Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES...

36
GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot I ”Operation of the Global Land Component” Framework Service Contract N° 388533 (JRC) PRODUCT USER MANUAL FRACTION OF VEGETATION COVER (FCOVER) VERSION 1 Issue I2.20 Organization name of lead contractor for this deliverable: VITO Book Captain: Bruno Smets Contributing Authors: Roselyne Lacaze

Transcript of Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES...

Page 1: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Gio Global Land Component - Lot I

”Operation of the Global Land Component” Framework Service Contract N° 388533 (JRC)

PRODUCT USER MANUAL

FRACTION OF VEGETATION COVER (FCOVER) – VERSION 1

Issue I2.20

Organization name of lead contractor for this deliverable: VITO

Book Captain: Bruno Smets

Contributing Authors: Roselyne Lacaze

Page 2: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 2 of 36

Dissemination Level PU Public X

PP Restricted to other programme participants (including the Commission Services)

RE Restricted to a group specified by the consortium (including the Commission Services)

CO Confidential, only for members of the consortium (including the Commission Services)

Page 3: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 3 of 36

Document Release Sheet

Book captain: Bruno Smets Sign

Date 23.03.2015

Approval: Roselyne Lacaze Sign

Date 23.03.2015

Endorsement: Michael Cherlet Sign Date 02.04.2015

Distribution: Public

Page 4: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 4 of 36

Change Record

Issue/Rev Date Page(s) Description of Change Release

28.03.2013 All

Initial version from FP7/geoland2 document,

Issue I1.31, 05.03.2012. Based upon

SPOT/VGT input data

I1.00

I1.00 23.04.2014 All Update to PROBA-V input data I2.00

I2.00 25.08.2014 All Clarifications and editorial changes after

external review; update of the validation results I2.10

I2.10 23.03.2014 33-37 Include conclusions of the full quality

assessment over one year of products I2.20

Page 5: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 5 of 36

TABLE OF CONTENTS

1 Background of the document ................................................................................................ 9

1.1 Executive Summary .................................................................................................................... 9

1.2 Scope and Objectives ................................................................................................................. 9

1.3 Content of the document ........................................................................................................... 9

1.4 Related documents .................................................................................................................. 10

1.4.1 Applicable documents ................................................................................................................................ 10

1.4.2 Input ............................................................................................................................................................ 10

2 Review of Users Requirements ............................................................................................ 11

3 Algorithm ............................................................................................................................ 13

3.1 Definition of the variable ......................................................................................................... 13

3.2 Outline ..................................................................................................................................... 13

3.3 The retrieval Methodology ....................................................................................................... 14

3.3.1 Input data ................................................................................................................................................... 14

3.3.2 Pre-processing steps ................................................................................................................................... 16

3.3.3 Biophysical retrieval methodology ............................................................................................................. 18

3.4 Limitations of the Product ........................................................................................................ 22

4 Product Description ............................................................................................................. 23

4.1 File Format ............................................................................................................................... 23

4.2 Product Content ....................................................................................................................... 25

4.2.1 Data File ...................................................................................................................................................... 25

4.2.2 Quicklook .................................................................................................................................................... 27

4.3 Product Characteristics ............................................................................................................ 27

4.3.1 Projection and Grid Information ................................................................................................................. 27

4.3.2 Spatial information ..................................................................................................................................... 28

4.3.3 Temporal Information ................................................................................................................................. 28

4.3.4 Data Policies ................................................................................................................................................ 29

4.3.5 Contacts ...................................................................................................................................................... 29

5 Validation Results ............................................................................................................... 30

6 References ........................................................................................................................... 35

Page 6: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 6 of 36

List of Figures

Figure 1: Main steps of the retrieval algorithm applied to PROBA-V VGT-like TOA reflectances to

get the biophysical variable (FCover). ..................................................................................... 14

Figure 2: PROBA VGT-like P segments contents ........................................................................... 16

Figure 3: Flow chart showing the principle of the biophysical retrieval methodology. ..................... 19

Figure 4: Cumulated frequency computed over the BELMANIP2 sites for FCover during the period

2003-2005 for the original CYCLOPES V3.1 products (dashed line) and after the „best

available product“ obtained by scaling CYCLOPES V3.1. ....................................................... 20

Figure 5: Neural network architecture used to estimate FCover from TOC reflectances and

illumination geometry. .............................................................................................................. 21

Figure 6: Theoretical performances of FCover neural network (presented as boxplots with the solid

line corresponding to the median and the gray levels containing respectively 50%, 90% and

99% of the population). ............................................................................................................ 21

Figure 7: Regions of FCover product. ............................................................................................. 24

Figure 8: The cut-out of tiles ........................................................................................................... 28

Page 7: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 7 of 36

List of Tables

Table 1: Spectral characteristics of VGT-2 and PROBA-V sensors ............................................... 15

Table 2: The global FCover products available .............................................................................. 19

Table 3: Definition of the continental tiles. ...................................................................................... 24

Table 4: Range of values and scaling factors of FCover and FCover uncertainty. ......................... 25

Table 5: Description of the quality flag of FCover. .......................................................................... 26

Table 6: Description of HDF-5 file attributes ................................................................................... 26

Table 7: Description of HDF-5 layer attributes ................................................................................ 27

Table 8: Color coding for quicklook FCover images ....................................................................... 27

Table 9: Summary of Product Evaluation (PROBA-V GEOV1). The plus (minus) symbol means

that the product has a good (poor) performance according to this criterion. ........................... 33

Page 8: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 8 of 36

List of Acronyms

ATBD Algorithm Theoretical Basis Document

BRDF Bidirectional Reflectance Distribution Function

CEOS Committee on Earth Observation Satellite

DBF Deciduous Broadleaf Forest

EBF Evergreen Broadleaf Forest

FAPAR Fraction of Absorbed Photosynthetically Active Radiation

GEOV1 Products including LAI, FAPAR, FCover Version 1

GCOS Global Climate Observing System

GIO GMES Initial Operation

GL Global Land

HDF Hierarchical Data Format

LAI Leaf Area Index

LPV Land Product Validation Subgroup of CEOS

MODIS Moderate Resolution Imaging Spectroradiometer

MODC5 MODIS vegetation products Collection 5

MVC Maximum Value Composite

NCEP National Centres for Environmental Prediction

NLF Needle Leaf Forest

NDVI Normalized Difference Vegetation Index

NIR Near Infrared

NMOD Number of valid observations used for the normalisation

PROBA-V VEGETATION sensor on PRoject for OnBoard Autonomy platform

PUM Product User Manual

QFLAG Quality Flag

RMSE Root Mean Square Error

SVP Service Validation Plan

SWIR Short Wave Infrared

TOA Top Of Atmosphere

TOC Top Of Canopy

VGT VEGETATION sensor onboard SPOT satellites

VNIR Visible and Near Infrared

WGS World Geodetic System

ZIP Compressed archive file format

Page 9: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 9 of 36

1 BACKGROUND OF THE DOCUMENT

1.1 EXECUTIVE SUMMARY

The Global Land (GL) Component in the framework of GMES Initial Operations (GIO) is earmarked

as a component of the Land service to operate “a multi-purpose service component” that will

provide a series of bio-geophysical products on the status and evolution of land surface at global

scale. Production and delivery of the parameters are to take place in a timely manner and are

complemented by the constitution of long term time series.

