Production of land cover map of Asia, Central Asia, and Middle East with emphasis of the development

1
Production of land cover map of Asia, Central
Asia, and Middle East with emphasis of the
development of ground truth database

• Ryutaro Tateishi, Hiroshi Sato, and Zhu Lin
• Center for Environmental Remote Sensing (CEReS),
  Chiba University
• Japan
• E-mail tateishi_at_ceres.cr.chiba-u.ac.jp
• URL http//asiaserv.cr.chiba-u.ac.jp/frame.htm

2
Outline Production of
land cover map - Geographic area - Legend
- Ground truth data - Classification -
Validation Development of ground truth database
- Global Land Cover Ground Truth Databases
3
Geographic areas for the mapping
Central Asia
Middle East
Asia
4

• ST (Sato Tateishi) land cover
• guideline legend
•  
• - based on LCCS (FAO)
• - modified Stibigs minimum requirement
• - include 65 crop names from FAOSTAT

5
ST Land Cover Guideline Legend (1/3)

• 1. Broadleaf Evergreen Forest
• 101 111 Crop list
• 2. Broadleaf Deciduous Forest
• 112 117 Crop list
• 3. Needleleaf Evergreen Forest
• 4. Needleleaf Deciduous Forest
• 5. Mixed Forest
• 6. Tree Open
• 31. Broadleaf Evergreen Woodland
• 32. Broadleaf Deciduous Woodland
• 33. Needleleaf Evergreen Woodland
• 34. Needleleaf Deciduous Woodland

Yellow classes of GT collection
6
ST Land Cover Guideline Legend (2/3)

• 7. Shrub
• 118, 119 Crop list
• 8. Herbaceous, single layer
• 9. Herbaceous with Sparse Tree/Shrub
• 10. Sparse Herbaceous / Shrub
• 11. Cropland
• 120 Rice, paddy
• 121 166 Crop list ( 104 Coconuts
• 
  165 Wheat )
• 12. Cropland / Natural Vegetation Mosaic
• 13. Tree-Water (Mangrove)
• 14. Wetland
• 15. Lichens / Mosses

7
ST Land Cover Guideline Legend (3/3)

• 16. Bare
• 35. Consolidated
• 36. Bare rock
• 37. Gravels, stones and boulders
• 38. Hardpan
• 39. Unconsolidated
• 40. Bare soil / Other unconsolidated
• materials
• 41. Loose and shifting sands
• 17. Urban
• 18. Snow / Ice
• 19. Water

8
Additional classes

• 201. Forest fire
• 202. Water (60-70) and many small
• islands with sand and salt (30-40)

9
Ground Truth of the whole area 338 sites, 31
classes (17 classes out of them are at global
level)
10
Global Land Cover Ground Truth database (GLCGT
database) The GLCGT database consists of
regional land cover ground truth (RLCGT) data.
The geographical size of a RLCGT data is
flexible, from a city size to a continental size.

--- metadata of RLCGT data (text) --- GT
land cover code data (raster) --- GT
site code data (raster) --- description
of GT sites (text) --- optional data
(text, raster, or any)
11
Regional Land Cover Ground Truth Area
12
Ground Truth of the whole area 338 sites, 31
classes (17 classes out of them are at global
level)
13

• Used data
• VEGETATION S-10 NDVI data sets
• Classification
• Maximum Likelihood method
• for monthly NDVI (January to November)

14
Land Cover Classification Result of Asia
15

• Confusion matrix of training data for 17 global
  level classes (producers accuracy)
• 11 clases are better than 90
• 3 classes are 80-90
• 3 classes are 60-80
• - shrub (75 ), the other parts were
  classified to
• ? bare areas, sparse herbaceous/shrub
• - herbaceous, single layer (69 )
• ? shrub, herbaceous with sparse
  tree/shrub,
• cropland natural vegetation
  mosaic,
• needleleaf deciduous forest
• - herbaceous with sparse tree/shrub (61 )
• ? sparse herbaceous/shrub, cropland
  natural vegetation
• mosaic, needleleaf deciduous
  forest

16
Validation by Mayauxs method (for 31 classes,
not for 17 global level classes)
Reason of unacceptable - three classes of
difficult training data - urban - detail
classes in bare areas
17

• Conclusions
• Asia, Central Asia and Middle East were
  classified to 31 land cover types (17 types at
  global level) by Maximum Likelihood method of
  monthly NDVI.
• - 338 GT sites (1-66 sites for each land cover
  class) were collected.
• - Geographic coverage of GT data is not enough
  because one type of land cover may have different
  data characteristics in different regions.

18

• Conclusions (continued)
• Refinement of ground truth data is necessary,
  especially shrub, herbaceous, and herbaceous with
  sparse tree/shrub
• Accumulation of ground truth data makes land
  cover map more accurate.
• Recommendation development of global land
  cover ground truth (GLCGT) database