Geoinformation, monitoring and assessment Environment

Posted December 1996

FAO Land Cover Classification: A Dichotomous, Modular-Hierarchical Approach

by Antonio Di Gregorio
Technical Officer, Remote Sensing and Land Cover Mapping
Project GCP/RAF/287/ITA East Africa
and Louisa J.M. Jansen
Associate Professional Officer, Soil Resources
FAO Land and Water Development Division


from a paper presented at the US Federal Geographic Data Committee (FGDC) Vegetation Subcommittee and Earth Cover Working Group meeting in Washington, DC 15-17 October 1996. The final document and minutes of a meeting between FAO and FGDC representatives (20-22 January 1997) have been posted by the FAO Agriculture Department at http://www.fao.org/waicent/faoinfo/agricult/AGL/AGLS/FGDCFAO.HTM


Summary

The FAO Land Cover Classification is a comprehensive standardised a priori classification system, designed to meet specified user requirements and created mapping exercises, but independent of the scale or means used. The proposed classification can be used as reference classification system because the used diagnostic criteria allow correlation with existing classifications/legends.

Land cover is defined as the observed physical cover including the vegetation (natural or planted) and human constructions which cover the earth's surface. Water, ice, bare rock or sand surfaces count as land cover.

Land Cover Classes are defined by the combination of a set of independent diagnostic criteria, the so-called classifiers, which are hierarchically arranged to assure a high degree of geographical accuracy. Because of the heterogeneity of land cover, the same set of classifiers can not be used to define all land cover types. The hierarchical arrangement may differ from one land cover type to an other. Therefore, the classification is designed according to two main phases:

  1. a dichotomous phase where eight major land cover types are distinguished; and
  2. a modular-hierarchical phase where the set of classifiers and their hierarchically arrangement are tailored to the major land cover type.

This allows the use of the most appropriate classifiers and reduces the total number of impractical combinations of classifiers. The classification system is set up in a flexible way which allows the user to stop at any time dependent on the level of detail required. The classification will supply the user with a mutually exclusive Land Cover Class, a unique Boolean formula, a nomenclature name and a unique numerical code. Both the numerical code and nomenclature name can be used to build up a map legend which can be linked to a user-defined name.

Further definition of the Land Cover Class can be achieved by adding attributes. Two types of attributes, which form separate levels in the classification, are distinguished:

  1. environmental attributes (e.g. climate, landform, altitude, soils/lithology and erosion);
  2. specific technical attributes (e.g. floristic, crop type and soil type).

Because of the complexity of the classification and the need for standardisation a program is under development which will assist the interpretation process, it will reduce heterogeneity between interpreters, and with interpretations over time. The FAO East Africa Project will be the first operational module of the AFRICOVER Programme and the first user of this program.


1. Historical background

Information on land use and land cover is required in many aspects of land use planning and policy development, as a prerequisite for monitoring and modelling land use and environmental change, and as a basis for land use statistics at all levels. A global agreement on the definition of both land use and land cover does not exist. As a result many classification systems and innumerable map legends exist, and maps and statistics from different countries, and in many cases even from the same country, are incompatible with each other. With escalating global needs due to increasing human population, there is an urgent need for better matching of land and its use to increase production while at the same time attempting to protect the environment, biodiversity, and global climate systems (Sims, 1995). Use of remote sensing makes it increasingly possible to map and monitor land use and land cover over wide areas.

From about 1988 the Soil Resources, Management and Conservation Service (AGLS) which has responsibility for land classification and land use planning, began to develop a conceptual basis for a quantitative and objective definition of land uses and their classification (Remmelzwaal, 1989). This resulted in the development of the Land Use Database in collaboration with ITC (de Bie, 1995).

In 1993 UNEP initiated a survey of existing land use and land cover classification methods (Young, 1993), which was discussed at an expert consultation in Geneva the same year. As a result it became clear that immediate adoption of a single global classification would be impractical. Instead it was agreed to concentrate on dialogue with numerous organisations and institutes in order to develop a joint approach.

In the meantime a number of International Geosphere Biosphere Programme (IGBP) initiatives had come into being, with strong links to universities in the United States and Europe. Contacts between the IGBP-DIS (Data and Information Systems) and IGBP-LUCC (Land Use and Land Cover Change) were developed to facilitate exchange of ideas and joint concept development.

