series exploration• Improved SNAPPY• Product groups (virtual stacks)• Improved support for multi-size products• GPF performance enhancements• Cloud access – data and processing• Machine Learning tools• Graph builder support for all operators• ESA SIP format support• OLCI & SLSTR Synergy L1C Tool

Page 3: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

Snow Products

• ESA SEOM Project – Sentinel 3 for Snow• Geological Survey Denmark (Jason Box)• Brockmann Consult (Olaf Danne)• Institute des Géoscience de l’Environnement (Maxim Lamare)• Meteo France (Marie Dumont)

• Data Products• Snow mask• Bare ice indicator• Polluted snow indicator• Spectral snow albedo• Broadband snow albedo• Specific surface areas• Snow grain diameter

Snow albedo is critical for climate studies!

Page 4: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

Snow - Ice - Cloud Classification

M. Lamare, S3Snow project

S3 rgb colour composite Cloud mask (red+blue)

O. Danne, S3Snow project

Page 5: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

Grain Diameter

J. Box, S3Snow PM5 Sep. 2018Greenland, 20180713

Page 6: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

Validation - SSA

Page 7: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

Validation – Spectral Albedo, Antarctica Dome-C

M. Lamare, S3Snow PM5 Sep. 2018

Page 8: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

Validation – Spectral Albedo, Antarctica ASUMA

M. Lamare, S3Snow PM5 Sep. 2018

Page 9: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

Validation – Spectral Albedo, Antarctica ASUMA

M. Lamare, S3Snow PM5 Sep. 2018

Page 10: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

Potential transfer to Proba-V

• Algorithm Requirements• Rayleigh correction no specific requirements• Cloud – snow screening

• S3: O2 band (height) & SWIR band• PB-V does not provide cloud height• PB-V SWIR band is better suited than S3 band 21 (1020nm)

• Bare ice blue, red and NIR bands• PB-V: Radiometric quality of bands to be tested

• Snow properties• The reflectance spectrum over (pure) snow can be modeled with 4 parameters 𝑙, 𝑅0, 𝑓,𝑚 . • 𝑙, 𝑅0, 𝑓,𝑚 allow determination of snow properties (which determine the reflectance)• RT inversion using 4 (arbitrary) bands in VIS/NIR range to retrieve the 4 parameters• Albedo, grain size and SSA are calculated from 𝑙, 𝑅0, 𝑓, 𝑚• S3: 400nm, 560nm, 865nm, 1020nm• PB-V: spectral and radiometric suitability of VIR-NIW-SWIR bands to be tested

• Antarctia Dataset of Proba-V • Valuable data source for scientific snow studies• Demonstration of application of Proba-V

• Greenland is observed in nominal observations of Proba-V• Complement S3 products

Page 11: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

Proba-V QWG #8

Sentinel-2 cloud shadow algorithm & processor

Carsten Brockmann, Dagmar Müller,

Grit Kirches, Michael Paperin, Olaf Danne, Tonio Fincke, Jan Wevers

06.11.2018

Hamburg

Page 12: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

Proba-V QWG 06.11.2018

S2 cloud shadowS2 cloud shadow processorCloud shadows can be calculated from geometry if cloud top and base are known (e.g. S3 allows cloud top height from O2 or temperature)But: no estimation of cloud top/base height from S2 bands possible; therefore there is no exact geometrical solution.Instead, a maximum cloud top height as a function of latitude is assumed.

Algorithms are based on sun geometry and the cloud mask, searching for dark pixel in an area of potential shadow.

Methodologies to identify cloud shadows:

1. Shifting the cloud mask towards the surface reflectance minimum along the illumination path

2. Clustering the surface reflectances within the potential shadow area for each cloud and find the darkest cluster.

3. Combination of 1. & 2.: Keeping only clustered shadow areas, which coincide with the shifted cloud mask

Potential cloud shadow area based on illumination geometry

Page 13: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

Proba-V QWG 06.11.2018

Shifted cloud mask

Best Offset Algorithm• Cloud mask is shifted along the

illumination path away from the sun.• At each step, the mean reflectance of

surface pixels underneath the shifted cloud mask is calculated.

