A Land Cover Map of Eurasia’s Boreal Ecosystems S. BARTALEV, A. S. BELWARD Institute for Environment and Sustainability, EC Joint Research Centre, Italy.

A Land Cover Map of Eurasia’s Boreal Ecosystems

S. BARTALEV, A. S. BELWARDInstitute for Environment and Sustainability, EC Joint Research Centre, Italy

D. ERCHOVCenter of Forest Ecology and Productivity, Russia

Glo

bal

Lan

d C

ove

r 20

00

SPOT 4 - VEGETATION Data

Type of data product

Standard S10 products including:

Spectral channels

NDVI

Angular information

Status map

Geographic window : 420N - 750N

and 50E -1800E

Time window : 3d decade of March 1999 – 1st decade of November 1999 (23 of S10 products)

STEP 1: Image pre-processing and generation of advanced data products

STEP2: Image classification

SPOT4-VGT S10

data

Contaminated pixels and

snow cover detection

Generated masks

Generation of the advanced data products

Seasonal mosaics

Wetness Index

Anisotropy Index

Wave-Likeness

Index

Snow Cover

ISODATA clustering of

seasonal mosaics

Initial labelling of

clusters

Spectral-temporal clusters

map

Semantic clusters

map

Decomposing of ambiguous

semantic clusters

Mono-semantic

clusters map

Land Cover Map

Merging of semantic clusters into

thematic classes

Land Cover Mapping Method

GIS Database

(topographic and thematic maps, DEM, forest inventory statistics and etc)

Derived Auxiliary Products

From Hall et al., 1998: "Algorithm Theoretical Basis Document (ATBD) for the MODIS Snow-, Lake Ice- and Sea Ice-Mapping Algorithms. Version 4.0"

Normalised Difference Snow Index

RR

RRNDSI

blu

swirblu

Detection of the contaminated pixels

Step 1: Detection of the pixels utterly contaminated by snow and

clouds with pre-specified thresholds

pixel with co-ordinates at fixed decade of observation t*

L

C

S

C

C

C set of pixels with presence of snow or/and ice

set of pixels with presence of clouds

set of pixels without presence of snow/ice or clouds

),( *tP Θ Θ

CSp CCC 1

Detection of the contaminated pixels

Steps 2J: Detection of the defective detectors and “slightly”

contaminated by snow/clouds pixels with adaptive thresholds

derived from time series of data

jpC s e t o f c o n t a m i n a t e d p i x e l s a t s t e p j

)( *ΘjswirM t h e m e a n o f ),( tR swir

*Θ

)( *Θjswir s t a n d a r d d e v i a t i o n o f ),( tR swir

*Θ

*Θ f i x e d c o - o r d i n a t e s

jp

jp CCC 1j

p

Seasonal Mosaics

spring

summer

autumn

Wave-Likeness Index (WLI)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

t1 t2 t3 t4 t5 t6 t7 t8 t9 … … … … … tn-1 tn

time of observation

ND

VI

(NDVI e, t e)

(NDVI max, t max)

(NDVI b, t b)

a

b -

1

b -

1

d

d

a

t

t t

p

NDVI

NDVINDVIWLI dctbSinaNDVI p ))(

2(

where

Cropland

Bi-spectral Gradient Wetness Index (BGWI)

SWIR

NIR

BGWI

Wetland BGWI-NDVI- BGWISummer MosaicNIR-MIR-RED

Pure Water

Analysing pixel

Surface Anisotropy Relative Linear Indexes (SARLI)

RED-NIR: Slope - Slope - Interception

NIR-MIR: Slope - Slope - Interception

SARLI is derived based on the linearised RPV Bi-directional Reflectance Distribution Function model to characterise a surface anisotropy properties

The SibTREES Land Cover Classes

Floristic differencesRemoved for GLC 2000

Global legend. New class =

Evergreen needle leaf

Available Country-wide Forest Inventory Data to compare with GLC2000 Map

Forest inventory database contains for each forest management unit the data on forest area, tree species composition, volume, area of non-forested land categories and some other information

Forest Management Units selected for comparison with GLC 2000 map

679 forest management units

GLC200 Forest Cover in comparison to Forest Inventory Data

GLC200 Forest Cover in comparison to Forest Inventory Data

GLC 2000 Map in comparison to SPOT-HRV image

SPOT-HRV Image

Simplified Forest Map

Simplified GLC 2000

Map

SPOT-VGT Image

Ongoing development to improve the Northern Eurasia’s GLC2000 Product

splitting some of the forest classes according to trees cover density

reducing of ambiguity between “cropland” and “grassland” classes

introducing to the map legend the mosaic classes such as “cropland/natural vegetation” and “forest/other vegetation”

comparison (pre-validation checking) with available forest inventory and other available land cover data to expose main divergences