The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008 
most of the cloudy contamination (Holben, 1986). Beside that, 
particular efforts were made to the GIMMS dataset including a 
correction of varying solar zenith angles due to orbital drift of 
the afternoon NOAA satellites and a correction of volcano 
aerosols (Vermote et al., 1997) emitted by El Chichón from 
March to April .1982 and Mount Pinatubo in June 1991. It is 
reported that the GIMMS dataset has high quality (Fensholt et 
al., 2006; Tucker et al., 2005). 
G b : the difference between a peak value and the nearest 
forward wave trough. 
Based on above variables, a wave packet meeting equation (1) 
is identified as “one-cropping”. Then the remotely sensed 
cropping index (Cl) is defined as the number of peaks of the 
NDVI time-series multiplies 100% which accord with the 
definition of “once cropping”. 
80 °E got 100°E 110°E 120 “E 
i i l i I . 
1 Heilongjiang, 2inter-Mongolia, 3Xinjiang, 
4Jilin, 5Liaoning, 6Gansu, 7Hebei, 8Beijing, 
9Shanxi, lOTianjin, llQinghai, 12Shaanxi, 
13Ningxia, 14Shandong, 15Henan, 16Jiangsu, 
17Anhui 
Figure. 1 Study area 
3. METHORDS 
3.1 Definition of the NDVI Time-series Based Cropping 
Index 
Based on the fact that the growing length of the mainly crops in 
China was above 90 days, the definition of one-cropping was 
“the crop with 3 months growing length and full coverage of the 
farm” (Liu, 1997). Four parameters were used to characterize 
the wave of “once cropping” in an NDVI curve (Figure.2). 
time/(15 days) 
Figure.2 NDVI profile of crop 
A>9 
G f >Px 50% 
G b >Px 50% 
(1) 
3.2 The Method of Cropping Index extracting 
As presented in Figure.3, the cropping index extracting 
procedure includes data preprocessing {stepl, step2), false 
peaks correcting {step3, step4), and Cl extracting {step5, step6). 
| MDl^I time-series | | Cloud flag for each PVn>VI point j 
Step 1 Linear interpolation of Cloudy NDVI values 
p= ■' i.- ' i 
Step 2 | Growing trend fitting by Savitzky-Golay filter | 
Step 3 
Step 4 
Step 5 
Step 6 
Figure.3 Flowchart of cropping index extracting procedure 
Step 1: Linear interpolation of cloudy NDVI values 
In this study, cloud flag data were used to improve the NDVI 
time-series by linear interpolation of the cloudy NDVI values 
(labelled • in Figure.4(a)). 
Step 2: Growing trend fitting by Savitzky-Golay filter 
This step was done exactly as reported in Chen et al. (2004). 
Figure.4(b) shows a growing trend curve obtained using this 
filter. 
Step 3: Finding the peaks by Twi-difference Algorithm 
The twi-difference algorithm used in this step is as follows: a 
new time-series {S2} was calculated following equation (2). 
SI, = 1, if NDVI> NDVI, 
51, = -1, if NDVI i X < NDVI. 
52, = Sl,-Sl M 
Amplitude P: the difference between maximal NDVI and 
minimal NDVI of a certain pixel within one year. 
Growing length 2: the interval between the nearest two wave 
troughs in the NDVI curve. 
Gf : the difference between a peak value and the nearest 
backward wave trough. 
Where i denotes the zth point in the NDVI time-series. Where 
52 ; equals -2 was recognized as a peak, and 2 a trough.