AARS Asia 30-second Land Cover Data Set
1. Introduction
Land cover is one of key environmental variables for global change studies such
as carbon circulation and also it is important variable for global/continental
scale land use planning in order to keep food supply for human and domestic
animals and to keep sustainable environment in the present age of human population
eruption.
However there was no reliable land cover data in global/continental scale. But
fortunately, land cover is one of the main features of the terrestrial environment
which can be extracted by remote sensing technique.
Therefore there are several organizations/groups such as IGBP, UNEP/FAO, CORINE
which are trying to develop land cover data set of global or continental area.
The Land Cover Working Group(LCWG) of the Asian Association on Remote Sensing(AARS)
has also developed land cover data set to meet both scientific needs (global
change studies) and social needs(global/continental scale land use planning).
2. Land cover classification system by LCWG/AARS
The AARS land cover classification system was developed through discussion with
members of the LCWG/AARS.
The land cover classification system was developed based on the following concepts.
(1) Basic concept of the developed land cover classification system
Main on-going projects on the development of land cover for continental/global
scale are IGBP-DIS land cover project and AFRICOVER project. The former aims
at global modelings for global change studies and the latter aims mainly at
land use planing for agricultural development. The LCWG/AARS proposes more general
land cover classification system which meets both scientific and social needs.
For scientific needs, the proposed classification system has similar land cover
classes to IGBP-DIS land cover classification system. Table 1 shows the corresponding
classes between the LCWG/AARS classification system and IGBP-DIS classification
system. Key land cover class for social needs is cropland. In the proposed LCWG/AARS
classification system, cropland is basically divided into three types such as
tree crops, shrub crops, and grass crops in order to match the classification
system for scientific needs.
Table 1 Corresponding classes between the LCWG/AARS classification system and IGBP-DIS classification system
|
|
LCWG/AARS class (class code) | IGBP-DIS class |
|
|
Evergreen broadleaf forest (18) | Evergreen broadleaf forests |
Evergreen needleleaf forest (36) | Evergreen needleleaf forests |
Deciduous broadleaf forest (74) | Deciduous broadleaf forests |
Deciduous needleleaf forest (90) | Deciduous needleleaf forests |
Evergreen shrubland (42) + Deciduous shrubland (92) | Closed shrubland |
Grassland (130) | Grasslands |
Grass crops (140) | Croplands |
Mixed vegetation (160) | Cropland/natural vegetation mosaic |
Wetland (170) | Permanent wetlands |
Bare ground (191) | Barren |
Perennial snow or ice (200) | Snow and ice |
Built-up area (210) | Urban and built-up |
Water (220) | Water bodies |
|
(2) Land cover classification system and legend
The word, "a classification system," has been used as the same meaning as "a
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. In the concept of LCWG/AARS,
a classification system is defined as a category system for the classification
work while a legend is a category system for the presentation of a classified
result. For example, though, in the AARS land cover classification system, forests
are divided into several categories, these divided categories can be merged
as forests in "a legend" when it is displayed. That is, multiple legends are
possible from the AARS land cover classification system.
(3) Class code
The classification system of this CD-ROM 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.
(4) Hierarchical system
Hierarchical system itself has been well adopted method for a classification
system. In some hierarchical classification systems, classes of the same level
has similar characteristics. However, in the classification system of this CD-ROM,
the same level in the hierarchy does not necessarily have the same classified
level. For example, Oil palm and Coconut are in the 7th level and Paddy and
Wheat are in the 4th level. This is because classes of forest are divided more
than classes of grassland.
(5) Interpretability
Continental or global land observation by satellite is often carried out by
AVHRR data with the resolution of 1km. In the future, MODIS of EOS-AM1 and GLI
of ADEOS-II with 250m resolution will be available. In the classes of forest
or shrubland, more easily interpretable classes by these satellite data are
put in the higher level of the 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 classification system uses a threshold
value of 3 meters height to distinguish shrubland from forest. Though three
classes of Forest, Shrubland, and Grassland are important, Forest and Shrubland
are combined in the 2nd level of hierarchical system because the discrimination
of Forest and Shrubland is difficult by low resolution satellite data such as
AVHRR.
