MONITORING OF LAND USE IN THE MAJOR GRAIN CROP PRODUCTION AREA 
USING TEMPORAL CHARACTERISTICS OF NDVI 
OBTAINED FROM COARSE RESOLUTION SATELLITE DATA 
S. Uchida 
Development Research Division, Japan International Research Center for Agricultural Sciences 
305-8686 Tsukuba, Ibaraki, Japan — uchidas(@jircas.affre.go.jp 
Commission VII, WG VII/2 
KEY WORDS: Agriculture, Land Use, Vegetation, Estimation, Monitoring, Pixel, Multitemporal 
ABSTRACT: 
NDVI obtained from satellite at the pixel over the area of crop cultivation would represent a specific temporal profile, which could 
be discriminated from that of other land use. In this study, the author attempted to estimate the acreage of winter wheat sown area 
using multitemporal NDVI obtained from NOAA-AVHRR data. The study site was selected in the Huang-Huai-Hai plain in China, 
where the most popular cropping pattern was double cropping of winter wheat and maize. First, the author examined a linear 
unmixing method, of which the value of NDVI at pixel was assumed as a composition of end-member values of NDVI in proportion 
to land use components. This method could be adopted if there was a high spatial resolution data taken in the appropriate period to 
estimate end-member values, and could be applied to the area where cropping season was almost resonant in the wide area. Secondly, 
he examined a relation of the ratio of winter wheat sown area in a pixel from the values of 2-temporal NDVI, which could show the 
relation with the components of land use. Estimated acreage by the either method showed a compatible spatial distribution with the 
acreage obtained from LANDSAT-ETM+ data. The advantageous point of the second method was that it would not require a high 
spatial resolution data for each year, supposed that the consistent parameters were adopted. By applying this method, the author 
could characterize the recent changes of winter wheat sown area in the northern part of the Huang-Huai-Hai plain. 
1. INTRODUCTION 
1.1 Background 
Satellite remote sensing has been expected as an efficient 
technology to monitor of agricultural land use, which could be 
changed depending on the various factors, for a wide area. 
Especially it should be useful for the area where the accurate 
statistics data was hardly obtained. Multi-spectral and 
adequately high spatial resolution sensor data showed enough 
capability to discriminate the sown arca of major crops, 
however, the timing of obtaining good quality data for this 
sensor was not optimized to detect the coverage of a specific 
crop. On the other hand, the spatial resolution of high temporal 
resolution sensor data was coarse compared with the unit of lot 
size and the pixel-based classification of it might cause false 
estimation of acreage of each land use type. 
Many researches were implemented to resolve the components 
within a pixel. They often employed the assumption that the 
spectral characteristics of pixel were represented by a linear 
combination of end-member spectral values of each pure 
component. This unmixing technique could be effective for the 
case when each end-member values showed physically 
distinctive characteristics. But for the case of agricultural land 
use discrimination, the category defined by the crop type might 
be confused with other categories containing vegetation cover. 
From this context, the analysis of temporal pattern of the 
specific index such as NDVI, which shows the distinctive 
characteristics by land use types, may have a kind of possibility 
to estimate the percentage of sown area of the specific crop in a 
pixel. 
138 
1.2 Objectives 
The objectives were firstly to develop a method to estimate the 
percentage of sown area of the major crop in a pixel of coarse 
resolution satellite data, and secondly to characterize the spatial 
and temporal features of cultivation of the major crop estimated 
by the method developed in this study. The author examined the 
temporal profile of NDVI to develop a method and applied it to 
estimate the winter wheat sown area in the major crop 
production area of China. 
2. MATERIALS AND METHOD 
2.1 Data 
The author obtained the 10-day maximum composite NDVI of 
NOAA-AVHRR for the period from 1994 to 2003 provided by 
SIDaB (Satellite Image Database System), which is operated by 
the Research Information Center, Ministry of Agriculture, 
Forestry and Fisheries of the government of Japan. He also 
collected 3 scenes of LANDSAT-ETM+ data (Path: 123, Row: 
32-34) taken on 19 May, 2001. 
2.2 Study Site 
The Huang-Huai-Hai plain is located in the castern part of 
China and selected as the study site (Figure 1). The dominant 
cropping pattern in this area is winter wheat and maize in 
summer, so that winter wheat is popularly cultivated except in 
the urban areas as well the mountainous areas. Winter wheat is 
generally seeded in October, flowered in May and harvested in 
June. Although the distance from the south to the north of plain 
Internatio 
Ure ao 
reaches al 
as small a: 
  
Figui 
2.3 Meth 
2.3... Li 
lies on line 
expressed 
where po) 
In 
This form 
This meth« 
From the t 
for each I: 
atmospher 
growing ct 
The end-n 
wholly oc 
estimated 
from LAN 
the end-mc 
used for of 
similar. 
In order to 
first LAN 
likelihood 
objective | 
drew a line 
against NI 
extrapolate 
à linear mi 
probability 
addition of 
density of 
one as sum 
232 2-4 
the feature 
flowering