The Product User Manual (PUM) is a self-contained document which gathers all necessary

information to use the product on an efficient and reliable way.

This document describes the GL Fraction of Vegetation Cover (FCover) Version 1 products,

derived from PROBA-V daily data. The FCover product is a 30-days composite, updated every 10

days using a moving window. The product is projected on a regular latitude/longitude grid with a

resolution of 1/112° (approx. 1km at the equator). It is delivered in 10°x10° tiles and in continental

tiles covering the whole globe (from 180°E to 180°W and from 75°N to 60°S). It is provided in multi-

band HDF5 format (by default format) containing the fraction of vegetation cover value, its

uncertainty, a quality flag and other additional information.

Note: The PROBA-V FCover V1 is disseminated as a “Demonstration” product, meaning that

some limitations still exist. Indeed the quality assessment of the PROBA-V normalized TOC

reflectances (TOC-r products), which are the input of neural networks to calculate the GEOV1 LAI,

FAPAR, FCover variables, is in progress. However, a preliminary analysis has showed systematic

positive differences between PROBA-V and SPOT/VGT TOC-r in near infrared and SWIR spectral

bands (PROBA-V TOC-r slightly higher than SPOT/VGT TOC-r). The reasons of such differences

are under investigation. Once fixed, a re-processing action of TOC-r and, then, of GEOV1 LAI,

FAPAR, FCover will be performed and a new quality assessment.

1.2 SCOPE AND OBJECTIVES

The PUM is the primary document that users have to read before handling the products.

It gives an overview of the product characteristics, in terms of algorithm, technical characteristics,

and main validation results.

1.3 CONTENT OF THE DOCUMENT

This document is structured as follows:

Chapter 2 recalls the users requirements

Chapter 3 presents a description of the algorithm.

Page 10: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 10 of 36

Chapter 4 describes the technical characteristics of the product.

Chapter 5 summarizes the validation procedure and the results.

1.4 RELATED DOCUMENTS

1.4.1 Applicable documents

AD1: Annex II – Tender Specifications to Contract Notice 2012/S 129-213277 of 7th July 2012

AD2: Appendix 1 – Product and Service Detailed Technical requirements to Annex II to Contract

Notice 2012/S 129-213277 of 7th July 2012

1.4.2 Input

Document ID Descriptor

GIOGL1_SSD Service Specifications Document of the Copernicus

Global Land Service.

GIOGL1_SVP Service Validation Plan of the Copernicus Global

Land service.

GIOGL1_ATBD_FCOVERV1 Algorithm Theoretical Basis Document of the

FCOVER Version 1 product

GIOGL1_ATBD_PROBA2VGT ATBD of the pre-processing module applied to

transform PROBA-V data in SPOT/VGT-like data

GIOGL1_ATBD_TOCRef ATBD of PROBA-V TOC reflectances

GIOGL1_PUM_TOCRef Product User Manual of the PROBA-V TOC

reflectances, Issue I2.00

GIOGL1_QAR-PROBAV-GEOV1 Quality Assessment Report of the PROBA-V GEOV1

(LAI, FAPAR, FCover Version1) products

1.4.3 External document

PROBAV-PUM Product User Manual of PROBA-V products,

available at http://proba-v.vgt.vito.be

Page 11: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 11 of 36

2 REVIEW OF USERS REQUIREMENTS

According to the applicable document [AD2], the user’s requirements relevant for Fraction of

Vegetation Cover are:

Definition:

o Fractional cover refers to the proportion of a ground surface that is covered by

vegetation.

Geometric properties:

o location accuracy shall be 1/3rd of the at-nadir instantaneous field of view

o pixel co-coordinates shall be given for centre of pixel

Geographical coverage:

o Geographic projection: regular lat-long

o Geodetical datum: WGS84

o Pixel size and accuracy: 1/112°; accuracy: minimum 10 digits

o Coordinate position: centre of pixel

o Window coordinates:

Upper Left:180°W-74°N

Bottom Right: 180°E 56°S

Frequency and timeliness:

o As a baseline the biophysical parameters are computed by and representative of

dekad, i.e. for ten-day periods (“dekad”) defined as follows: days 1 to 10, days 11 to

20 and days 21 to end of month for each month of the year.

o As a trade-off between timeliness and removal of atmosphere-induced noise in

data, the time integration period may be extended to up to two dekads for output

data that will be asked in addition to or in replacement of the baseline based output

data.

o The output data shall be delivered in a timely manner, I. e. within 3 days after the

end of each dekad.

Ancillary information:

o the number of measurements per pixel used to generate the synthesis product

o the per-pixel date of the individual measurements or the start-end dates of the

period actually covered

o quality indicators, with explicit per-pixel identification of the cause of anomalous

parameter result

Accuracy requirements: wherever applicable the bio-geophysical parameters should meet

the internationally agreed accuracy standards laid down in document “Systematic

Observation Requirements for Satellite-Based Products for Climate”. Supplemental details

Page 12: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 12 of 36

to the satellite based component of the “Implementation Plan for the Global Observing

System for Climate in Support of the UNFCCC”. GCOS-#154, 2011”.

As the FCover is not considered as an Essential Climate Variable, no specific accuracy

requirement is specified in the above document.

The above requirements are indicative. They have to be adapted and clarified by the User Board of

the Global Land service.

Page 13: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 13 of 36

3 ALGORITHM

The algorithm has been defined by INRA (Institut National de REcherche Agronomique) in the

framework of the FP7/geoland2 project. It generates the fraction of vegetation cover (FCover)

associated with the leaf area index (LAI) and the Fraction of absorbed PAR (FAPAR). These

products are known as the GEOV1 products (Baret et al., 2013).

3.1 DEFINITION OF THE VARIABLE

FCover corresponds to the gap fraction for nadir direction. It is used to separate vegetation and soil

in energy balance processes, including temperature and evapotranspiration. It is computed from

the leaf area index and other canopy structural variables and does not depend on variables such

as the geometry of illumination as compared to FAPAR. For this reason, it is a very good candidate

for the replacement of classical vegetation indices for the monitoring of green vegetation. Because

of the linear relationship with radiometric signal, FCover will be only marginally scale dependent

(Weiss et al., 2000). Note that similarly to LAI and FAPAR, only the green elements will be

considered, either belonging both to the overstorey and understorey.

The Copernicus Global Land FCover Version 1 product is derived from CYCLOPES V3.1 (Baret et

al., 2007) products.

3.2 OUTLINE

The PROBA-V FCover Version 1 is derived from the SPOT/VGT-like Top of Atmosphere (TOA)

PROBA-V reflectances generated by the PROBA2VGT module [GIOGL1_ATBD_PROBA2VGT].