(7) Harmonization
The AARS land cover classification system has a harmonized characteristics with
IGBP-DIS classification system because it is the main global land cover classification
system for global change studies by the use of remote sensing. Threshold values
of 60% of canopy cover for Forest or 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 because some shrubs are higher than 2 meters which is the
threshold by IGBP-DIS classification system. 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. A reason not to select FAO/UNESCO
thresholds is that two thresholds of 40% and 70% are difficult to discriminate
by low resolution remote sensing images.
(8) Inclusion of Asian 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, 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 in the hierarchical classification
system.
For further reading about AARS land cover classification
Ryutaro Tateishi, Wen Cheng Gang and L. Kithsiri Perera, Land cover classification
sytem for continental/global applications, Asian Pacific Remote Sensing Journal,
Vol.10, No.1, pp.1-9, July 1997
3. Ground truth collection
Ground truth data in this CD-ROM means geographically specified regions which
are identified one of classes in the AARS land cover classification system by
class code. Collection of good ground truth data is a key issue for reliable
land cover mapping.
Ground truth data were collected mainly from existing thematic maps by the cooperation
of the working group members. The used maps are listed in the Appendix 1 at
the bottom of this document. Some of ground truth data were collected by field
survey in Central Asia such as Kazakhstan, Uzbekistan and Turkmenistan. The
following three field trips were performed with the cooperation of WG member
of Kazakhstan.
(1) From August 23, 1996 to September 2, 1996 from Almaty to Akmola(Tselinograd) of Kazakhstan
(2) From July 5, 1997 to July 23, 1997 from Akmola(Tselinograd) to Kustanaj of Kazakhstan
(3) From April 26,1998 to May 8, 1998 from Almaty of Kazakhstan, through Uzbekistan, to Ashkhabad of Turkmenistan.
In order to see examples of photographs taken in the field surveys, please click
here.
Ground truth data of 31 land cover classes were collected from 19 types of information
sources which are thematic maps and field surveys. Geographical regions of each
ground truth data are recorded in the data set, "gt.gif", and the information
sources for each ground truth data are also recorded in the data set, "gtsc.gif".
In order to see the ground truth data, please click here.
4. Used data
4.1 AVHRR data
Global Land 1-km AVHRR Data Set was used as the source of satellite data. 10-days
composite data of AVHRR NDVI, channel 4, and channel 5 were used for this project.
NDVI data from April 1, 1992 to March 31, 1993 and channel 4 and channel 5 data
from April 1, 1992 to October 31, 1992 were used.
For further information about Global Land 1-km AVHRR Data Set:
http://edcwww.cr.usgs.gov/landdaac/1KM/1kmhomepage.html
4.2 Elevation data
The Global Land One-kilometer Base Elevation(GLOBE), Version 1.0 , was used
in this project. GLOBE data is a global 30 arc-second grid digital elevation data.
For further information about the GLOBE:
http://www.ngdc.noaa.gov/seg/topo/globe.shtml
4.3 Digital Chart of the World(DCW) data
The DCW is a 1:1,000,000 scale vector base map of the world with 17 attibute
layers. The seashore lines and national boundaries were used in this project
for geometric registration and product's display.
