1. The Land Cover Map of Northern Eurasia method, product and initial users' feedback Global Land Cover 2000
S. Bartalev, A. Belward EC JRC, Italy   D. Erchov and A. Isaev CFEP RAS, Russia

2. SPOT 4 - VEGETATION Data
Type of satellite data : S10 product, including Sun -Earth - Sensor angular data
Geographic window : 420N - 750N and 50E -1800E
Time windows : two time-windows were considered:
Data from March to November of year 1999 were used to produce the land cover classification
Data from June to August of year 2000 were employed for the burned area class updating

3. Land Cover Mapping Method
Image pre-processing and generation of advanced data products
Image classification
SPOT4-VGT S10 data
Seasonal mosaics
Initial labelling of clusters
ISODATA clustering of seasonal mosaics
Spectral-temporal clusters map
Wave-Likeness Index
Semantic clusters map
Contaminated pixels and snow cover detection
Generation of the advanced data products
Anisotropy Index
Wetness Index
Decomposing of ambiguous semantic clusters
Merging
of semantic clusters into thematic classes
Generated masks
Mono-semantic clusters map
Snow Cover
GIS Database
(topographic and thematic maps, DEM, forest inventory statistics and etc)
Derived Auxiliary Products
Land Cover Map

4. Normalised Difference Snow Index
Snow/Ice
Clouds
Water
Vegetation
Red channel
NDSI

5. Contaminated pixels detection
Step 1: Detection of the snow related pixels
Step 2: Detection of the pixels contaminated by clouds
Step 3: Detection of the pixels contaminated by defective SWIR detectors

6. Snow-free duration within observation period
Observation period considered is of year 1999

7. Seasonal Mosaics
spring
summer
autumn

8. The seasonal dynamic of the land cover types’ spectral signatures
RED
Bright soil
Dark soil
Water
Spruce forest
Larch forest
Pine forest
Broadleaf forest
Grassland
NIR
Soil line
Max LAI line
Spring
Summer
Autumn

9. Wave-Likeness Index (WLI)
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
NDVI
(NDVI e, t e)
(NDVI max, t max)
(NDVI b, t b) a
b -1 d
Cropland
where

10. Comparison WLI and Land-Use Map for Northern Kazakhstan
Wave-Likeness Index
Land Use Map

11. Bi-spectral Gradient Wetness Index (BGWI)
SWIR
NIR
BGWI
Wetlands
BGWI-NDVI- BGWI
Summer Mosaic
NIR-MIR-RED
Pure Water
Analysing pixel

12. Thematic interpretation of clusters
Both spectral properties and environmental criteria were involved into experts' analysis to assign thematic labels for the clusters:
(i) relations between spectral properties and land cover characteristics (green biomass, water contents and etc.).
(ii)  geographical distribution of natural and anthropogenic phenomena, environmental relation and processes, and etc.
Spectral properties
Semantic hypothesis
Spectral clusters
Thematic classes
Environmental criteria
Thematic interpretation process

13. Use of the advanced data products for land cover mapping
Seasonal dynamic of the clusters' spectral signatures with respect to soil-line was essential to assign main vegetation and land cover types
BGWI was employed to assign clusters to either wetland or dryland cover types
WLI separated cropland from other vegetation types
Snow cover duration was used to localise classes belonging to the tundra
Anisotropy indexes were used to separate deciduous forest and humid grassland, and also dark evergreen needle-leaf forest and Palsa-bogs classes

14. Environmental criteria for the thematic interpretation of clusters
Geographical location
Physiographic factors (climate, altitude above the see level and etc.)
Spatial pattern (compactness, dispersivity, and etc.)
Spatial context
Known facts regarding natural and anthropogenic disturbances in the ecosystems
Natural ecosystem processes (successions, phenology)

15. The land cover map updating
Summer of 1999
Summer of 2000
Forest burns occurred during fire season of year 2000 were detected to update the land cover map

17. Land Cover Map legend OTHER VEGETATION TYPES AND COMPLEXES TUNDRA
WETLANDS
FORESTS
SHRUBLANDS
GRASSLANDS
NON-VEGETATED LAND
COVER TYPES

18. 6 7 1 2 5 8 4 3

19. Land Cover Map : 1. Moscow region

20. Land Cover Map : 2. Northern part of East Siberia (Evenkiya)

21. Land Cover Map : 3. Altai region

22. Land Cover Map : 4. Baykal lake region

23. Land Cover Map : 5. Central part of Yakutia region

24. Land Cover Map : 6. Magadan region

25. Land Cover Map : 7. Kamchatka peninsula

26. Land Cover Map : 8. Sakhalin island

27. The Land Cover Map validation
Qualitative validation
Elimination of macroscopic errors in the land cover map
Evaluation of map acceptance by potential regional users
Quantitative validation
Cross-comparison with existing data on the land cover (national statistic, maps and etc.)

28. Qualitative Validation Approach
Validation is based on 20 х 20 regular grid cells
Systematic evaluation by group of experts for different ecosystem types (forest, tundra, cropland and etc.) using available reference data and experts’ knowledge
Final revision of the land cover map based on the validation database records to eliminate macroscopic errors

29. Qualitative Validation Data Base

30. Sources of reference data for Qualitative Validation

31. Quantitative Validation Approach
Use of the national forest statistic and land-use map for the comparison with GLC 2000 land cover map
Estimation of the land cover classes' area at the level of administrative regions of Russia
Statistical cross- comparison of the land cover classes' proportion

32. Administrative regions of Russia

33. Forested area from GLC 2000 map against forest statistics
Total forest cover of Russia, thousand ha
Total official forest statistic of Russia, 1998
Russian Forest Service Data, 1998
Forest map of former USSR (ed. A.S. Isaev, 1990)
Northern Eurasia map, GLC 2000
Percent of forest cover by administrative regions of Russia, %
National forest statistic, 1998
Northern Eurasia map, GLC 2000
Forest Service statistics, 1998
Northern Eurasia map, GLC 2000

34. GLC 2000 Map in comparison to SPOT-HRV image
SPOT-VGT Image
SPOT-HRV Image
Simplified Forest Map
Simplified GLC 2000 Map

35. GLC 2000 map in comparison to forest map of Russia
Non-changed forest classes
Deciduous Broadleaf forest is replaced by Evergreen Needleleaf
Evergreen Needleleaf forest is replaced by Deciduous Broadleaf
Forests (mainly Deciduous Broadleaf) is not included into National forests statistics

36. Cropland area from GLC-2000 map against IIASA's Land Use map
Cropland area is derived from both maps at the level of administrative regions of Russia.

37. Use of GLC 2000 Map for forest fire monitoring in Russia
GLC 2000 map is employed by Moscow Regional Environment Department for forest fire monitoring. Active fires (on the map in the red colour) are detected from NOAA-AVHRR satellite data.

38. CONCLUSIONS
A new Northern Eurasia land cover map has been created as a part of Global Land Cover 2000 project
The EC Joint Research Center and Russian Academy of Science have established this map to support forest and land management throughout Northern Eurasia
The map is made up of a series of advanced products derived from the S10 VEGETATION-SPOT4 data, including seasonal mosaics, snow duration, directional properties describing anisotropy, wetness index, phonological descriptors

39. CONCLUSIONS
The land cover map has been qualitatively validated by group of regional experts and compared with national forest statistics and land-use data. Validation has shown satisfactory accuracy and reliability of the map
Comparison of the land cover map with existing forest map indicated areas of the forest type changes as result of logging, fires and forest successions and demonstrated limitation of Russian forest inventory system
The land cover map is already employed by forest fire monitoring system of Moscow region and number of other potential users is foreseeing