The retrieval methodology is described in Baret et al., (2013) and summarized in Figure 1. The

TOA reflectances are pre-processed to obtain normalized Top of Canopy (TOC) reflectances,

according to the methodology applied to get the PROBA-V TOC-R products

[GIOGL1_ATBD_TOCRef and GIOGL1_PUM_TOCRef]. Then, the method to get the FCover from

TOC reflectances is based upon a neural network.

Page 14: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 14 of 36

Figure 1: Main steps of the retrieval algorithm applied to PROBA-V VGT-like TOA reflectances to get

the biophysical variable (FCover).

3.3 THE RETRIEVAL METHODOLOGY

3.3.1 Input data

The input data of the FCover Version 1 processing line are the SPOT/VGT-like Top of Atmosphere

(TOA) PROBA-V reflectance products generated by the PROBA2VGT pre-processor

[GIOGL1_ATBD_PROBA2VGT].

3.3.1.1 PROBA-V instrument

Since May 2013, a new VEGETATION sensor is operational on board of a PROBA2 platform, also

known as PROBA-V. Although the platform is designed to provide continuity to the SPOT-

VEGETATION sensor, it has been designed to provide an increase in resolution from 1km to 1/3

km.

To achieve this increase in resolution, the PROBA-V instrument is composed of three cameras.

Each camera is composed of 2 sensors, the VNIR sensor and the SWIR sensor. The VNIR sensor

provides through 3 detectors from one strip each, the blue, red and NIR bands. The SWIR sensor

provides through a single detector, built up from three strips, the SWIR band. Each camera can be

individually switched on/off. The NIR band observes the Earth first, followed by the RED, BLUE,

and SWIR bands. As a result, an observation time difference of 12s exists between the NIR and

SWIR bands.

An image consists of one or more [image] bands, each consisting of a given number of columns

and rows (the pixels). A PROBA-V segment covers a swath of about 500 km in case of the centre

camera, and a swath of about 700 km in case of the left / right camera. With a Field Of View of

102.4°, it reaches a swath of 2295 km, flying at an altitude of 820 km with an equatorial pass-over

Page 15: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 15 of 36

time between 10:30 and 11:30. The current overpass time is 10:45. More details are given in the

PROBA-V Product User Manual [PROBAV-PUM].

To compose the 1 km pixels, a compositing operation of the different cameras is performed.

Different cameras (left, center, right) provide images for different segments. The segments are

then composited into 1 image by using a Maximum Value Composite (MVC) (Tarpley et al. 1984,

Holben, 1986) approach on NDVI to provide the end user products. Details on the PROBA-V 1km

data are given in the ATBD of the pre-processing module [GIOGL1_ATBD_PROBA2VGT].

The spectral characteristics of PROBA-V are close, but different, to those of SPOT/VGT-2 (Table

1).

Spectral band SPOT/VGT-2 PROBA-V

Centre (nm) Width (nm) Wavelength (nm)

B0 (blue) 450 40 447 – 493

B2 (red) 645 70 610 – 690

B3 (NIR) 835 110 770 – 893

SWIR 1665 170 1 570 – 1 650

Table 1: Spectral characteristics of VGT-2 and PROBA-V sensors

3.3.1.2 PROBA VGT-like data

In order to re-use the existing processing chain ingesting SPOT-VGT P segments, a module has

been developed to transform the daily PROBA-V TOA tiles data into files compatible with the daily

SPOT-VGT P segments. This module performs:

1. Radiometric correction to keep the reflectance values consistent with the SPOT-

VGT time-series.

2. Format transformation to enable the re-use of TOC-r operational workflow.

3. Integration of meteorological ancillary data in TOA products.

All three operations are described in detail in the ATBD [GIOGL1-ATBD-PROBA2VGT].

The resulting products are SPOT-VGT compatible products. Every output product is a zip file

containing a set of files as shown in Figure 1. In addition, the SPOT-VGT compatible product,

representing Top-Of-Atmosphere data, contains two ancillary meteorological data files,

representing the water vapour and the ozone, in the ZIP contents.

Page 16: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 16 of 36

Figure 2: PROBA VGT-like P segments contents

3.3.2 Pre-processing steps

Before the ingestion into the neural network, the TOA reflectances are first pre-processed to get

normalize TOC reflectances.

3.3.2.1 Radiometric calibration

As the PROBA-V VGT-like input TOA reflectances are compatible with the SPOT/VGT-2 TOA

reflectances, the same calibration coefficients are used.

3.3.2.2 Cloud screening

Hagolle et al. (2004) developed a cloud screening method based on surface reflectance in the blue

band (B0) computed using the SMAC code (Rahman and Dedieu, 1994) and corrected only from

Rayleigh scattering and gaseous absorption. The 1 km spatial resolution GLC2000 global land

cover classification (Bartholomé and Belward, 2005) was used to drive the threshold values used

in blue band (B0). For barren, sparse herbaceous, artificial surfaces and associated areas, mosaic

Page 17: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 17 of 36

cropland/shrub cover, and shrub cover classes were tuned: the pixel is flagged as cloudy if (TOC-

r(B0)> 0.24) and (TOC-NDVI < 0.05). For the other classes, the original algorithm of Hagolle et al.

(2004) is applied. To separate snow from clouds a snow detection algorithm was finally

implemented based on spectral variation of snow (Dozier, 1989). Thresholds associated to each of

the four bands were empirically tuned. Note that cloud detection is not active over salt lakes. Cloud

screening was validated at global scale during April 1999 by comparison to quasi-simultaneous

cloud fraction observations observed over 11500 synoptic nebulosity ground stations provided by

Météo-France (Berthelot et al., 2004). Comparison with FP7/CYCLOPES project cloud mask

shows good performances over Europe, with however some over detection above snow, and under

detection in tropical regions, due to the difficulty to identify semi transparent clouds.

Once cloudy pixels are identified, cloud shadows are first projected over the surface taking into

account view and sun directions, assuming that clouds are at a 3 km altitude. The cloud mask is

spatially extended over a 2 x 2 pixels radius to prevent small contamination of clear pixels in the

cloud vicinity. Pixels are then declared ‘suspect’ over a 3 x 3 radius centered on the cloud detected

pixels. Note that cloudy pixels are not spatially extended over snowy pixels. Each valid pixel (not

cloudy, not shadowed, not saturated or not out of orbit) may thus be declared either clear (all the

bands valid), suspect (in the vicinity of cloudy/shadowed pixels), or snowy (with all the bands valid

or with just B0 and/or B2 (red) bands invalid).