The AFRICOVER initiative by the Environmental Information and Natural Resources Service (SDRN) on the definition of a Land Cover Classification was initiated during the expert consultation held in Addis Ababa, 1994. One of the main recommendations was to establish three international working groups which should define the technical specifications for the AFRICOVER Project in order that the project becomes a truly normative programme. One of the working groups, the Working Group on Classification and Legend, defined a standardised classification which could be used for mapping land cover in all African countries. Several papers and preliminary reports (Negre, 1995; Barisano, 1996) were prepared in which existing classifications/legends were analysed as well as nomenclatures and proposals for the standardised classification made. The concepts of the latter were developed (Di Gregorio & Jansen, 1996) discussed and approved at the Dakar meeting in July 1996. Finally the co-operation between the AFRICOVER Working Group and the above mentioned FAO and UNEP initiatives should be mentioned. At a meeting in Rome early 1996, the approaches which were developed in parallel, merged into one common approach.

The overall objective of the joint initiative is to answer to the need for standardisation and to develop a common integrated approach to all aspects of land use and land cover. This implies a methodology which is applicable at any scale, and which is comprehensive in the sense that any identified land cover or land use anywhere in the world can be readily accommodated.


2. Definition land cover and land use

The distinction between land cover and land use is fundamental. In previous classifications and legends the two were often confused. They are defined as follow (Sims, 1995; De Bie, 1995):


3. Classification and legend

Classification is an abstract representation of the situation in the field using well-defined diagnostic criteria, the classifiers. Sokal (1974) defined it as "the ordering or arrangement of objects into groups or sets on the basis of their relationships". A classification system describes the names of the classes and the criteria used to distinguish them. A classification is therefore:

  1. scale independent; and
  2. independent of the means used to collect information (whether satellite imagery, aerial photography or field survey or a combination of them are used).

3.1 Classification structure

3.1.1 Hierarchical Systems
Classification systems come in two basic formats, hierarchical or non-hierarchical. A hierarchical classification offers more flexibility because of its ability to accommodate different levels of information, starting with structured broad-level classes which allow further subdivision into more detailed sub-classes.

3.1.2 A Priori and A Posteriori Systems
Classification can be done in two ways, i.e. a priori and a posteriori.

An a priori classification is based on classes defined before actual data collection takes place. This means that all possible combinations of classifiers must be dealt with in the classification. Basically in the field each sample plot is identified and labelled according to the classification adopted. This method is used extensively in plant taxonomy and soil science. The main advantage is that classes are standardised independent of the area and the means used.

A posteriori classification means that classes are defined after clustering similarity or dissimilarity of samples. The Braun-Blanquet method is an example of this approach. The advantage of this type of classification is its flexibility compared to the implicit rigidity of the a priori classification. On the other hand, because a posteriori classification depends on the area described it is unable to define standardised classes.

3.2 Legend

A legend is the application of a classification in a particular area using a defined mapping scale and particular data set. Therefore, a legend may contain only a proportion, or sub-set, of all possible classes of the classification. Thus, a legend is:

  1. scale and cartographic representation dependent (e.g. occurrence of mixed classes);
  2. data and mapping methodology dependent; and
  3. different legends which do not have a common reference classification system can not be compared and correlated.

4. Conceptual basis

4.1 Criteria for a (Reference) Land Cover Classification

There are many classification systems in existence throughout the world. However, there is no single internationally accepted land cover classification system. Such a system should meet the criteria that:

  1. it must be comprehensive.

  2. it should be an a priori classification system as defined and explained above.

  3. it should be a common reference basis for all derived (and when possible existing) classifications.

  4. it meets the needs of a variety of users (it should not be single project oriented) which may take only part of the classification and develop from there according to their own specific needs.

  5. it must be arranged in a hierarchical structure to be used at different scales and at different levels of detail allowing cross-reference of local/regional with continental/global maps without loss of information. Some existing classifications are designed to be used at a specific scale and/or consider only or mainly classes derived from satellite imagery.

  6. it must be able to describe all land cover features as derived from its general definition, e.g. it must be able to describe ice as well as forest. A proportion of the existing classifications are vegetation classifications, e.g. UNESCO and Yangambi. Other land cover features, such as cultivated areas, bare land or ice, are not considered.

  7. it must be adaptable to the variety of land cover types (all possible combinations of the classifiers should be considered).

  8. a class must be defined by a combination of well-defined diagnostic criteria, the classifiers. In most classifications there is an unclear or unsystematic description of the classifiers from which the class should be derived.

