Home Articles Land cover classification of Asia using 8km AVHRR data

Land cover classification of Asia using 8km AVHRR data

ACRS 1996

Global Environment

Land cover classification of Asia using 8km AVHRR data

Ryutaro Tateishi and Cheng-Gang Wen
Center for Environemental Remote Sensing (CEReS), Chiba University
1-33 Yayoi-cho, Inage-ku, Chiba 263, Japan
Fax: +81-43-290-3857
E-mail : [email protected]


Abstract

This study has been carried out as an activity of Land Cover Working Group(LCWG) of the Asian Association on Remote Sensing (AARS). The purpose of this study is to develop land cover dataset of the whole Asia with a grid of four minutes. This study consists of the following four components. (1) Establishment of land cover classification system, 92) Collection of ground truth data based on the established land cover classification system from Working group members, (3) Extraction of phonological rules form decision tree method by statistical analysis of phonological feature data of ground truth data for each class. As a result, four-minute land cover dataset was produced and it is planned to be distributed with ground truth dataset and metadata.


1. Introduction

The Land cover Working Group(LCWG) was established in the Asian Association on Remote Sensing (AARS) in October 1993. The final goal of the Working Group is to develop 30 second (approximately 1km resolution) land cover data set of the whole Asia. At this moment, the Working Group consists of 49 members from 29 Asian and Oceanian countries. The working Group tries to develop four minute (approximately 8km) land cover data set of the whole Asia using 12-montly 8-km global AVHRR NDVI data of 1990. Data source is 10-days composites of NOAA/NASA Pathfinder Land Cover Data Set with 8-km resolution in 1990 (Agbu 1994). The dataset is based on Goode Interrupted Homolosine Projection. Parameters contained in the data set include NDVI, Cloud flag, Quality control flag, Scan angle, solar zenith angle, Relative generate cloud free image, the monthly composite data was from 10-days composite data by maximum composite method.

Land cover data is one of the key environmental variable. It is necessary for global study such as carbon circulation and also important for global/continental scale land use planning which is necessary to keep food supply for human and domestic animals in the present age of human population eruption. However there is no reliable land cover data in global/continental scale. Several organizations/groups such as IGBP and UNEP/FAO are trying to develop land cover data set of global or continental area. The purpose of IGBP’s land cover project is global change study, and UNEP/FAO puts on land use planning in Africa. The LCWG of AARS has general purpose for land cover mapping and focused on Asian and Oceanian regions.


2. Land cover classification system

the land cover classification system was proposed a shown in Table 1 which is based on the following concepts.