3.3.2.3 Atmospheric correction

Atmospheric effects are corrected using the SMAC code (Rahman and Dedieu, 1994) for the four

spectral bands. Input atmospheric characteristics are derived from currently available products:

NCEP Meteorological data for pressure and water vapour, and TOMS/TOVS for ozone

(ftp://oceans.gsfc.nasa.gov). The aerosol optical thickness was derived from the minimum of

monthly values of MODIS aerosol products (MOD08_M3) (Kaufman et al., 1997). A climatology of

these values was built over the period covering March 2000 up to December 2006. However,

MODIS aerosol products present missing values and unexpected large AOT550 values

(AOT550>0.6) that were removed, mostly corresponding to transition between deserts and

vegetated areas. The gaps were thus filled with the simple latitudinal gradient proposed by

Berthelot and Dedieu (Berthelot and Dedieu, 1997). A spatial smoothing procedure is finally

applied to minimize the possible discontinuities between 1°× 1° pixels and to reach the l 1/112°

spatial resolution.

3.3.2.4 Normalization and compositing

The linear reflectance model of Roujean et al. (1992) is inverted to normalize the surface

reflectances and remove the bidirectional effects due to changes in solar and viewing angular

configurations during the synthesis period of 30 days. A Gaussian weighting function is applied to

the observations of the synthesis period: the maximum weight is put on the last observation of the

Page 18: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 18 of 36

period since these products are near real time. This model describes the bidirectional reflectance

as a linear combination of three geometrical kernels (fi), weighted by three coefficients. The

compositing method is improved using prior information (Hagolle et al., 2004): a confidence interval

for the coefficients of the Roujean model is used to constrain the least-squares fit. The constrained

BRDF model is finally used to detect iteratively and discard dates with residual cloud contamination

or atmospheric effects. The current algorithm is applied at a 10-day frequency on surface

reflectances acquired during a moving temporal window of 30 days to generate directional

parameters for each spectral band. Normalized reflectances are then computed with a solar zenith

angle corresponding to the median value of observed during the compositing period and a viewing

observation at nadir.

More details can be found in the ATBD [GIOGL1_ATBD_TOCRef] or in the PUM

[GIOGL1_PUM_TOCRef] of the PROBA-V TOC-Ref product.

3.3.3 Biophysical retrieval methodology

3.3.3.1 Principle of the algorithm

The biophysical algorithm (common to LAI, FAPAR and FCover Version 1) is based on already

existing ones to capitalize on the efforts accomplished and get a larger consensus from the user

community. Figure 3 shows the several steps used for each of these variables. Following the

published literature on products validation (Garrigues et al., 2008; Weiss et al., 2007), the best

performing existing products were selected and combined to take advantage of their specific

performances while limiting the situations where products show deficiencies. The selected

products are re-projected onto the SPOT/VGT plate-carrée 1/112° grid, smoothed through time

and interpolated at the 10 days frequency. Then the products are combined and eventually scaled

to compute the fused product that is expected to provide globally the ‘best’ performances. The

fused products are generated for few years over the BELMANIP2 set of sites that is supposed to

represent the possible range of surface types and conditions over the Earth (Baret et al., 2006).

Neural networks are then calibrated over this set of sites to relate the fused products to the

corresponding SPOT/VGT normalized TOC reflectances from the FP7/CYCLOPES project (Baret

et al., 2007).

Page 19: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 19 of 36

Figure 3: Flow chart showing the principle of the biophysical retrieval methodology.

3.3.3.2 Selection of the products used to generate the “fused products”

Only one global FCover product was available: the CYCLOPES dataset (Table 4)

Products FCover Temporal frequency

(days)

Time period

Projection Reference

CYCLOPES V3 10 1999-2007 Lat-lon (Baret et al., 2007)

Table 2: The global FCover products available

Existing Products

Selection

Product 1 Product 3

Fusion

FusedProduct

CYCLOPESL3a

Neural NetCalibration NNT Coefs

CYCLOPESL3a

Neural NetCalibration

GEOV1Product

AlgorithmCalibration

Application

SmoothingInterpolation

SmoothingInterpolation

Scaling

CYCLOPES Normalized TOC

reflectancesPROBA-V TOC-

Ref product

Neural Net

PROBA-V FCoverV1 product

Page 20: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 20 of 36

3.3.3.3 Product fusion

Since only the CYCLOPES products were available globally, no fusion with other products was

possible. However, several evaluations have shown that CYCLOPES FCover products were

suffering from a significant systematic underestimation (Verger, 2008). This was corrected for by

applying a scaling factor to the CYCLOPES V3.1 products ( ). This factor (

was

corresponding to the inverse of the FCover value for the 99% cumulated frequency (Figure 4) that

should be expected to be very close to 1.0 since it should correspond to very dense canopies:

Where is the value that will be used for calibrating the neural network.

Figure 4: Cumulated frequency computed over the BELMANIP2 sites for FCover during the period

2003-2005 for the original CYCLOPES V3.1 products (dashed line) and after the „best available

product“ obtained by scaling CYCLOPES V3.1.

3.3.3.4 Calibration of the neural network

The scaled FCover product was used to calibrate neural network using as inputs the normalized

TOC reflectance in the three SPOT/VGT bands (red, near infrared and short wave infrared) along

with the cosine of the sun zenith angle corresponding to 10:00 local time (§ 3.2.2.4) (Figure 5).

Page 21: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 21 of 36

Figure 5: Neural network architecture used to estimate FCover from TOC reflectances and

illumination geometry.

The learning data set available was split into a set of cases containing 90% of the available data

while the 10% remaining part was used for hyper-specialization and theoretical evaluation (test

data set).

The theoretical performances of the network were evaluated over the test data set. Results show

that the training of the network was quite efficient, with relatively small RMSE value. The algorithm

is unbiased as expected (Figure 6).

Figure 6: Theoretical performances of FCover neural network (presented as boxplots with the solid

line corresponding to the median and the gray levels containing respectively 50%, 90% and 99% of

the population).

3.3.3.5 Quality flags and uncertainties

In addition to the quality flags used to describe TOC-Ref products, additional information is also

provided for FCover product:

Input out of range. This represents the consistency of the measured PROBA-V input

reflectances with those used in the training data base. A flag is raised when observations

are outside the training definition domain.

B2

B3

SWIR

Norm

Norm

Norm

Norm

Norm S

S

S

S

S

L Var

cos(s)

Page 22: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 22 of 36

Output out of range. If the output is outside the expected range of the biophysical

variable, a flag is raised.

Estimated uncertainties: This represents the expected error expressed in RMSE between

the estimated and the actual biophysical values. As a first approximation, this can be

derived from the theoretical performances of the algorithm as evaluated over an

independent data set. A specific network is trained to relate the estimated uncertainties to

the input reflectance and observation geometry values.

3.4 LIMITATIONS OF THE PRODUCT

The main limitations of the product are related to:

The temporal resolution: the products are provided at a 10 days sampling interval.

However, they are built from 30 days composite of PROBA-V observations. The original

products used to calibrate the algorithm were also based either on 30 days temporal

resolution (CYCLOPES) or 20 days (MODIS) 8 days within ±10 days.

Presence of water bodies or snow. The performances of the products in presence of

water bodies or snow (either pixels mixed with water bodies or snow, or water bodies or

snow representing a significant fraction of the background) are expected to be degraded.

This may be detected using the ‘input out of range’ flag.

