not only very low, but also greatly different due to the various 
samples they used in validation. 
Until now, traditional manual interpretation is still the 
most effective method to assess the precision of the 
classification and it is also regarded as the highest precision 
classification method. In addition, there is no precedent 
comparison of global data set with the thematic products 
among all the research. Due to the wetland category’s 
complexity, the wetland is usually treated as other different 
types in different land cover classification schemes, and this is 
the reason why there is such a low accuracy for wetlands in all 
the above comparison research among the global land cover 
data sets. Based on the Landsat TM images across China, Niu 
et al. (Niu ef al., 2009) had completed the wetland mapping by 
manual interpretation. The two global land cover data 
sets-GLC2000 and MOD12Q1 were evaluated based on China 
wetland mapping products in this study, and the results of 
evaluation are discussed. 
2. DATA SOURCES AND PROCESSING 
2.1 Data preparation 
GLC2000 global land cover data have been produced by an 
international partnership of 30 research groups coordinated by 
the European Commission’s Joint Research Centre, based 
primarily on SPOT 4-VEGETATION daily 1-km data from 
November 1999 to December 2000(Loveland ef al., 2000). The 
global classification scheme is assigned to a LCCS land cover 
legend (Herold et al., 2008) . The MODIS land cover product 
(MOD12Q1) is based on the spectral information supplied by 
the MODIS sensor on-board Terra. All monthly inputs have 
been produced from MODIS Levels 2 and 3 data between 
November 2000 and December 2001 and include seven spectral 
bands, the enhanced vegetation index (EVI), spatial texture, 
land surface temperature, snow cover, elevation and a water 
mask (Strahler ef a/., 1999). The classification combines prior 
and posterior probabilities to assign the most probable class for 
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 
   
each location on the globe based on the IGBP classification 
scheme with 17 classes (Loveland et al, 2000). The 
classification uses a universal supervised approach with a 
multi-temporal decision tree algorithm and selects the training 
region from the high resolution image together with the 
ancillary data sets (Friedl ef al., 2002). 
The wetland map of China was produced by IRSA (Institute of 
Remote Sensing Application Chinese Academy of Sciences) 
based on completely manual interpretation with the minimum 
cartographic unit area of 9 hectares (Niu ef al., 2009). The 
wetland classification system is based on the Ramsar 
Convention and the classification of China National Forest 
Bureau during the first wetland survey between 1995 and 
2001(Gong et al., 2010). 
2.2 The crosswalk between different classification systems 
for the wetland 
The wetlands contain different land types in each global land 
cover data set because the global data sets and the reference 
data adopt different classification systems. But we cannot 
divide the available wetland types into more detailed classes so 
as to make the wetland class one-to-one correspondence among 
these different classification systems. Therefore the IGBP 
classification system that contained the least wetland-related 
types was chosen as the standard one. Then wetland types in 
the reference data set and wetland-related landcover types in 
the LCCS were converted to the IGBP system (table 1). There 
are two wetland-related types in the MOD12Q1 including 
permanent wetland and water. Permanent wetland can be 
considered as peatland to a very great extent (Pflugmacher ef 
al., 2007). In order to be distinguished from other wetland 
terms, permanent wetland was named as “peatland” and water 
was named as “wetland water” in our research. In addition, the 
paddyfields which were not included in the reference data sets 
were not assessed. 
Table 1 the crossover between different classification systems 
MOD12Q1 GLC2000 
Reference data 
  
  
IGBP legend (2/17) LCCS legend (4/22) 
the wetland of china legend (14/15) 
  
  
  
Value Class name Value Class name Value Class name 
Tree Cover, ; 
7 resulady flooded 11 Intertidal zone/Shoal/Bay 
Tree Cover, 12 Marine marshes 
P : 8 regularly flooded, 
ermanen ; 
11 wetland saline water 14 Estuarine Deltas/ | sandy 
Regularly flooded islands 
15 Shrub and/or 
Herbaceous 22 Flood wetlands 
Cover 24 Inland marshes 
13 Estuarine water 
15 Lagoons 
21 River 
23 Lakes 
0 Water Bodies 20 Water bodies 31 Reservoirs / Ponds 
32 Artificial river channels 
33 seawater fish farms/salt flats 
35 landscaping and recreational 
  
water bodies