  1. Reason of the proposal of new land cover classification system
    the main on -going land cover projects of continental/global scale are IGBP-DIS land cover project and AFRICOVER project. The former aims at global modeling for global change study and the latter aims at land use planning mainly for agricultural development. This study proposes more general land cover classification system which meets both scientific and social needs. For scientific needs, the proposed classification system has similar classes to IGBP-DIS land cover classification system. Key class for social needs is cropland. In the proposed system, cropland is basically divided into three types such as tree crops, and grass crops in order to match the classification system for scientific needs.
  2. Land cover classification system and legend
    the word, “a classification system”, has been used as the meaning as ‘a legend”. Land cover legend has been decided based on user needs in a country or in project. The classified result is presented by the legend and its classification work has also been done according to the legend. In this study, a classification system is defined as a category system for the classification work while a legend I a category system for the presentation of a classification can be merged as forests in a legend when it is displayed. That is, multiple legends are possible from one classification system. What authors propose here is a land cover classification system, not a legend.
  3. Class ID No.
    The proposed 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 possible. Class code is recorded in one byte.
  4. Hierarchical System
    Hierarchical system itself has been well adopted method for classification system. In some hierarchical classification systems, classes of the same level has similar characteristics. However, in the proposed system, Oil palm and Coconut are in the 7th level and Paddy and Wheat are in the 4th level. This is because types of forest are more complicated than types of grassland.
  5. Interpretability
    Continental of global land observation by satellite is often carried out by AVHRR data with the resolution of 100m. In the future, satellite data with 250 meter resolution such as of EOS-AMI1 and GLI of ADEOS-II will be available. In classes of forest or shrubland, more easily interpretable classes by these satellite data set in the higher level of hierarchical classification system. For example, “Evergreen” and “Deciduous” are in higher levels than “Forest” and “shrubland” because discrimination of Evergreen and Deciduous is easier than that of Forest and Shrubland.
  6. Forest, Shrubland, and Grassland
    For the purpose of global change studies, the discrimination of vegetation into forest, shrubland, and grassland is important. Shrubs is small woody plants that are branched from the base. The proposed system used a threshold value of 3 meters to distinguish shrublad from forest.
    Since the discrimination between forest ad shrubland by low-resolution satellite remote sensing data is difficult, two classes, “Forest” and “Shrubland” are combined into a larger class, “Forest or shrubland”. There classes such as “Forest or shrubland (code: 12)”, “Grasland (code: 130)”, and “Mixed vegetation (code 160)” are proposed for discrimination forest, shrubland, grassland, and their combination.
  7. Harmonization
    The proposed classification system has a harmonized characteristics with IGBP-DIS classification system because it is the main global classification system for use of remote sensing. Threshold values of 60% of canopy cover for forest of shrubland and 10% of vegetation cover for Non vegetation are selected in order to match the IGBP-DIS classification system. However the threshold of tree height discriminating shrubland from forest is decided as 3 meter. Regarding thresholds for forest, FAO and UNESCO have different values: over 40% canopy cover for open forest and over 70% for closed (or dense) forest. One reason to selected IGBP-DIS threshold of 60% is the proposed classification system has an emphasis on global change studies, and the other reason is two thresholds of 40% and 70% are difficult to discriminate by low resolution remote sensing images.
  8. Inclusion of Asia main land cover types and flexibility
    In the lower level of hierarchical classification system, Asian types of land cover were included, for example Coconut of Philippine, Pasture of Mongolia, Rubber and Oil palm of Malaysia, and Paddy of Sri Lanka. The proposed system is flexible because other types of land cover class can be added at the lowest level f hierarchical classification system.


3. Ground truth collection

Ground truth data were collected mainly from existing maps by the cooperation of the Working Group members. Ground truth data of 33 types of land cover classes were collected from WG members. A few number of ground truth data were added by ground survey of Kazakstan and Vietnam. The collected ground truth data based on the proposed land cover classification system are planed to be distributed together with the developed land cover data set.


4. Extractin of phonologically features and classification method

In this study, a clustering, K-means method, was applied for monthly composite NDVI data of 1990 independently both in global scale and in Asian and Oceanian regions. 78 clusters in global scale and 80 clusters in Asia and Oceania region were derived, respectively. Since AVHRR data in Pathfinder Data Set in the northern high latitude region over 55 degree north are not available in winter season because of low sun elevation angle, a clustering was also applied for monthly data from March to November. Furthermore according to Loveland’s method (Loveland 1944), there are the other four kinds of phonological features were derived as follows.

(1) onset : the month in which the NDVI first rose threshold value, it corresponds to the time of appearance of green vegetation at the beginning of the growing season.

(2) duration : the number of months when the NDVI reached or exceeded a threshold value, it is similar to the length of growing season.

(3) peak : the month in which the maximum NDVI occurred, it corresponds to the time of
maximum vegetation activity.

(4) total : the mean value of NDVI from January to December 1990 period, it reflects total vegetation activity.

By the statistical analysis of phonological features of the ground truth data, discrimination rules were derived. These derived rules were integrated in order to determine tree classification algorithm.

ACRS 1996

Global Environment

Land cover classification of Asia using 8km AVHRR data


5. Distribution of he developed land cover data set

Four minute grid land cover data set will be distributed in spring 1997 with ground truth data, phonological image data derived from monthly NDVI, and metadata with complete description abut the data.


6. Acknowlegment

the authors would like to thank the members of Land Cover Working Group of the Asian Association on Remote Sensing (AARS) for their cooperation to this project.