Vegetated areas in the desert. Some known vegetated areas like oasis or irrigated crops

can appear like permanent “no data” pixels.

Time lag: Considering the compositing approach, and the delay necessary for processing,

there is, at final, a time lag of about 2 weeks between the representative date of the

products and its availability. This can be a limitation for NRT monitoring applications.

Validity domains are defined for the inputs (normalized spectral surface reflectances) of the neural

networks. If inputs are outside the validity domain, a flag is raised on the normalized reflectances.

Despite of that, the FCover is assessed by the neural network, and kept if it belongs to its physical

domain.

The neural network is calibrated with the FCover best estimates among existing products, and the

RMSE between predicted and actual Fcover values is about 0.037.

Page 23: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 23 of 36

4 PRODUCT DESCRIPTION

The PROBA-V FCover V1 product follows the following naming standard:

g2_BIOPAR_FCOVER_<YYYYMMDDHHMM>_<AREA>_<SENSOR>_V<Major.Minor>

where

<YYYYMMDDHHMM> gives the temporal location of the file. YYYY, MM, DD, HH, and MM

denote the year, the month, the day, the hour, and the minutes, respectively.

<AREA> gives the spatial coverage of the file. In our case, <AREA> is HxxVyy, the name of

the 10°x10° tile or the continent short name

<SENSOR> gives the name of the sensor used to retrieve the product, so PROBAV

<Major.Minor> gives the version number of the product. “Major” increases when the

algorithm is updated. “Minor” increases when bugs are fixed or when processing lines are

updated (metadata, color quicklook, etc...).

4.1 FILE FORMAT

The FCover products are distributed in zip archives. The archive contains following files:

multi-band HDF-5 format containing the following layers:

o FCOVER: Fraction of Vegetation Cover

o FCOVER-ERR: uncertainty on the variable

o FCOVER-QFLAG: quality flag

o NMOD: number of valid observations during the synthesis period

o LMK: land-sea mask based upon the GLC-2000 land cover map

a xml file containing the metadata conform to INSPIRE2.1. An xsl file allows displaying a

friendly view of the metadata file.

A quicklook in a coloured geo-tiff format. The quicklook sub-sampled to 25% in both

horizontal and vertical direction from the FCOVER layer.

A text file containing the copyright of the product.

Each file corresponds to a tile of 10° x 10°. A dedicated typology has been chosen to locate easily

each tile when reading its name. The globe is divided in 36 tiles in longitude and 18 in latitude

according to Figure 7. Only the green tiles, covering land, are generated and hence available. In

Page 24: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 24 of 36

winter season, the northern hemisphere has less observations, hence the number of available tiles

could be less. The products are also provided by continents (grouping of a set of tiles) as defined

in Figure 7, and named as in Table 3. The 6 colored rectangles, covering most land of our

continents, represent the available continents.

Figure 7: Regions of FCover product.

The Africa and South-America continent (AFRI and SOAM continental tiles in Table 3) are also

distributed through EUMETCast (see Section 4.3.5 for access information).

Short name Continent

AFRI Africa: 30°W – 60°E, 40°N – 40°S

ASIA Asia: 60°E – 180°E, 80°N – 0°N

EURO Europe: 30°W – 60°E, 80°N – 30°N

NOAM North America: 170°W – 50°W, 80°N – 10°N

OCEA Oceania: 100°E – 180°E, 0°S – 50°S

SOAM South America: 110°W – 30°W, 20°N – 60°S

Table 3: Definition of the continental tiles.

Page 25: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 25 of 36

4.2 PRODUCT CONTENT

4.2.1 Data File

The physical range of FCover and uncertainty of FCover are given in Table 4. The physical values

are retrieved by:

PhyVal = DN / ScalingFactor + Offset

where the scaling factor and the offset are given in the tables.

As an example, FCover is a byte for which a data value of 150 equals a physical value of 0.6.

FCover FCOVER-Err

Minimum physical value 0 0

Maximum physical value 1.0 0.2

Maximum DN value 250 40

Invalid 255 255

Out of Range by value superior to the max physical value 253 253

Out of Range by value inferior to the min physical value 254 254

Scaling factor 250 200

Offset 0 0

Table 4: Range of values and scaling factors of FCover and FCover uncertainty.

The parameter NMOD corresponds to the number of measurements used to process the FCover,

and hence reflects the number of valid observations during the synthesis period. The physical

values are equal to the digital number within the range [0, 60].

The parameter LMK is a land cover map based upon the GLC2000 product. It integrates the class

“salt lakes” at global scale. The digital numbers are the sum of 200 and of the class number of the

GCL2000 map.

The quality flag (QFLAG) must be read bit per bit (Table 5). Its description is given in the table

below: bit 15 and bit 16 have no meaning. The QFLAG is coded on 2 bytes with an invalid value

equal to 65535.

Page 26: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 26 of 36

Bit = 0 Bit = 1

Bit 1: Land/Sea Land Sea

Bit 2: Snow status Clear Snow

Bit 3: Suspect No suspect Suspect

Bit 4: Aerosol status Pure Mixed

Bit 5: Aerosol source Modis Climato

Bit 6: Input status OK Out of range or invalid

Bit 7: LAI status OK Out of range or invalid

Bit 8: fAPAR status OK Out of range or invalid

Bit 9: fCover status OK Out of range or invalid

Bit 10: B2 saturation status OK Saturated

Bit 11: B3 saturation status OK Saturated

Bit 12: Filtering status Not filtered Filtered

Bit 13: Gap filling status Not filled Filled

Bit 14: NDVI status OK Invalid

Table 5: Description of the quality flag of FCover.

The multi-band HDF-5 file contains a number of HDF-5 attributes on the file-level, see Table 6 and

on the layer-level, see Table 7.

Attribute name Attribute description

Archive_facility Name of archiving center

Centre Name of processing/dissemination center

Ellipsoid_name Name of ellipsoid system

Geodate_name Name of date system

LAT Lattitude (top left)

LONG Longitude (top left)

Instrument_ID Sensor name

Overall_quality_flag Quality indicator

Pixel_size Size of a single pixel (in square)

Product_time Time of product creation

Product_algorithm_version Version of algorithm during processing

Projection_name Name of projection system

Region_name Area name

Satellite Platform name

Table 6: Description of HDF-5 file attributes

Page 27: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 27 of 36

Attribute name Attribute description

Class Type of layer (DATA)

Nb_bytes Data type (e.g. uint8, uint16, …)

Order_bytes Order data type (1= intel)

Product Layer name

Missing_value Data number representing missing value

Offset Offset factor to retrieve physical value

Scaling_factor Scale factor to retrieve physical value

Table 7: Description of HDF-5 layer attributes

4.2.2 Quicklook

The quicklook is a geo-referenced tiff file. The spatial resolution is sub-sampled, using nearest

neighbour resampling, to 25% in both directions, hence a quicklook is 1/16th of the size of the data

layer.

The quicklook is coded in 1 byte, using the same offset and scale parameters as the data layer.

The quicklook is provided with an embedded color table (Table 8).