-> Mean reflectance as function of relative offset, starting from a cloud pixel.

Statistics are calculated for land, water and combined pixels independently.

For each S2 tile of a granule:Scaled mean reflectances underneath the shifted cloudmask as a function of the shift along the illumination path.Minimum of the averaged functions (black) gives the best offset.

offset

Pros- Simple and fast approach.- Good first guess for cloud shadow mask and averaged

distance between cloud and its shadow.

Cons- Stable statistics only for shift of entire cloud mask, not

individual clouds.- Single cloud top height for all clouds is assumed.

Page 14: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

Proba-V QWG 06.11.2018

Cluster analysis

Cluster Analysis Algorithm• For each individual cloud, the potential

shadow area is clustered.• The darkest clusters, if below a certain

threshold, are supposed to represent the cloud shadow.

Pros- Individual solution for each cloud.- Even shadows of unidentified clouds can be detected, if they

lie within a potential shadow area.

Cons- Shadows inbetween clouds are not found, because they are

too bright (adjacency effects of clouds)- Bare soil can be part of the darkest clusters.

Summary

• Both methods are strongly dependent on the quality of the pixel classification and the given cloud mask.

• The estimation of the potential cloud shadow area is depending on the sun and view geometry -> apparent sun azimuth angle.

The results of Cluster Analysis and the Shifted cloud mask are compared and combined, analysing distance and brightness information to adjust for the restrictions of both methodologies.

Page 15: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

Proba-V QWG 06.11.2018

Potential cloud shadow: Apparent sun azimuth

Apparent sun azimuth• Projecting clouds to the surface at view_zenith

> 0, distorts their true positions (nadir view). • Nevertheless, the true position gives rise to

the shadows.• As the cloud height is unknown, their position

can not be corrected. • Instead, the apparent sun azimuth angle is

estimated from the viewing and sun geometry.

Potential shadow area based on Apparent Sun azimuth

Potential shadow area based on Sun azimuth in S2 product at edge of the swath

view

sun

Projected

position

cloud

Actual position

(nadir)

shadow

Sun azimuth

Projected

position

Actual position

(nadir)shadow

Apparent Sun azimuth

But: Finding a single, representativeview azimuth angle for a granuleis not trivial.

Page 16: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

Proba-V QWG 06.11.2018

Examples 20180212 T32QNFRGB

Clustered shadow

Coinciding shadow

Shifted shadow

Page 17: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

Proba-V QWG 06.11.2018

Examples 20161226 T45RYL

NDSI<0 and quality_flag.bright

Cloud postprocessing

RGB

Clustered shadow

Coinciding shadow

Shifted shadow

Page 18: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

Proba-V QWG 06.11.2018

Processing overview

Left: clustered cloud shadow. Many bare soil areas are flagged as coud shadow.

Middle: shifted cloud mask (with adjustments)

Right: Keeping only clustered shadow areas, which coincide with the shifted cloud mask

Page 19: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

Proba-V QWG 06.11.2018

SNAP Processor

• Algorithm is fully implemented in SNAP(IdePix for S2)

• Processing time depends on spatial resolution• Number of pixel

• Complexity of geometric shapes

• We calculate only on 60m

• Typical processing time for 1 L1C granule = 1 min 30 sec on a Desktop PC

Page 20: Proba-V SNAP Toolbox S3 Snow IdePix Cloud Shadow · SNAP Status and Update Serco Business •iCOR incompatibility with SNAP resolved by VITO •SNAP evolution KO 16.11.18 • Time

Serco Business

Summary - Recommendations

1. Evolution of Proba-V Toolbox (re-iterated from previous QWG)• Add Proba-V support to IdePix

• Add Proba-V support to those Soil & Vegetation Radiometric & Water Indices which are not supported by Global Land Service

2. Study to transfer snow retrieval from S3 to PB-V• Could become a processor in Proba-V Toolbox

• Could be made available on MEP together with Antarctica dataset

3. Compare cloud shadow from PB-V with S2 method in IdePix