  9. classes must be mutually exclusive and unambiguous.

  10. a clear distinction must exist between the type of classifiers used. Often no underlying common principle has been identified and used to define land cover. Often a mixture of different types of classifier, e.g. land cover and environmental classifiers such as climate, geology and landform. These factors influence land cover but are not inherent features of it. This type of combinations are frequently applied in an irregular way and often do not follow any hierarchy. This leads to confusion in the final nomenclature.

  11. the diagnostic criteria or classifiers used in the classification must be selected because of easy measurement and permanence (they must be easy recognisable in the field and independent of season).

  12. it should be suitable for mapping and monitoring purposes.

  13. it must be scientifically sound and practically oriented.

At present there is in some cases an absence of or solely slight compatibility between two classifications, or between classification and legend. The practical implications of these facts hamper the possibility of the use of these classification results by a wide audience.

4.2 The FAO Land Cover Classification

The general criteria as explained above are the conceptual basis for the proposed classification system. The main conditions to fulfil are:

  1. Emphasis must be given not on the name of a class, but on the classifiers defining the class. The main land cover class is then defined by the combination of a set of independent diagnostic attributes (classifiers).
  2. The classifiers must be hierarchically arranged in order to assure at the highest levels of the classification a high degree of geographical (i.e. mapping) accuracy of the class considered.

To fulfil these requirements two problems arise:

  1. Land cover according to its definition is dealing with a heterogeneous set of classes which can not be defined with the same set of classifiers.
  2. Even when classifiers are the same, their hierarchical arrangement may be different.

In addition, the free combination of the whole set of classifiers involved would lead to an enormous number of combinations (which are, in most of the cases, not present in the field).

The fulfilment of all these requirements without incurring the problems mentioned, is the objective of the design of the proposed method.

4.2.1 Design Criteria
The classification is designed according to two main phases:

  1. a dichotomous phase where a subdivision is made to define eight major land cover categories (see paragraph 4.2.2) from which point onwards,
  2. a so-called modular-hierarchical phase starts. In this phase the creation of a land cover class is based on the combination of a set of pre-defined classifiers. These classifiers are tailored to each of the eight major land cover types (see paragraph 4.2.3).

The tailoring allows the use of most appropriate classifiers to define land cover classes derived from the major land cover types, and at the same time reduces the total number of impractical combinations of classifiers. This results in a class defined by a Boolean formula showing each classifier used, a unique number for the GIS, and a name which can be both the provided standard name or user-defined.

4.2.2 Dichotomous Phase

As stated above a dichotomous key is used at the main level of classification to define the major land cover classes (Figure 1).

Figure 1. The Dichotomous Key of the Land Cover Classification
DICHOTOMOUS
PHASE
VEGETATED NON-VEGETATED
Terrestrial
A1
Aquatic or
regularly flooded land
B2
Terrestrial
B1
Aquatic or
regularly flooded land
B2
Cultivated
terrestrial
A11
Natural/semi-
natural terrestrial
A12
Cultivated aquatic
A23
Natural/semi-
natural aquatic
B24
Built-up &
associated areas
B15
Bare areas
B18
Artificial
water bodies
B27
Inland water
B28

At the main level a distinction is made between:

  1. A. Vegetated Areas: areas which have a vegetative cover1 (which may consist of woody, herbaceous, trees, shrubs, forbs, graminoids, mosses/lichens) of at least four percent during at least two months of the year. This class is determined by the presence of vegetation and the time factor.
  2. B. Non Vegetated Areas: areas which have a total vegetative cover of less than four percent during at least 10 months of the year. This class is determined by the absence of vegetation and the time factor.

At the second level a distinction is made in:

These distinctions result at the third level in eight major Land Cover Categories:

4.2.3 Modular-Hierarchical Phase

In this phase the creation of the Land Cover Class is given by the combination of a set of pre- defined pure land cover classifiers. This set of classifiers is different for each of the eight main land cover types, this difference is due to the tailoring of the classifiers to their respective type (Figure 2)

Figure 2. Overview of the Modular-Hierarchical Phase of the Eight Major Land Cover Categories

Cultivated Terrestrial
A11
IA
Physiognomy
B
Spatial aspect
C
Crop combination
IID
Cultural practices
IIIL
Landform
M/N
Lithology/soil
IVO
Climate
P
Altitude
Q
Erosion
VS
Crop type

