Bo EN a SE 
N ET NES 
    
types are in sequence Croplands (12), Barren or Sparsely 
Vegetated (16), Grasslands (10), Open Shrublands (7), Mixed 
Forest (5), Woody Savannas (8) and Evergreen Coniferous 
Forest (1). 
Croplands, grasslands and bare areas in both global data sets 
are the landcover types where wetland water omission is most 
likely to occur because of the coarse spatial resolution of global 
data sources. The inherent limitation of coarse spatial 
resolution during landcover classification mapping is well 
known (Latifovica et al., 2004), and the mixed pixels (the 
heterogeneity landscape of the landcover categories) contribute 
largely to product error. The second reason is the difference of 
the acquired date of data sources. Usually wetland water 
changes dynamically along with the seasons and the changes 
are huge due to the distinct seasonal precipitation pattern across 
China. An example is the confusion between established water 
bodies in the reference data and the regularly flooded shrub 
and/or herbaceous cover data obtained by GLC2000. 
The reason for wetland water being mistakenly classified as 
snow and ice is the different acquisition date of satellite images 
between the global data set and the reference data. For example, 
the great omissions of wetland water in area of north Tibet, 
southwestern of Qinghai, the Boston lake region and the 
Sayram lake area was caused possibly by using winter images 
of GLC2000. In addition, the crystallization of a salt lake is 
similar to a water lake in the satellite images, and this will also 
cause some errors in the classification results (Hong et al, 
2006). 
Table 5 the omission proportion (%) between the global 
land cover products and reference data 
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 
  
  
SIC Wetland Peatland MOD12Q1 Wetland Peatland 
code water code water 
1 1.23 0.12 0 42.93 5.08 
2 1.16 4.17 1 3.06 0.48 
3 0.00 0.00 2 0.97 0.16 
4 4.29 2.17 3 0.04 0.30 
5 0.47 2.96 4 0.40 0.61 
6 0.03 0.09 3 5.23 4.15 
9 07% 0.14 6 1.33 0.52 
10 0.00 0.00 7 6.78 13.86 
11 2.35 1.06 8 3.37 2.81 
12 0.05 0.54 9 0.92 1.18 
13 15.95 45.88 10 7.17 31.09 
14 2.62 6.45 11 0.66 0.09 
15 4.09 3.34 12 14.51 25.02 
16 19.97 17.33 13 2.45 0.41 
17 0.09 0.34 14 1.30 1.24 
18 0.27 1.64 15 0.01 0.00 
19 4.36 8.05 16 8.88 13.00 
20 38.64 4.97 
21 3.48 0.58 
22 0.12 0.01 
23 0.05 0.02 
  
**Note: Fresh water regularly flooded Tree Cover (7) and 
saline water regularly flooded Tree Cover (8) was not included 
in GLC2000 data over parts of China. 
3.3.2 Spatial distribution of omission of wetland: Though the 
omission of wetland in both global landcover data sets 
distributed widely and sparsely across China, it mainly 
happened in the Northeast China region, Northwest China 
region, the Tibetan Plateau, the low reaches of the Yangtze 
River and parts of the coastline. 
In Northeast China, apart from mixed forest (5) and open 
shrubland (7) in the MODI12Q1 data set, herbaceous cover 
(closed-open) (13) and cropland (16) are the most likely 
landcover types to be confused with wetland in both global data 
sets. The omission of wetland is most occurred spatially in 
Sanjiang Plain, part of Song-Nen Plain, areas along the banks 
the Hai La Er upriver, the Hulun buir Grassland, the foot of 
Greater Higgnan Mountains and Lesser Khingan Range. Errors 
in the MODI2QI data set were also found around Liaohe 
estuary in Liaoning province. Apart from the distribution of 
marshland in Northeast China, one of the great commodity 
grain bases and Three Forests Zones in China are also 
distributed alternately in this region, which may account for the 
mistakes between wetland types and cropland, forests and grass 
landcover types. 
In the northwest China, the omissions of wetlands are very 
serious and are mainly distributed in Xinjiang, Qinghai and 
parts of Inner Mongolia. The bare area (19), herbaceous cover 
(closed-open) (13), sparse herbaceous or sparse shrub cover (14) 
in the GLC2000 data set and barren or sparsely vegetated(16), 
grass land (10) and open shrubland (7) in the MOD12Q1 data 
set are the most confused landcover types, respectively. The 
confusion with the Herbaceous Cover(closed-open) (13) and 
Sparse herbaceous or sparse shrub cover (14) largely show the 
mosaic distribution of the river plains of the Tarim River and 
the Yili River, Boston Lake, and Aydingkol lake. At the same 
time, wetland types distributed along rivers in North Qinghai, 
central and west Xinjiang and parts of the Alxa regions of Inner 
Mongolia, are mistakenly classified as bare areas (19) in global 
land cover data sets. 
Except for barren or sparsely vegetated(16) and open shrubland 
(7) in MOD12Q1 data set, the herbaceous cover(closed-open) 
(13) in both global data sets is the most mistakenly interpreted 
landcover type confused with wetland in the Tibetan Plateau 
and Zoige regions in Sichuan province, where the known Zoige 
swamp and all kinds of meadows are distributed widely. In 
comparison, the confusion of wetland with farmland occurs 
mostly around inland water bodies and shorelines in East China, 
such as Poyang Lake in Jiangxi province. 
Overall, the spatial distribution of the confusion between 
MOD12Q1 data set and the reference data is similar to that of 
GLC2000 data set, which is possibly controlled by the spatial 
patterns of wetland in China. 
4 CONCLUSIONS 
To assess the suitability of existed global land cover products 
for global wetland research, the precision of wetland-related 
landcover types in two global land cover data sets, GLC2000 
and MOD12Q1, was calculated and analyzed by referencing 
China wetland maps. Some conclusions or suggestions are as 
follows: 
(1). The precision of wetland water is higher than that of