5. Use of the ratio of land surface temperature(Ts) and NDVI
Several studies (Janodet 1994, Lambin and Ehrlich 1995) demonstrate the advantage
of combining NDVI with land surface temperature (Ts) data derived from AVHRR
channel 4 and 5 for single year land cover classification. Ts is related, through
the surface energy balance equation, to surface moisture availability and evapotranspiration,
as a function of latent flux (Carlson 1981). The combination of Ts and NDVI
time series allows to characterise surface conditions both in terms of fractional
vegetation cover, surface moisture status and surface resistance to evapotranspiration
(Goward and Hope 1989, Nemani 1993). The ratio between Ts and NDVI is an adequate
measure of the biophysical information contained in the Ts-NDVI space because
it mainly quantifies variations in both Ts and NDVI which are characterized
by a negative Ts-NDVI relation, i.e., the variations which are bio-physically
meaningful (Lambin and Ehrlich 1995). Since Ts displays the opposite trend to
NDVI when moving from sparse to dense vegetation landscapes, the use of the
ratio Ts/NDVI increases the capability of discrimination of vegetation classes.
The ratio of Ts/NDVI has been interpreted biophysically as regional surface
resistance to evapotranspiration (Nemani and Running, 1989). This concept provides
theoretical support for using this ratio in land cover analysis.
6. Data preprocessing
6.1 Extraction of land surface temperature (Ts) data
The land surface temperature, Ts (in Kelvin), was derived from channel 4 and channel 5 using
split window algorithm (Price 1984). The formula to derive Ts in centigrade
is as follows.
Ts =T4 + 3.33 (T4-T5)
Where T4 and T5 are brightness temperature (in Kelvin) of AVHRR channel 4 and channel 5
. The formula assumes a constant surface emissivity of 0.96.
6.2 Monthly compositing
The maximum value composite method (Holben, 1986) was applied to AVHRR NDVI
and Ts 10-days composite data to get monthly composite data. The maximum value
of NDVI and Ts were selected independently for every month. Ts responds both
to short-term variations in energy balance related to rainfall events and changes
in soil moisture, and to seasonal changes. The monthly composite of Ts data
artificially removes the short time scale variations in 10-days composite Ts,
leaving only the seasonal trend. It mainly includes lower frequency information,
which is related to land cover types (Lambin and Ehrlich,1995). The ratio of
the maximum Ts and maximum NDVI ratio were calculated for every month. The actual
calculation of the ratio was done by the formula (Ts - 160) / (NDVI + 1)
in order to avoid zero value in a numerator and a denominator. However, for
simplicity, this ratio is referred to as Ts/NDVI in this manuscript.
6.3 Transformation of map projection and geometric registration
The map projection of NOAA AVHRR 1-km Land Data Set is based on the Interrupted
Goode Homolosine projection. Monthly NDVI data and the derived Ts/NDVI data
based on this projection were transformed to latitude/longitude projection (Plate
Carree Projection) with 30 second grid.
The Digital Chart of the World(DCW) was used as reference of geometric registration.
The vector data of seashore lines in the DCW was transformed to raster data
with 30 second grid.
By comparing NDVI and Ts/NDVI data with the 30-second DCW seashore line data,
there are one pixel difference at most parts of Asia and about three or four
pixels difference at some regions in high latitude. Then geometric registration
was applied using 250 GCPs in the DCW. The positional accuracy after geometric
registration was 0.5 pixel(one pixel: 30-second grid) in these 250 GCPs.
After transformation of map projection and geometric registration, the following
rectangular region were extracted to cover the whole Asia.
Location of the center of the upper left pixel:
25deg 0min 15sec east in longitude, 89deg 59min 45sec north in latitude
Location of the center of the lower left pixel:
25deg 0min 15sec east in longitude, 14deg 59min 45sec south in latitude
Location of the center of the upper right pixel:
165deg 0min 15sec west in longitude, 89deg 59min 45sec north in latitude
Location of the center of the lower right pixel:
165deg 0min 15sec west in longitude, 14deg 59min 45sec south in latitude
The pixel numbers of the extracted region is 20,400 by 12,600.
6.4 Dataset preparation for classification
By the data preprocessing described above, the following data were prepared
for the classification.