Natural/Semi-Natural Terrestrial
A12
IA
Life form & cover
of the main layer
B
Spatial distribution
C
Height of main layer
IID
Leaf type (life cycle)
E
Leaf phenology
IIIF
Stratification
(layering)
G
Cover
H
Height
IVL
Landform
M/N
Lithology/soil
VO
Climate
P
Altitude
Q
Erosion
VIT
Floristic

Cultivated aquatic
A23
IA
Physiognomy
IIB
Spatial aspect
IIIC
Cultural practices
IVO
Climate
P
Altitude
VS
Crop type

Natural/Semi-Natural Aquatic
A24
IA
Physiognomy & cover
of the main layer
B
Height of
main layer
IIC
Water persistence
IIID
Leaf type
E
Leaf phenology
IVF
Stratification
G
Cover
H
Height
VO
Climate
P
Altitude
Q
Water quality
VIT
Floristic

Built-up and associated areas
B15
IA
Surface aspect

Bare Areas
B16
IA
Surface aspect
IIB
Macropattern
IIIL
Landform
O
Climate
IVP
Altitude
Q
Erosion
VM/N
Soil type / lithology

Artificial Waterbodies
B27
IA
Persistence
IIB
Depth
C
Sediment
load
IIIO
Climate
P
Altitude
IVV
Quality

Inland Water
B29
IA
Physical status
B
Persistence
IIC
Depth
D
Sediment load
IIIO
Climate
P
Altitude
IVV
Quality

The land cover classifiers can be combined with so-called attributes. Two types of attributes are considered (which are not explained in detail here):

  1. environmental a/o other type of attributes 2); and
  2. specific technical attributes (e.g. floristic, crop type, soil type and water quality)

The user is obliged to start with the pure land cover classifiers. However, at any time the user can stop, dependent upon the level of detail required and he/she can derive a land cover class. Further definition of this class can be achieved by adding a combination of any of the other types of attributes.

Due to the fact that the classification is designed for mapping purposes (the system gives high priority to mapability) the user needs to follow specific rules:

  1. A higher level must be fulfilled before going to a lower level (because the mapability is high at high levels and decreases with lower levels).

  2. Within each level there are essential pure land cover classifiers (in the figures in bright green) and a further subdivision of them (in the figures of the modular-hierarchical phase in light blue) which defines in greater detail the class. The latter type of classifier is optional and does not need to be fulfilled.

  3. All essential classifiers within one level should be determined before going to a lower level.

  4. At any time inside a level the user can stop, and a mutually exclusive class is defined.

  5. All land cover classes defined in such way are hierarchically arranged.

  6. At any time the user can further define the land cover class using environmental a/o specific technical attributes.

  7. Each land cover class is defined by a Boolean formula (combination of classifiers used), a unique code (numerical) and a name (nomenclature).

  8. The land cover class can be combined with environmental a/o specific technical attributes at any level. The attributes will add a second, separate, code to the land cover class.
4.2.4 Advantages of the Adopted Method

From the conceptual point of view the advantages of the proposed classification are:

  1. It is a real classification system in the sense that it covers all possible combinations of classifiers. Some combinations are excluded due to the conditions which are elements of the classification system. These conditions are clearly explained.

  2. A given land cover class is clearly defined by a set of independent classifiers. The classifiers are clearly differentiated in type, being pure land cover classifiers, environmental and other classifiers and discipline specific classifiers, avoiding an unclear mixture of these.

  3. The classification is truly hierarchical: the difference between a land cover class and a further subdivision of this class is given through the addition of new classifiers. The more classifiers used the greater the detail of the defined land cover class.

  4. The classification can be used as reference classification system. In fact, the emphasis given to the set of classifiers defining the class allows easy correlation between existing classification/legend and the proposed one.

  5. The specific design of the classification allows incorporation into GIS and databases. The pure land cover classes can be used in overlay procedures to make combinations with e.g. climate, physiography, to create new classes.

From the practical point of view:

  1. The classification is designed for mapping. The hierarchical arrangement of classifiers is set up to assure at a high level an precise mapping accuracy (clear definition of boundary between two land cover classes).
  2. It facilitates the integration of different types of data.

  3. It is highly flexible, responding to the information available or gathered in a given area or for the time and budgetary constraints of a project. This means that within one land cover map mapping units will contain the maximum available information but this quantity of information may differ between mapping units. This will not affect the homogeneity of the resulting map.

  4. It rationalises the field data collection. Because the classes are defined by a combination of classifiers, field surveyors should detect the single classifiers and not deal with the final class name. This means that the field survey can be done independent of, or parallel to the interpretation process.