References

  • Agbu,P.A. and M.E.James, “The NOAA/NASA pathfinder AVHRR Land Data Set User’s Manual”, Goddard Distributed Active Center, NASA, Space Flight Center, Greenbelt. September 1994.
  • Loverland, T.R., J.W.Merchant, D.O. Ohlen, and J.F. Brown, :Development of a Land-Cover Characteristics Database for the Conterminous U.S., Photogrammetric Engineering and Remote Sensing , Vol. 57, No. 11, pp.1353-1463, 1994.
  • Tateishi, R., C.Wen, and K.Perera, “Working Group Report and Land Cover Database of Asia”, Proc. 15th ACRS, 17-23 Nov., 1994, pp. M-3.
  • Tateishi, R.(ed), “Report of the International Workshop on Global Databases”, International Archives of the Photogrammetry and Remote Sensing , Vol. XXX, Part 4W1, Boulder, 30-31 May, 1995.
  • UEP/FAO, “Report of the UNEP/FAO Expert Meeting on Harmonizing Land Cover and Land Use Classification “, Geneva, 23-25 November, 1993.

Table 1(a) Proposed land cover classification system (LCWG, AARS) July 1996


Table 1 (b) Explanation of land cover classes

10:Vegetation
vegetation but cannot be interpreted
into any class from 12 to 184;
hereafter the meaning of the underlined
part is written like (12-184)

12:Forest or shrubland (14-120)
forest or shrubs canopy cover is>60%

14:Evergreen forest or shrubland (16-60)
Canopy is never without green filiage.
Evergreen canopy cover >60%

16:Evergreen forest (18-36)
Forest canopy cover is>60. Tree
height is exceeding 2 meters.

18:Evergreen broadleaf forest (20-33)

20:Natural evergreen broadleaf forest

23:Oil palm

24:Coconut

33:Other evergreen broadleaf tree crops

36:Evergreen needleaf forest

42:Evergreen shrubland (44-57)
Shrubs canopy cover is >60%. Tree
height is less than 3 meters.

44:Natural evergreen needleaf shrubland

46:Evergreen shrub crops (47-57)

47:Tea

57:Other evergreen shrub crops

10%<forest canopy cover <60%
10%<shrub canopy cover <60%

70:Deciduous forest of shrubland (72-110)
With an annual cycle of leaf-on and
Leaf-off periods. Deciduous canopy
cover >60%.

72:Deciduous forest (74-90)

Forest canopy cover is >60%. Tree
height exceeding 3 meters.

74:Decidous broadleaf forest (76-87)

79:Rubber

87:Other deciduous broadleaf tree crops

90:Deciduous needleaf forest

92:Deciduous shrubland (94-107)
Shrub canopy cover is >60%. Tree
height is less than 3 meters.

94:Natural decidous shrubland

96:Deciduous shrub crops (97-107)

97:Cotton

107:Other deciduous shrub crops

110:Deciduous forest and shrubland

10%<forest canopy <60%
10%<shrub canopy cover <60%

120:Mixed forest or shrubland
Neither evergreen nor deciduous forest
or shrubs exceeds 60% of coverage.

130:Grasslad (132-157)
Tree and shrub cover is less than 10%.

132:Natural grassland / pasture

140:Grass crops (141-157)
Include cereal and other grass type crop

141:Paddy

142:Wheat

143:Suarcane

144:Corn

146:Wheat and rice

157:Other grass crops

160:Mixed vegetation
10% <forest or shrub canopy<60%
10%<grass cover <60%
include savanna and mixed land with cropland, shrubland,
Forest and few houses

170:Wetland (172-174)

172:Mangrove

174:Swamp
Any type of wetland with vegetation
except mangrove

180:Little vegetation (182-184)
Vegetation cover is more than 10% at
the peak season

182:Tundra

184:Other little vegetation

190:Non vegetation (191-210)
Vegetation cover is less than 10% at
Any time of a year

191:Bare ground (192-195)

192:Rock

193:Stones or gravel

194:Sand

195:Clay

200:Perennial snow or ice

220:Water (222-226)

222:Inland water
Lake, pond, river,rservoir

224:Water with seasonal change
Inland water with dry period

226:Tidal flat

For example, Evergreen forest and shrubland (60) is included in Evergreen forest or shrubland (14), Forest or shrublad (12) and Vegetation (10). Therefore an explanation of Evergreen forest and shrubland (60) included the explanation of the classes, 14,12 and 10. That is, the explanation, “Evergreen canopy cover>60%”, is included implicitly for Evergreen forest and shrubland (60),