- Ts/NDVI : seven monthly data from April to October 1992
- Maximum NDVI : the maximum monthly data from April 1992 to March 1993
- Minimum NDVI : the minimum monthly data from April 1992 to March 1993
- Digital elevation data
All these data are registered together in 30-second grid in latitude/longitude.
7. Classification
Land cover classification was done by the following steps.
(1) Clustering of monthly Ts/NDVI from April to October
(2) Finding classification rules for decision tree method
(3) Classification by decision tree method
(4) Post-classification modification
Clustring by ISODATA was applied to monthly Ts/NDVI from April to October 1992.
One hundred clusters were obtained as a result of clustering.
Rules for decision tree method were found by comparing cluster numbers, maximum
NDVI, minimum NDVI, digital elevation data of the ground truth regions. Out
of one hundred clusters, 46 clusters can be directly assigned to one of sixteen
land cover classes. The rest of clusters are assigned land cover classes using
threshold values of maximum NDVI, minimum NDVI or digital elevation data. The
threshold value of 0.15 for maximum NDVI is used to discriminate "vegetation"
and "non vegetation". The threshold value of 0.23 for minimum NDVI is used to
discriminate evergreen vegetation and the others.
After classification by decision tree method, there are some unnatural patterns
in classified image due to noises in AVHRR or the effect of mosaicking to produce
10-days composite AVHRR. To eliminate these undesirable patterns, land cover
classes of these portion were substituted by the higher level land cover class
in hierarchical classification system.
8. Conclusion
By the method describing in the previous sections,
(1) Asia 30-second land cover data set ("land.raw") and
(2) Asia 30-second ground truth data ("gt.gif" and "gtsc.gif") were produced.
References
Brown, J.F., Loveland, T.R., Merchant, J.W., Reed, B.C., Using Multisource Data
In Global Land Cover Characterization. Photogrammetric Engineering and Remote
Sensing, Vol. 59, No.6, pp. 977-987, 1993
Carlson, T.N. and Dodd, J.K.. Remote estimation of surface energy balance, moisture
availability and thermal inertia. Journal of Applied Meteorology. Vol. 20, pp.67-87, 1981
Ehrlich, D. and Lambin, E.F. Broad scale land cover classification and interannual
climatic variability. International Journal of Remote Sensing, Vol. 17, No.
15, pp.845-862, 1996
Eidenshink, J.E. and Faundeen, J.L. The 1-km AVHRR Global Land Data Set, International
Journal of Remote Sensing, Vol.15, pp.3443-3462, 1994
Gregorio,A., AFRICOVER Land Cover Classification, International Working Group
Meeting by FAO, Dakar, 29-31 July 1996
Goward, S.N. and Hope, A.S.. Evaporation from combined reflected solar and terrestrial
radiation: Preliminiary FIFE results from AVHRR data. Advances in Space Research.
Vol. 9, pp.239-249, 1989
Holben, B.N., Characteristics of maximum-value composite image from temporal
AVHRR data. International Journal of Remote Sensing, Vol. 7, pp.1417-1434, 1986
Janodet, E. Cartographie fonctionnelle decosystems tropicaux africainsa partir
de levolution temporelle de leurs reponses radiometriques. These de Doctorat.
Universite de Paris XI Orsay. Paris, France., 1994
Kalensky, Z.D. Regional and Global Land Cover Mapping and Environmental Monitoring
by Remote Sensing. International Archives of Photogrammetry and Remote Sensing
Vol.31, ISPRS Commission IV, Vienna Austria, 1996
Kerber, A.G., and Schuut, J.B., Utility of AVHRR channel 3 and 4 in land cover
mapping. Photogrammetric Engineering and Remote Sensing. Vol. 52, pp. 1877-1883,
1986
Lambin, E.F., and Ehrlich, D., Combining vegetation indices and surface temperature
for land cover mapping at broad spatial scales. International Journal of Remote
Sensing, Vol.16, pp.573-579., 1995
Loveland,T.R., J.M. 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.1453-1563, November 1991
Loveland, T.R. and Belward, A.S. The IGBP-DIS global 1km land cover data set,
DISCover: first results. International Journal of Remote Sensing, vol. 18, No.
