International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
ACCURACY EVALUATION OF TWO GLOBAL LAND COVER DATA SETS OVER
WETLANDS OF CHINA
Z. G. Niu**, Y. X. Shan*, P. Gong?"
* State Key Laboratory of Remote Sensing Science, Institute of Remote Sensing Applications, Chinese Academy
of Sciences, Beijing, 100101, zhgniu(ggmail.com
? Institute for Global Change Studies Tsinghua University, Beijing, 100084
Theme or Special Session: VII/4: Methods for Land Cover
KEY WORDS: GLC2000, MODIS, China, Wetland, Remote sensing, Accuracy Evaluation
ABSTRACT:
Although wetlands are well known as one of the most important ecosystems in the world, there are still few global wetland mapping
efforts at present. To evaluate the wetland-related types of data accurately for both the Global Land Cover 2000 (GLC2000) data set
and MODIS land cover data set (MOD12Q1), we used the China wetland map of 2000, which was interpreted manually based on
Landsat TM images, to examine the precision of these global land cover data sets from two aspects (class area accuracy, and spatial
agreement) across China. The results show that the area consistency coefficients of wetland-related types between the two global data
sets and the reference data are 77.27% and 56.85%, respectively. However, the overall accuracy of relevant wetland types from
GLC2000 is only 19.81% based on results of confusion matrix of spatial consistency, and similarly, MOD12Q1 is merely 18.91%.
Furthermore, the accuracy of the peatlands is much lower than that of the water bodies according to the results of per-pixel
comparison. The categories where errors occurred frequently mainly include grasslands, croplands, bare lands and part of woodland
(deciduous coniferous forest, deciduous broadleaf forest and open shrubland). The possible reasons for the low precision of
wetland-related land cover types include (1)the different aims of various products and therefore the inconsistent wetland definitions
in their systems; (2) the coarse spatial resolution of satellite images used in global data; (3) Discrepancies in dates when images were
acquired between the global data set and the reference data. Overall, the unsatisfactory results highlight that more attention should be
paid to the application of these two global data products, especially in wetland-relevant types across China.
1. INTRODUCTION
Wetlands are among the most valuable ecosystems in the world
and supply highly valuable services for human welfare
(Costanza, 1997). Accurate information on global wetland
areas and their spatial distribution is therefore important for
wetland management and research. However, there are few
global wetland mapping efforts at present. The Ramsar Sites
Database includes 1757 internationally important wetland sites
covering 158 countries. [http://ramsar.wetlands.org/Database
/AbouttheRamsarSitesDatabase/tabid/812/language/en-US/Def
ault.aspx]. The Global Lakes and Wetlands Database (GLWD),
which has been developed jointly by the World Wide Fund for
Nature (WWF) and the Center for Environmental Systems
Research, University of Kassel, Germany, provide the current
generation of a variety of existing maps, data and information
on the global wetlands together with the application of
Geographic Information System. (Lehner et al, 2004) .
However, the application of the GLWD faces challenges
because the information cannot be updated in time. Though
GlobWetland II project, launched by the European Space
Agency (ESA) in collaboration with the Ramsar Secretariat in
2009, aimed principally to develop a Global Wetlands
Observing System information system, only over 200 wetland
sites in the Mediterranean basin were included in the system.
Until now there has been no global wetland data products based
on identical satellite remote data. Therefore, the global land
cover data sets, though which were not designed for the
extraction of wetland, are the only available data sets at global
scale.
China, with the unique Tibetan wetlands, has extensive rich
wetland types and the total wetland area (except for rice
paddies) is about 3.595*10°km® (Niu,et al, 2009). China
wetlands, therefore, are representative of the world’s wetlands
both in amount and wetland types, and can be used to evaluate
the precision of the wetland-related types in global land cover
datasets.
Since the 1990s, a series of improvements have been made in
Land Use/Cover mapping based on the remote sensing data
source at global scale. So far, there are four global land cover
data sets available with 1 km spatial resolution, which include
IGBP DISCover, UMD land cover products, the MODIS land
cover products and Global Land Cover 2000 (GLC2000) , in
which the AVHRR, MODIS, VEGETATION images are used
respectively. Another new data set - Globcover Land Cover
(2005 and 2009) with 300m spatial resolution - has been
produced recently by ESA through an international partnership
at global scale.
All landcover types, including wetland, of the global land cover
data sets were evaluated during existed research. Herold et al.
(2008) found that precision of the peatlands from MODI12Ql1
and GLC2000 was 38.1% and 45.9% respectively. Chandra
Giria's (2005) study makes a strong case for the inconsistency
of the global land cover data sets which mainly occurs in the
wetlands where the coefficient is only 36.66%. Ran ef al. (2010)
used China land use/cover data to evaluate the existing four
data sets across China and found that peatland type of IGBP
DIScover data set had the highest precision, although it was
just 38%. However, the peatland precision from GLC2000 and
MOD12Q1 data set reached only 0.15% and 0.29%,
respectively. For the UMD data set, as yet there is no separate
peatland category in its classification system. The precision of
the water from IGBP DIScover, UMD, GLC2000 and
MOD12Q1 are 9.25%, 35.12%, 9%, and 9.43% respectively.
The above wetland precision of MOD12Q1 and GLC2000 was