Table 8: Color coding for quicklook FCover images

4.3 PRODUCT CHARACTERISTICS

4.3.1 Projection and Grid Information

The product is displayed in a regular latitude/longitude grid (plate carrée) with the ellipsoïd WGS

1984 (Terrestrial radius=6378km). The resolution of the grid is 1/112° (approx. 1km at the equator).

The reference is the centre of the pixel. It means that the longitude of the upper left corner of the

pixel is (pixel_longitude – angular_resolution/2.)

Each tile covers an area of 10°x10°, while the continental tiles cover a multiple of these 10°x10°

tiles as defined in Table 3.

Page 28: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 28 of 36

Figure 8: The cut-out of tiles

Starting from the upper left center pixel and taking 10° or 1120 pixels into account, the final tile is

an array of 1121 lines and 1121 columns taking the bottom/right upper/left half pixel into account

(Figure 8). As such there is an overlap between the tiles. For example, the last column of tile

H18V4 is the first column of the tile H19V4, and the last line of H18V4 is the first line of the tile

H18V5.

Note that the same reasoning holds also for the continental tiles.

4.3.2 Spatial information

As shown in Figure 7, the FCover product is provided, hence from longitude 180°E to 180°W and

latitude 75°N to 60°S. Any areas cover latitude above 75°N and below 60°S is filled with no-values

(255 for Byte and 65535 for Uint16).

4.3.3 Temporal Information

The FCover products are a 30-days composite, updated every 10 days using a moving window.

The temporal information “YYYYMMDDHHMM” in the filename is the “actual” date corresponding

10° = 1120 pixels

1121 pixels = 10.0089°

10° = 1120 pixels

1121 pixels = 10.0089°

Page 29: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 29 of 36

to the product value: the sum of weights of the weighting function (see paragraph 3.3.2.4) is the

same on each side of this date.

For example, the file g2_BIOPAR_FCOVER_200408130000_H17V04_VGT_V1.1.h5 corresponds

to the synthesis period from 26/07/2004 to 25/08/2004. The day 13/08/2004 is the day where the

sum of weights of the weighting function is the same on each side of this date.

Important note: The “actual” date is the date to take in consideration for any analysis.

The temporal information is detailed in the metadata (xml file) by 2 fields "EX_TemporalExtent :

beginPosition" and "EX_TemporalExtent : endPosition" as “YYYY-MM-DDTHH:MM::SS” giving the

beginning and the end of the 30-days time period, respectively. The endPosition of the 30-days

time period is always set to the 05th, 15th or 25th day of the month.

4.3.4 Data Policies

Any use of the original PROBA-V FCover products implies the obligation to include in any

publication or communication using these products the following citation:

"The product was generated by the land service of Copernicus, the Earth Observation programme

of the European Commission. The research leading to the current version of the product has

received funding from various European Commission Research and Technical Development

programmes. The product is based on PROBA-V 1km data ((c) ESA and distributed by VITO).”

The user accepts to inform the production and distribution center of their publications through the

following address: [email protected].

4.3.5 Access and Contacts

The PROBA-V FCover V1 product is available through the Copernicus Global Land service

website at the address: http://land.copernicus.eu/global/products/fcover

Accountable contact: European Commission Directorate – General Joint Research Center

Email address: [email protected]

Scientific contact: Institut National de la Recherche Agronomique (INRA), France

Email address: [email protected]

Production and distribution contact: VITO

Email address: [email protected]

Page 30: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 30 of 36

5 VALIDATION RESULTS

5.1 VALIDATION PROCEDURE

The quality assessment of the PROBA-V GEOV1 products (LAI, FAPAR, FCover) during the first

year of data was conducted following the validation procedure described in the Copernicus Global

Land Service Validation Plan [GIOGL1_SVP] in agreement with the CEOS LPV best practices for

validation of LAI products. This study was focused on the consistency of PROBA-V and

SPOT/VGT GEOV1 products for the 6-months overlap period (mid November 2013 – mid May

2014). Moreover, MODIS C5 products were considered for inter-comparisons during the overlap

period and for the whole 2014. The following main criteria were assessed: completeness, spatial

consistency, temporal consistency, smoothness, statistical assessment of discrepancies with

similar products, and accuracy.

5.2 MAIN RESULTS

Product completeness

PROBA-V GEOV1 products shows a slight higher fraction of missing values (around 5% in overall)

than the SPOT/VGT GEOV1 products. The spatial and temporal distribution of the missing values

was consistent with that of SPOT/VGT, except over northern latitudes and equatorial areas.

Missing values for Needle Leaf Forest (NLF) reach up to 80% in the European wintertime. The

larger discrepancies were found over the Evergreen Broadleaf Forest (EBF) type, showing

PROBA-V fraction of missing values up to 60%. Per continental region, the larger discrepancies

are observed in South America, where the contribution of EBF forest is very important.

The temporal length of missing values was very consistent between PROBA-V and SPOT/VGT

GEOV1 products, with MODIS C5 products showing shorter length of gaps.

Spatial consistency

PROBA-V GEOV1 maps and difference maps with SPOT/VGT GEOV1 shows a very good

consistency, with most of the values within the optimal levels of 0.5 for LAI and 0.05 for FAPAR

and FCOVER. The analysis of residuals over global data confirms the good spatial consistency

between PROBA-V and SPOT/VGT with more than 80% of data within optimal levels (up to 92%

for LAI). Nevertheless, a slight overestimation of PROBA-V FCover was detected mainly over EBF

biomes and the South America, which shows the highest vegetation cover values during this

period.

Differences between PROBA-V and MODIS show also large fraction of pixels within the optimal

levels for the LAI (>72%). Larger discrepancies were found in areas dominated by EBF or

Deciduous Broadleaf Forest (DBF) in South America, Africa and Asia. However, for the FAPAR,

Page 31: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 31 of 36

both products shows large spatial inconsistencies around the world for the whole 2014 year as

previously reported between SPOT/VGT and MODIS C5. MODIS provides larger FAPAR values

over semi-arid regions whereas PROBA-V provides larger FAPAR values over more vegetated

regions. Better agreement is found for densest equatorial forests.

The spatial variability in homogeneous zones shows very similar results between PROBA-V and

SPOT/VGT over three areas (forest, shrublands, herbaceous) studied. Very good repeatability

(spatial precision) is achieved. MODIS shows similar results for LAI but very low spatial variability

for FAPAR. Over a selected heterogeneous area, showing an important gradient of vegetation,

PROBA-V GEOV1 showed very good dynamic range, good consistency with SPOT/VGT and the

larger spatial variability.

Temporal consistency

PROBA-V temporal variations were found very consistent with that of SPOT/VGT products for the

different biomes, as well as with MODIS products. The cross-correlation between PROBA-V and

SPOT/VGT temporal variations was higher than 0.9 typically in more than 70% of the sites for the

different biomes with the exception of EBF types where much lower correlations were observed.

The cross-correlation with MODIS products was also satisfactory, with a range of 50-70% of the

samples with correlations better than 0.9.

