DETECTING CROP ROTATIONS IN CHINA 
USING AVHRR IMAGERY AND ANCILLARY DATA 
J.K. Guo ?, J. Y. Liu *, G.M. Huang ”, D.F. Zhuang *, H.M. Yan *, G.P. Zhang* 
a Institute of Geographical Science and Natural Resources Research, CAS, 11A Datun Road, Anwai, Beijing, 100101, China- 
(Guojk, zhuangdf, yanhm)@lreis.cn, liujy@igsnrr.ac.cn 
b Chinese Academy of Surveying and Mapping, Beitaiping Road, No.16, HaiDian, Beijing, 100039, China - 
huang.guoman@163.net 
c National Meteorological Center, Zhongguancun South Street, 46, HaiDian, Beijing, 100081, China - zhanggp@cma.gov.cn 
KEY WORDS: Land use, Crop, Detection, Classification, Imagery, Multitemporal 
ABSTRACT: 
The situation of cropland use in China is very complicated. In many areas, the cropland is used in multi-cropped ways. There is a 
need for better information on the area and distribution of cropland using in different cropping rotation systems, but it is not easy to 
get it in traditional census ways. This paper focuses on the methodology of crop rotations detection in China using multitemporal 
satellites images. Two agricultural regions located in the middle of China were chosen as the study areas. The dataset used here 
includes 10 days composites NDVI (36 periods) obtained from the NASA Pathfinder AVHRR Land dataset, land-cover dataset 
derived from TM images, and the ground based agricultural monitoring data. The discrete Fourier transform was applied to the 
NDVI data set on a per pixel basis for the whole cropland of the study areas and then the additive and the first four harmonics 
(amplitude and phase) were classified using ISOLATE unsupervised classification algorithm for both regions respectively. Crop 
information derived from local stations and the Chinese cultivated system regionalization map were used to assess the accuracy of 
the result. The result of this study showed that the methodology used in this study is, in general, feasible for detecting crop rotations 
  
in China. 
1. INTRODUCTION 
The situation of cropland use in China is very complicated 
because of huge population and limited cropland. Roughly half 
of the cropland in China is used in multi-cropped ways—the 
sequential cultivation of an ordered succession of crops on the 
same land in a year. The information about the crop rotations 
in China is very important for assessments of the potential of 
food production and the impact of multi-cropping on 
biogeochemical cycling of carbon and nitrogen in 
agroecosystems. There is a need for better information on’ the 
area and distribution of cropland using in different cropping 
rotation systems but it is not easy to get it in traditional census 
ways. Remote Sensing has been shown to be very useful for 
analysis and mapping crop rotations (Panigrahy, 1997; Qiu, 
2003). 
Time series of Normalized Difference Vegetation Index (NDVI) 
have been used widely in studies on land cover characteristics 
and vegetation phenology because of its correlation with green 
plant biomass and vegetation cover. The cropland used in 
different kinds of crop rotations exhibit distinctive patterns of 
NDVI variation in a year so it is feasible to derive the crop 
rotation information from analysis the temporal change in 
NDVI values. T 
Some methods have been used to obtain the characteristics of 
NDVI time series such as signal decomposition (Moody, 2001) 
and key NDVI metrics analysis (Reed, 1994). Discrete Fourier 
transform (DFT) as a signal decomposition method can extract 
periodic responses through expressing a NDVI time series 
curve with the sum of an additive term and a series of 
sinusoidal waves and it has been used in analysis and 
evaluation land surface dynamics and phonologies (Moody, 
2001; Olsson, 1994), land cover classification (Andres, 1994), 
crop type identification (Jakubauskas, 2001, 2002), and 
mapping agroecological zones or producing bioclimatological 
regionalization (Menenti, 1993; Azzali, 2000). It has been 
shown that the Fourier harmonics are sensitive to seasonal 
variability in vegetation productivity and discrete Fourier 
analysis is an objective and concise summarization of the 
temporal signature that is sensitive to systematic changes in 
vegetation (Andres, 1994; Moody, 2001). 
In this study we detected the situation of crop rotations in two 
of nine agricultural regions in China using discrete Fourier 
analysis and unsupervised classification approach with 
AVHRR time series NDVI and ancillary data. 
2. DATA USED 
2.1 Study Areas 
In this paper, Yellow-Huai-Hai Rivers’ Region (HH) and Loess 
Plateau Region (HT), two of the nine agricultural regions based 
on the Chinese general agricultural regionalization map (Sun, 
1994) were used for the study (Fig.1). Both regions are located 
in the middle of China and cover the total or part of 12 
provinces and metropolis. There is a mix of single and double- 
cropping in these areas. Single cropped winter or spring wheat, 
spring maize, cotton, rapeseed and double cropped winter 
wheat and summer maize (or cotton, rice) are the main crop 
rotations types in these two regions. 
2.2 Data Used 
The dataset used in this study includes 10 days composites 
AVHRR time series NDVI (36 periods), land-cover dataset 
234 
  
Internati 
obtains 
region: 
region: 
phonol 
A time 
AVHR 
technic 
The da 
Land d 
the twc 
in. IS 
corres 
instead 
in 199 
Goode' 
to Albe 
researc 
Croplai 
asam 
areas. 
percent 
1999/2 
(Liu, 1¢ 
The grc 
local s 
phonok 
area we 
regiona 
commit 
the resu 
Figure 
3.1 Dis 
Discrete 
method