  5. It facilitates the standardisation of the interpretation process contributing to its homogeneity. In fact, the interpreter is not dealing with a final class name but is dealing with one classifier at the time. This reduces heterogeneity between interpreters and with interpretations over time.

  6. It is multi-user oriented. Because the class is defined by a set of classifiers every user can make a re-selection based upon the classifier(s) of interest.
  7. It is designed to map at a variety of scale, from small-to large-scale.

5. Classification program

Because of the complexity of the classification and the need for standardisation a program is under development which will assist the interpretation process, it will reduce heterogeneity between interpreters, and with interpretations over time. The program contains, besides a glossary, the following modules:

  1. Classification: The program starts at the dichotomous phase where one derives the main land cover type. At this phase a choice between two options is made. Dependent on the land cover type selected the user enters the modular-hierarchical phase where the classifiers are defined. In this phase the selection of one classifier at a high level may have consequences for the options available at a lower level. The system is build up in such a way that automatically choices which are no longer valid in connection with a chosen classifier at a high level become inactive. Because the system will be used in several countries by different teams data consistency needs to be guaranteed. At any level the user can ask for the land cover class and store its Boolean formula, its code and class name in the module called "Legend". Any user having derived at a certain class will store the same Boolean formula and same code. The class name can be the given class name or may be user- defined.

  2. Legend: For every mapping unit the user goes again through the same sequence of screens to derive the land cover class which is subsequently stored in the "Legend". To facilitate the interpretation process selection of the land cover class for a repeatedly occurring mapping unit can be done directly from the "Legend" file.

  3. Field Survey Data: The sample site is described and other relevant information can be stored. The data is automatically classified using the information and "translating" it into the classifiers.

  4. Translator Tool: Existing classifications and legends can be translated into the reference classification and vice versa.

The program is still under development and may in the near future be included in an image interpretation software program. An "Agricultural Statistics" module is expected to be developed, which will group automatically the classes according to the classes of the FAO Production Yearbook.

The East Africa Project, which will cover 11 countries of the Nile River Basin, will be the first operational module of the AFRICOVER Project and the first user of the program. The Project will become operational in late 1996. This Project will evaluate and validate the classification and its program. A number of pilot studies are foreseen in the framework of the AFRICOVER Project which will do the same in other agro-eclogical regions in Africa.


References

Barisano, E. 1996 - ... de Bie C.A., van Leeuwen, J.A. and P.A. Zuidema 1996 - "The Land Use Database: A Knowledge Based Software Program for Structured Storage and Retrieval of User-Defined Land Use Data Sets". User's Reference. Includes software diskette.

Di Gregorio, A. and L.J.M. Jansen 1996 - "The AFRICOVER Land Cover Classification Scheme: A Dichotomous, Modular-Hierarchical Approach". Working Paper with the Proposal for the International Working Group Meeting, Dakar 29-31 July 1996.

Gyde Lund, H. 1995 - "FGDC vegetation (classification) standards" - Version 31 July 1995 - USDA - Washington, USA - pp 14.

ITE/WCMC 1995 - "Harmonisation of Global Land Nomenclatures (GGNLCLU)".

Negre, T. 1995 - "Report of the Preparatory Mission: Outlines for the AFRICOVER Classification System". Draft Document. FAO, Internal Document. pp 47.

Sims, D. 1995 - "Background Note On Ongoing Activities Relating to Land Use and Land Cover Classification". FAO, Internal Document, Rome. pp. 7.

Sokal R. 1974 - "Classification: Purposes, Principles, Progress, Prospects" - in Science vol.185, number 4157 - pp 111-1123.

UNESCO. 1973 - "International classification and mapping of vegetation" - Paris, France - pp 32.

Young. 1994 - "Towards international classification systems for land use and land cover - A preliminary proposal prepared for UNEP and FAO" - FAO - Rome, Italy - pp 45.

White, F. 1983 - "The vegetation of Africa - A descriptive memoir to accompany the UNESCO/EATFAT/UNSO vegetation map of Africa" - UNESCO - Paris, France - pp 356.


Further information

Chief, Environmental Information and Natural Resources Service (SDRN)
FAO Research, Extension and Training Division
Viale delle Terme di Caracalla
00100 Rome, Italy
fax: ++39 6 522 55731
email: SDR-Registry@fao.org



Back to Top FAO Homepage