GEOV1 temporal variations over two Deciduous Tropical Forest Enviro-Net sites

(http://www.enviro-net.org) were found consistent with ground data. However, PROBA-V FCover

shows unreliable seasonal patterns over some desertic sites (e.g. Sahara) as it was previously

reported for SPOT/VGT GEOV1 products.

Intra-Annual Precision (smoothness)

PROBA-V GEOV1 products shows a good intra-annual precision, almost identical to that of

SPOT/VGT equivalent product, and much better than MODIS.

Statistical Analysis of discrepancies

Scatter-plots and uncertainty statistics between GEOV1 (PROBA-V vs SPOT/VGT) products over

BELMANIP 2.1 sites show the optimal consistency between both products, with overall correlations

higher than 0.96, no mean bias, and overall discrepancies (RMSE=0.3, 0.03, 0.04 for LAI, FAPAR

and FCover respectively) lower than the GCOS requirements on accuracy for LAI and FAPAR.

For the LAI product, very little systematic differences were found, with some bias (higher values for

PROBA-V) for LAI values higher than 6. For the FAPAR, very good consistency was found with

median values very close to zero regardless the interval of FAPAR considered. For the FCover,

systematic differences between PROBA-V and SPOT/VGT are larger, with a slight but systematic

overestimation of PROBA-V for FCover values larger than 0.4. The different behavior of the LAI,

Page 32: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 32 of 36

FAPAR and FCover can be partly explained due to the different impact of the input variations as

each variable uses a different neuronal network.

The systematic and overall discrepancies between PROBA-V GEOV1 and MODIS C5 products are

larger and show a larger dynamic as a function of the GEOV1 product value. For the LAI, large

scattering was found for LAI between 3 and 5, and the bias can be larger than 2. For the FAPAR, a

clear tendency was observed in the systematic differences. MODIS provides larger values for the

lower FAPAR values (<0.3), as MODIS overestimates FAPAR for very low values (Camacho et al.,

2013). However, GEOV1 provides larger values than MODIS mainly for FAPAR larger than 0.5.

The discrepancies between PROBA-V GEOV1 and MODIS C5 are similar to that found between

SPOT/VGT and MODIS C5:

Analysis per biomes:

For the LAI, very good agreement between PROBA-V and SPOT/VGT was found for all biomes,

with no systematic discrepancies. For the FAPAR, the agreement is optimal for all biomes with no

systematic deviations. For the FCover, no bias was found for non-forest types, whereas for forest

biomes PROBA-V shows slightly higher values than SPOT/VGT. Larger discrepancies (mean bias

of 0.04) and overestimation was found in EBF type, where maximum values (FCover=1) were

generally observed.

Scatter-plots between PROBA-V and MODIS LAI products shows a large scattering for EBF type,

and higher LAI values for DBF, NLF and Cultivated sites were obtained with PROBA-V. For

Herbaceous and Shrubs/Sparse, the agreement is quite good with low bias. Regarding the

FAPAR, an asymmetric behavior in Cultivated, Herbaceous and DBF sites was observed with a

tendency of PROBA-V to overestimate the MODIS FAPAR estimates for large values, whereas

MODIS products overestimates PROBA-V for low values. This tendency was observed for all

biomes types except in EBF and NLF sites with larger FAPAR values.

Analysis per continental region

PROBA-V and SPOT/VGT products show very good agreement for all the continental regions with

almost no bias for all the regions and products. Only for the FCover in South America, a large

systematic overestimation was found as this region presents the highest vegetation values.

The comparison between PROBA-V and MODIS products shows much larger residuals and RMSE

values for all the regions. The LAI larger discrepancies were also found in South America, as large

discrepancies were found for EBF and Cropland types which are dominant classes in this

continent. For the FAPAR, a positive difference was observed in North America, South America

and Europe whereas a negative tendency was found in Oceania.

The specific analysis over Africa shows in general larger discrepancies among the PROBA-V

GEOV1, MODISC5 and LSA SAF products for the three variables. The discrepancies between

PROBA-V and MODIS or LSA SAF LAI or FAPAR are however better than discrepancies between

Page 33: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 33 of 36

the two operational satellite products used as references. The large discrepancies observed for the

FCover are partly explained by a systematic bias in the LSA SAF products for low vegetation

values.

Accuracy Assessment

The accuracy of PROBA-V GEOV1 LAI product was very close to GCOS requirement using limited

concomitant data (RMSE=0.54) or using additional non-concomitant references (RMSE=0.51). For

the FAPAR the accuracy is also quite good (0.11 for all data), but a slight overestimation was

observed mainly as compared to concomitant data (bias=0.09). The FCover shows the worst

performance, with a systematic positive bias observed for mainly for forest and cropland sites (up

to 0.15 for concomitant data) and overall error of 0.14. PROBA-V shows similar performances

than MODIS for both LAI and FAPAR products.

Concluding remark

The validation results of the PROBA-V GEOV1 products over one year (2014) period shows very

good spatial and consistency with the SPOT/VGT GEOV1 products for the overlap period.

However, a positive bias as compared to SPOT/VGT has been detected for the FCover mainly for

values larger than 0.5. This bias seems to be confirmed by the limited ground observations

available. All the criteria evaluated including accuracy assessment shows positive results (Table

9). The main drawback of the product is the completeness which is slightly lower than in

SPOT/VGT GEOV1 products.

Table 9: Summary of Product Evaluation (PROBA-V GEOV1). The plus (minus) symbol means that

the product has a good (poor) performance according to this criterion.

QA

Criteria Performance Comments

Product

Completeness -

Main limitations over Northern latitudes in wintertime and

Equatorial areas.

Spatial

Consistency +

Optimal spatial consistency between PROBA-V and SPOT/VGT

GEOV1 products. Most of the differences between both lies

within 0.5 for LAI and 0.05 for FAPAR/FCOVER

Good repeatability over well known homogenous areas (Dense

Forest and Shrublands).

Good variability for known spatial gradients.

Larger discrepancies between PROBA-V and MODIS over EBF

and DBF for LAI and globally for FAPAR (similar than

SPOT/VGT and MODIS).

Temporal

Consistency +

Good consistency of PROBA-V GEOV1 temporal variations, as

compared to SPOT/VGT GEOV1 and MODIS C5. Cross-

Page 34: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 34 of 36

correlations between PROBA-V and SPOT/VGT GEOV1 (higher

than 0.9 in more than 70% of the sites for most of biomes

except in EBF).

Realism of temporal profiles over deciduous forest. Unreliable

seasonality of the FCover over some desertic sites.

Intra-Annual

Precision + Very smooth temporal profiles

Statistical

Analysis of

Discrepancies

+

Optimal consistency between PROBA-V and SPOT/VGT

GEOV1 (RMSE=0.3, 0.03, 0.04 for LAI, FAPAR and FCover

over BELMANIP2.1

Larger discrepancies between both GEOV1 products and

MODIS C5.

FCover PROBA-V shows higher values than SPOT/VGT mainly

for forest sites.

