Legend vs. classification scheme, the challenge for GLC 2000

1
Legend vs. classification scheme, the challenge
for GLC 2000

• H.-J. Stibig

2
GLC2000 Workshop Legend
Main topics

• legend for the GLC2000 product
• approach to be chosen for land cover
  classification
• how do we classify global land cover?
• which land cover classes should be included?
• what legend do we finally need?

• Use of the data set should not be restricted to
  global level
• how to achieve a homogenous classification of
  land cover at the
• global level
• providing relevant information as far as
  possible also for the
• regional and the national level

3
GLC2000 Workshop Legend
land cover information is required
a) for different purposes b) at different
geographical (management and planning
) levels

• e.g. for
• assessment of land resources
• forest inventories
• hydrological models
• vegetation - atmosphere
• interaction models

• global
• regional
• sub-regional
• local

• thematic information needs can range
• from few land cover classes (forest / non-forest)
  
• to large number of classes
• continuous range of values

4
GLC2000 Workshop Legend

• Basic objective of GLC2000 project at global
  level
• to map the distribution and the extent of main
  land cover classes

• uniform and consistent data set
• comparable across regions and countries
• defined reference year 2000
• compatibility to IGPB classes

• Data set should provide a baseline
• for assessment of land resources
• for monitoring land cover change in a global
  context
• input for global climate modeling
• reference data for the implementation of
  international conventions

• different information needs at global level

5
GLC2000 Workshop Legend
aiming at a broader range of applications

• address the regional and sub-regional level
• e.g. monitor processes deforestation,
  desertification, ..
• link to the national level where possible
• number of large less, developed countries with
  outdated national databases where the GLC2000
  could provide baseline information on land and
  vegetation cover

• GLC2000 should
• provide land cover information to the best
  detail possible
• allow to derive regional land cover
  classifications

6
GLC2000 Workshop Legend
Example regional needs for Africa

7
Example needs for arid sub-region (Africa)
8
GLC2000 Workshop Legend
Example (Sub-)Regional Needs Siberia

9
GLC2000 Workshop Legend
different information impact on the definition
of requirements a legend

• Predefined legend
• linked to defined purpose limited value for
  other applications
• forces the user to squeeze the vegetation types
  into a
• predefined scheme

10
GLC2000 Workshop Legend

• Limited number of broad global land cover
  classes
• limit possibilities of describing vegetation at
• regional level
• limit the range of possible applications

• Large number of detailed regional land cover
  classes
• globally -gt number of slightly different classes
• risk of overlap if not well defined
• (open forests vs. woodland, grassland vs. steppe)
• difficulty of handling and interpreting a large
  number of
• classes
• risk of incompatibility of classes when grouping
  at global level

11
GLC2000 Workshop Legend

• Problem of class names
• a ) different regional use of class names
• b ) different understanding in how to interpret
  specific classes
• incompatibility between regions or data sets

• forest (IGBP) gt 60 of tree cover
• many of the Siberian forests with tree cover less
  than 60 would not be called forest - does not
  correspond to regional understanding of forest
• forest (FAO) gt 10 of tree cover
• what is the information content of the class
  finally?
• South American Pampas
• to Steppe, Savannah or Grassland?

12
GLC2000 Workshop Legend
Proposed Flexible classification system
instead of a fixed legend (FAO LCCS)

• land cover types are described by adding up a
  series of land cover
• classifiers and attributes
• no predefined class names and legend
• compatibility at higher level and the final
  legend are achieved by
• grouping according to a selection of these
  classifiers and attributes

13
GLC2000 Workshop Legend
14
GLC2000 Workshop Legend
Building a legend with forest gt 40 tree cover
- ( FAO dense forest)

Forest gt 40
15
GLC2000 Workshop Legend
Building a legend with vegetation cover with
woody component

Woody vegetation
16
GLC2000 Workshop Legend
Building a legend with class impact of
agriculture

Agricultural Impact
17
GLC2000 Workshop Legend
Advantages

• classification of land cover can be done at
  different geographical level at
• different detail, depending on the information
  available
• regional subclasses
• adding regional species information to a global
  forest class
• legend can be formed by grouping of significant
  classifiers and attributes
• according to a specific purpose

• improved possibilities to harmonize
  classifications between regions
• improved compatibility to existing data sets
  (-gtmonitoring capabilities)
• contribution to standardizing land cover
  classification

• Prerequisite
• minimum set of common classifiers and attributes
  to be present
• in order to build the global classes

18
GLC2000 Workshop Legend
Implications for mapping method

• Remote sensing (SPOT VEGETATION)
• spectral information signatures of land cover
  classes
• temporal information seasonal variation of
  spectral signatures
• Limitations in detail of land cover
  classification
• parameters not assessable
• coarse resolution

• foresee integration of ancillary information
• (as a standard procedure ?)

• Step 1 classification (digital) of a single date
  or multi-temporal data set
• Step 2 adding ancillary information
• digital data layers
• manually defined ROI

19

GLC2000 Workshop Legend

20

GLC2000 Workshop Legend Stratification
Layer Eco-floristic zone

21
T GLC2000 Workshop Legend GLC2000 Workshop
Legend
GLC2000 Workshop Legend ROIs- manual
stratification

22
GLC2000 Workshop Legend
Points for discussion

• feedback to the approach of a flexible
  classification system
• based on classifiers
• allowing regional flexibility in class definition

• do we need to define a Minimum Legend for the
  global level?
• if yes which land cover classes?

23
GLC2000 Workshop
Your feedback on the LCCS tool?
Do LCCS definitions meet the regional needs for
land cover classification?

• LCCS specific / technical
• need to specify a minimal set of classifiers for
  building up a global legend?
• need to define additional rules to avoid
  ambiguous definitions?
• need to require certain attribute information ,
  e.g. such as climate?

• are the classifiers of LCCS in a format in order
  to allow operational re-grouping?
• training course on LCCS required

24
GLC2000 Workshop

• Ancillary data
• how to integrate ancillary information in the
  classification procedure?
• need to specify as standard what kind of
  ancillary information to be
• used
• classifications on e.g. ecofloristic zones,
  soils
• data sets of geo-physical parameters

• Methods of mapping
• need to define the mapping methodology