15, pp.3289-3295, 1997
Nemani, R.R. and Running, S.W.. Estimation of surface resistance to evapotranspiration
from NDVI and thermal-IR AVHRR data. Journal of Applied Meteorology. Vol.28,
pp.276-284, 1989
Nemani, R., Pierce, L., and Running, S. Developing Satellite-derived Estimates
of Surface Moisture Status. Journal of Applied Meteorology, Vol. 32, pp.548-557, 1993
Price, J.C., Land surface temperature measurements from the split-window channels
of the NOAA-7 Advanced Very High Resolution Radiometer, Journal of Geophysical
Research. Vol.89, pp.7231-7237, 1984
Running,S.W., T.R. Loveland and L.L. Pierce, A Vegetation Classification Logic
Based on Remote Sensing for Use in Global Biogeochemical Models, AMBIO Vol.23,
No.1, pp.77-81, February 1994
Running, S.W., T.R. Loveland and L.L. Pierce, Vegetation Classification Logic
Based on Remote Sensing for Use in Global Biogeochemical Models. AMBIO Vol.
23, No. 1, pp.77-81, February 1994
Tateishi,R.(Ed.), Report of the International Workshop on Global Databases,
International Archives of Photogrammetry and Remote Sensing, Vol.XXX, Part 4W1,
Boulder, 30-31May, 1995
Tateishi, R. and Wen, C.G., Global Four Minute Land Cover Data Set. Journal
of the Japan Society Photogrammetry and Remote Sensing, Vol.36, No.4, pp.62-74,
1997
The International Geosphere-Biosphere Programme: A Study of Global Change (IGBP)
of the International Council of Scientific Unions. The DIS 1km Land Cover Data
Set. Global Change NewsLetter, No. 27, September, 1996
Townshend, J.R.G.(Edit), Improved Global Data for Land Applications, Global
Change Report No.20, 1992
Tucker, C.J., Townshend, J.R., and Goff, T.E., African land cover classification
using satellite data. Science, Vol.227, pp.369-375., 1985
UNEP Environment Assessment Programme for Asia and the Pacific. Land Cover Assessment
and Monitoring, Malaysia. Environment Assessment Technical Reports, November,
1997, VOL 9-A
UNEP Environment Assessment Programme for Asia and the Pacific. Land Cover Assessment
and Monitoring, Thailand. Environment Assessment Technical Reports, VOL 8-A,
November, 1997
UNEP/FAO, GEM Report series No.25, Report of the UNEP/FAO Expert Meeting on
Harmonizing Land Cover and Land Use Classifications, Geneva, 23-25 November
1993
Wen, C.G and Tataishi, R. Classification Methodology for Land Cover Mapping
Using Global Monthly 8-km AVHRR Data. Proceedings of the 16th Asian Conference
on Remote Sensing, S-7, Nakhon Ratchasima, Thailand, 20-24 November, 1995
Wen, C.G and Tataishi R. Production of Land Cover Dataset of Asia. Proceedings
of IGBP/BAHC-LUCC Joint Inter-Core Projects Symposium on Interactions between
the Hydrological Cycle and Land Use/Cover, Kyoto, Japan, pp.34-37, 4-7 November,
1996
Wen, C.G and Tateishi, R. AARS Global Four-minute Land Cover Data Set. Proceedings
of the 18th Asian Conference on Remote Sensing, JS-1, Kuala Lumpur, Malaysia,
20-24 October, 1997
Wyatt,B., D.Briggs and H.Mopunsey, CORINE: An information system on the state
of the environment in the European Community, Building Databases for Global
Science (book),Taylor & Francis, pp.378-396, 1988
Appendix 1. Land use and land cover maps used for ground truth collection