International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004 
  
  
Inter 
— 
NDVI ann amp and  NDVI ann max are applied to undi 
deciduous grass type vegetation land to separate single agn 
cropped agricultural fields from natural grassland. These two Areas of each land cover category are compared with the Tofa 
categories have quite similar spectral and phenological existing land use map “1 KM Grid Data of landuse and com 
characteristics. Therefore, this criterion (if NDVI_ann_amp is landcover of China” (abbreviated as CASW data). This map disc 
higher than threshold and NDVI_ann_max is higher than is about | km grid data of land use and land cover of China gras 
threshold, then it is categorized as single cropped in 1996. The value of a grid in every land use layer is a of p 
agricultural field) has no supportive evidence and it is percent. The percent is about the real area of the land use utili 
derived fully empirically by trial and error approach. Nine type in this gird over | km? area and it is from the statistic of vas 
land cover categories (water, non-vegetation,less-vegetation, ^ the area attribute of vector data of land use of China at mod 
evergreen forest, evergreen grassland, deciduous forest, 1:100000 (CASW, 2001). The areas of five land use impl 
deciduous grassland, single cropped agricultural field, and categories (forest, agricultural field, grassland, less- reso. 
double cropped agricultural field) are established from the vegetation, and non-vegetation) are figured out for each and 
viewpoints of the parameterization of the water balance province by aggregation of the grids and appropriate and 
model used in this study, and the accuracy of the land cover categories of each map. The results are shown in Figure 4. valu 
classification. In general, parameterization of the land 
surface properties in the physical model is expected to be 30 3s 
difficult, and the accuracy of the land cover classification 3 , 5 + 
will be deteriorate, along with the increase in the class 85 às The 
categories. Therefore minimum number of the categories is 20 . 4 T . ment 
established in this stage, and category might be added 15 20 . MOL 
hierarchically with a focus on outcomes from the model. * t se ee io ce 
5 2 : i &' valu 
3. RESULTS AND DISCUSSION ra T ene 
0 5 10 151 202725 30 0 $ 1013552201 25^ 30!:35 
The land cover classification map is shown in Figure 3. The (a) Forest (b) Agricultural field b 
threshold values are determined manually using image but 
processing software. According to the visual interpretation, ii 9 reco 
non-vegetation, less-vegetation, evergreen forest, and 25 ° mpl 
double cropped agricultural field have relatively high B . 20 ext 
agreements with existing land cover maps and satellite . diffi 
2 : : A 10 15 field 
based classification maps, though the categories of land e *. 
cover are not fully corresponded each other. These categories ° ut o . $ 
have distinct characteristics in NDVI or reflectance. On the "e gi? 
: : : : $9 v e Refe 
other hand, single cropped agricultural field is not o 8** oai dese Fried 
sufficiently discriminated from grassland, and it is 0 5 10 T 20 à. 5 10 is 20° 25 3 
overestimated especially in the northern region from eastern s i Cove 
Siberia to northeastern China, and southern region around (c) Grassland (D Less veretation Emm 
Xizang (Tibet), Sichuan, and Qinghai provinces. In these 30 40 Fried 
regions, the multi-temporal metrics of grassland and 25 MOD 
agricultural field used in this study show the indistinctive a a 30 e ; Envil 
characteristics as the result of the similarity in the spectral 
reflectance in two bands, thus NDVI, and vegetation ts e 20 : .? Liu, . 
phenology. 10 * PELA integ 
5 18, A Inter 
; - o p? 2500 
0 5 1051529 25 30 0 10 20 30 40 
Pal, ? 
(e) Non-vegetation (f) Agricultural + Grassland decis 
Sensi 
Figure 4. Comparison of land use area with CASW data 
(horizontal: CASW, vertical: this study, unit: 10000km?) Pan, 
in CI 
indic 
The good correlations are shown in forest and non- Remo 
vegetation, though classification result has the tendencies 
to overestimate relative to the CASW data. In case of Runn 
Figure 3. Land cover classification map agricultural field, most provinces where there are small areas veget 
(legend from top to bottom: evergreen forest, deciduous of agricultural field are resulted in underestimations. In analy 
forest, evergreen grassland, deciduous grassland, single contrast, grassland shows countertrend, that is, provinces Ref 
er 
cropped agricultural field, double cropped agricultural field, 
less-vegetation, non-vegetation, and water) 
that have small grassland areas have trends of 
overestimation, but in provinces with larger grassland, 
838