Land Cover Classification System for Continental/Global Applications
Ryutaro Tateishi*, Wen Cheng Gang*, and L. Kithsiri Perera**
Center for Environmental Remote Sensing (CEReS), Chiba University
1-33 Yayoi-cho, Inage-ku, Chiba263, Japan
Fax: +81-43-290-3857
Email: tateishi@rsirc.cr.chiba-u.ac.jp
Weathernews Inc-
D21, 1-3, Nakase, Miharna-ku,Chiba261, Japan
Fax: +81-43-274-5012
Abstract
A new land cover classification system is proposed for continental/ global applications. In this study, a classification system is distinguished from a legend. A classification system is a category system for the classification work while a legend is a category system for the usage of a classified product. A legend is based on a classification system and different legends are possible from one classification system. The ..proposed classification system has a hierarchical structure with the possible extension of classes up to 255. The approximate resolution of satellite remote sensing image for the proposed classification system is from 25Om (future GLI sensor on ADEOS-II, Japan) to 8 km (sampled AVHRR). However detail land cover categories are included in the classification system in order to correspond to other maps and ground survey-
Another important feature of the system is that easily classified categories are in the higher level in the hierarchical classification system. This classification system is developed as the activity of the working group, "l-km Land Cover Database of Asia"(LCWG), in the Asian Association on Remote Sensing (AARS).
I. Introduction
One general problem of land cover legend is that there are many kinds of legends in different countries and different projects. This fact prevents data comparison, exchange and integration. This is why UNEP/FAO has a project on harmonization of land cover and land use classifications (UNEP/FAO 1993). The goal of this project is to make clear the relationships among different land cover/use classification systems both r within and between countries, both within and between applications, from national to regional to global ::- scales. In the International Workshop on Global Databases organized by the International Society for ~.. Photogrammetry and Remote Sensing (ISPRS) Working Group IV/6, the same problem was discussed (tateishi 1995a). One question in the discussion was "Is the standard legend possible?" The answer was No because users community is diverse and there are a variety of needs for land cover. However it was suggested to develop several standards for each of the users community. The important thing here is that a legend must reflect needs of users. It must be also noted that needs are not static. They will evolve over time.
The Land Cover WG(LCWG) of AARS was established in October 1993. The objectives of the production of land cover database of Asia by the LCWG, AARS are:
- First, global change studies (from scientific needs) and
- Secondly, land use management and planning (from social needs)
In 1994, the LCWG tried to classify land cover of Asia using Global Vegetation Index (GVI) data with the nominal resolution of 16 km. A land cover classification system was developed first for this trial (Tateishi1994). In 1995, the LCWG IS carrying out land cover classification using NOAA / NASA Pathfinder
AVHRR Land Data Set with the nominal resolution of 8 km. Through this activity, the first land cover classification system was modified (Tateishi 1995b). At the working Group meeting during the International Symposium on Vegetation Monitoring held at Chiba University from August 29-31, 1995, and through letter communications, many valuable comments and opinions were received for the modified c land cover classification system. The land cover classification system in this paper reflects these comments and opinions.
2. Classification system and legend
The word, "classification system", has been used as the same meaning as "legend". Land cover legend has been decided based on user needs in a country or in a project. The classified result is presented by the legend and its classification work has also been done according to the legend. When a legend has a hierarchical structure, a higher level category is not necessarily easier to be classified than a lower level category. For example, if "forest" and "shrubland" are higher categories than "evergreen" and "deciduous", higher categories are more difficult to classify by multi temporal low resolution satellite data. For this reason, a classification system is distinguished from a legend in this study. A classification system is a category system for the classification work while a legend is a category system for the presentation of a
classified result. That is, multiple legends are possible from one classification system. What authors propose here is a land cover classification system, not a legend.
3. Proposed land cover classification system;
Table l(a) and (b) shows a proposed land cover classification system which has the following
characteristics.
- Hierarchical structure
- Maximum number of possible land cover classes: 255
- Harmonization with internationally accepted land cover classification system
- Inclusion of Asian main land cover types A)
- Possibility to extend globally applicable classification system
- flexibility
- Unique classification among forest, shrubland and grassland
- Separated cropland into tree crops, shrub crops and grass crops
(a): There is seven levels at maximum. The highest level has three categories: Vegetation, Non vegetation, and Water. Going to the lower level, it becomes more difficult to classify by satellite remote sensing. For example, discrimination of forest and shrubland by low-resolution satellite remote sensing data is more difficult than that of evergreen and deciduous. This is why forest ( code: 16 or 42) and shrubland( code:72 or 92) is in lower level than evergreen (code: 14) and deciduous (code:70). Though each level does not correspond to a specific resolution or a map scale, classes of only higher level can be classified by low- resolution satellite remote sensing data. The lower level classes are included for ground truth collection or comparison with detail maps.
(b ): The classification system consists of 59 classes including 47 classes for vegetation, 8 classes for non vegetation, and 4 classes for water. Addition of new classes up to 255 is possible. Class code is recorded in one byte."
