■I 
979 
2008 
nearest 
ition (1) 
sensed 
of the 
with the 
(1) 
extracting 
, false 
, step6). 
NDVI point 
olation 
false 
:edure 
le NDVI 
7 values 
:er 
. (2004). 
using this 
nthm 
follows: a 
(2). 
(2) 
. Where 
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
Step 4: Elimination of false peaks in the NDVI curve based 
on the definition of one-cropping and counting the number 
of false peaks i N > 
X, G f , G b of each wave packet can be calculated when all the 
peaks and troughs were recognized in step 3, and then false 
peaks were identified according to the definition of “one- 
cropping” and the number of the false peaks (N) was counted .If 
N is zero, the correcting of NDVI time-series will finish 
(Figure.4(c)), and then do step5. Otherwise, replace the false 
peaks by linearly interpolated values and return to step3. 
time/(15 days) 
108 120 132 144 
time/(15 days) 
Figure.4 Data processing of NDVI time series 
Step 5: Exporting the cropping index of each pixel 
The cropping index of ith pixel is “C7, =MX 100”, where M is 
the number of “once cropping” waves. 
Step 6: Calculating cropping index in administration units 
Average Cl of certain administration unit can be calculated 
following equation (3): 
(3) 
Where n in the above expression is the number of the total 
arable land pixels in the administration unit. 
3.3 The Method of Cropping Index Variation Extracting 
Since cropping anomalies often occurred during 1982-2003, the 
Least Absolute Deviation (LAD) linear regression was 
employed instead of Ordinary Least Square (OLS) regression to 
calculate the change trend of cropping index (VCI) during 1982- 
2003. Comparing to OLS, LAD could reduce the sensitivity to 
outliers effectively and provide a robust and plausible estimate. 
The detailed information about LAD regression can be found in 
Powell (1984). 
4. RESULTS AND DISCUSSION 
4.1 Precision Evaluation 
Precision evaluation was carried out as the comparison between 
the Cl extracted trough our new method and that calculated 
from statistical data at province scale. As mentioned above, 
cropping index is defined as the ratio of the total seeding area to 
the arable land area. Annual seeding area of every year during 
1982 to 2003 is available in “China Agriculture Information 
Net” (http://www.agri.gov.cn/sjzl/ nongyety.htm). However, 
there is no available data about arable land area of each year, so 
the arable land area in 1996 could be approximately regarded as 
the arable land area from 1996 to 2003 (also available in China 
Agriculture Information Net). Based on the statistical data 
above, the annual Cl of each province was calculated. 
Generally, the remotely sensed cropping index shows high 
accordance with statistical data at province scale (R 2 =0.9213, 
/*<0.001, slope= 1.0775) (Figure.5), suggesting the reliability of 
the proposed method. 
statistical Cl 
Figure. 5 Correlation of remotely sensed Cl and statistical Cl 
4.2 Spatial Pattern of Cropping Index 
The 22-year average Cl varied evidently among different 
provinces (Table 1), with Hebei, Shandong, Henan, Jiangsu, 
Anhui exhibiting high values, and Heilongjiang, inter-Mongolia, 
Xinjiang, Gansu, Shanxi, Qinghai, Ningxia exhibiting relatively 
lower values. As presented in Figure.6, the cropping index 
shows an increasing trend from northeastern and northwestern 
provinces (about 100) to southeastern ones (about 200). The 
spatial distribution of cropping index extracted by the newly 
proposed method was consistent with the actual Chinese 
cropping system reported by Shen et al. (1983). 
Province 
Average cropping index from 1982 to 
2003 
Heilongjiang 
88.99 
inter- 
73.61 
Xinjiang 
79.56 
Jilin 
90.43 
Liaoning 
91.88 
Gansu 
83.04 
Hebei 
123.34 
Beijing 
107.99 
Shanxi 
87.13 
Tianjin 
104.68 
Qinghai 
77.07 
Shaanxi 
95.32 
Ningxia 
73.78