Accuracy ± Good accuracy with limited ground dataset for LAI and FAPAR.

Positive bias for FCover.

The details of the analysis are reported in a Quality Assessment Report [GIOGL1_QAR_PROBA-

V-GEOV1] available on http://land.copernicus.eu/global/products/FCover.

Furthermore, the quality assessment of the PROBA-V normalized TOC reflectances (TOC-r

products), which are the input of neural networks to calculate the GEOV1 LAI, FAPAR, FCover

variables, is in progress. However, a preliminary analysis has showed systematic positive

differences between PROBA-V and SPOT/VGT TOC-r in near infrared and SWIR spectral bands

(PROBA-V TOC-r slightly higher than SPOT/VGT TOC-r). The reasons of such differences are

under investigation. Once fixed, a re-processing action of TOC-r and, then, of GEOV1 LAI, FAPAR,

FCover will be performed and a new quality assessment exercise will be carried out.

Page 35: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 35 of 36

6 REFERENCES

Baret, F., M. Weiss, R. Lacaze, F. Camacho, H. Makhmara, P. Pacholcyzk, B. Smets, 2013.

GEOV1: LAI and FAPAR essential climate variables and FCOVER global time series

capitalizing over existing products. Part1: Principles of development and production.

Remote Sensing of Environment 137: 299–309.

Baret, F., Hagolle, O., Geiger, B., Bicheron, P., Miras, B., Huc, M., Berthelot, B., Weiss, M.,

Samain, O., Roujean, J.L. et al., 2007. LAI, fAPAR and fCover CYCLOPES global products

derived from VEGETATION. Part 1: Principles of the algorithm. Remote Sensing of

Environment, 110: 275-286.

Baret, F., Morissette, J., Fernandes, R., Champeaux, J.L., Myneni, R., Chen, J., Plummer, S.,

Weiss, M., Bacour, C., Garrigue, S. et al., 2006. « Evaluation of the representativeness of

networks of sites for the global validation and inter-comparison of land biophysical products.

Proposition of the CEOS-BELMANIP”. IEEE Transactions on Geoscience and Remote

Sensing, 44(7: special issue on global land product validation): 1794-1803.

Baret, F. and Guyot, G., 1991. Potentials and limits of vegetation indices for LAI and APAR

assessment. Remote Sensing of the Environment, 35(2-3): 161-173.

Bartholomé, E. and Belward, A.S., 2005. GLC2000: a new approach to global land cover mapping

from Earth Observation data. International Journal of Remote Sensing, 26(9): 1959 - 1977.

Berthelot, B. and Dedieu, G., 1997. Correction of atmospheric effects for vegetation data. In: G.

Guyot and T. Phulpin (Editors), Physical measurements and signatures in Remote Sensing.

Balkema, Courchevel (France), pp. 19-25.

Berthelot, B., Quesney, A., Hagolle, O., Cabot, F. and Bruniquel, V., 2004. The CYCLOPES

project: cloud screening and atmospheric correction, VEGETATION users conference,

Antwerp (Belgium).

Camacho, F., Cernicharo, J., Lacaze, R., Baret, F., and Weiss, M. (2013). GEOV1: LAI, FAPAR

Essential Climate Variables and FCover global time series capitalizing over existing products.

Part 2: Validation and intercomparison with reference products. Remote Sensing of

Environment 137: 310–329.

Dozier, J., 1989. Spectral signature of alpine snow cover from Landsat 5 TM. Remote Sensing of

Environment, 28: 9 - 22.

Garrigues, S., Allard, D., Baret, F. and Weiss, M., 2006. Influence landscape spatial heterogeneity

on the non-linear estimation of leaf area index from moderate spatial resolution remote

sensing data. Remote Sensing of Enviroment, 105: 286-298.

Page 36: Gio Global Land Component - Lot I - ICDC Datenzentrum · 2015. 9. 28. · GIO-GL Lot 1, GMES Initial Operations Date Issued: 23.03.2015 Issue: I2.20 Gio Global Land Component - Lot

GIO-GL Lot 1, GMES Initial Operations

Date Issued: 23.03.2015

Issue: I2.20

Document-No. GIOGL1_PUM_FCOVERV1 © GIO-GL Lot1 consortium

Issue: I2.20 Date: 23.03.2015 Page: 36 of 36

Garrigues, S., Lacaze, R., Baret, F., Morisette, J., Weiss, M., Nickeson, J., Fernandes, R.,

Plummer, S., Shabanov, N.V., Myneni, R. et al., 2008. “Validation and Intercomparison of

Global Leaf Area Index Products Derived From Remote Sensing Data”. JGR, 113(G02028).

GCOS-154 (2011). Systematic Observation Requirements for Satellite-based Products for Climate

Supplemental details to the satellite-based component of the Implementation Plan for the

Global Observing System for Climate in Support of the UNFCCC - 2011 Update, WMO,

Geneva, Switzerland.

Hagolle O., A. Lobo, P. Maisongrande, F. Cabot, B. Duchemin, A. de Pereyra, 2004, “Quality

assessment and improvement of temporally composited products of remotely sensed

imagery by combination of VEGETATION 1 and 2 images”, RSE, 94, 172-186.

Holben, B. N. (1986). Characteristics of maximum-value composite images from temporal AVHRR

data, International Journal of Remote Sensing, 7, 1417-1434.

Kaufman, Y.J., Tanré, D., Remer, L.A., Vermote, E.F., Chu, A. and Holben, B.N., 1997.

Operational remote sensing of tropospheric aerosol over land from EOS moderate

resolution imaging spectroradiometer. Journal of Geophysical Research, 102(D14): 17051-

17067.

Rahman, H. and Dedieu, G., 1994. SMAC: a simplified method for the atmospheric correction of

satellite measurements in the solar spectrum. International Journal of Remote Sensing,

15(1): 123-143.

Roujean, J.-L., M. Leroy, and P.Y. Deschamps, 1992, “A bidirectional reflectance model of the

Earth’s surface for the correction of remote sensing data”, JGR, vol. 97, 20,455-20,468.

Tarpley, J. P., Schneider, S. R., & Money, R. L. (1984). Global vegetation indices from NOAA-7

meteorological satellite. Journal of Climate and Applied Meteorology, 23, 491 – 494.

Verger, A., 2008. Analisi comparativa d'algorismes operacionals d'estimacio de parametres

biofisics de la coberta vegetal amb teledeteccio, Universitat de Valencia, Valencia (Spain),

277 pp.

Weiss M., F. Baret, R. Myneni, A. Pragnère, and Y. Knyazikhin, 2000, “Investigation of a model

inversion technique for the estimation of crop characteristics from spectral and directional

reflectance data”, Agronomie, 20, 3-22.

Weiss, M., Baret, F., Garrigues, S., Lacaze, R. and Bicheron, P., 2007. “LAI, fAPAR and fCover

CYCLOPES global products derived from VEGETATION. part 2: Validation and

comparison with MODIS Collection 4 products”. RSE, 110: 317-331.