Mathematical and Computer Modelling 54 (2011) 1037–1043
Contents lists available at ScienceDirect
Mathematical and Computer Modelling
journal homepage: www.elsevier.com/locate/mcm
Mapping rice planting areas in southern China using the China
Environment Satellite data
Jinsong Chen a,b , Jianxi Huang c,∗ , Jinxing Hu a
a
Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, China
b
Institute of Space and Earth Information Science, The Chinese University of Hong Kong, Hong Kong
c
College of Information and Electrical Engineering, China Agricultural University, Beijing, China
article
info
Article history:
Received 13 August 2010
Accepted 4 November 2010
Keywords:
China Environment Satellite HJ-1A and B
Rice planting area
Normalized difference vegetation index
abstract
The objective of this research is to investigate the potential of application of China
Environment Satellite HJ-1A/B in monitoring rice cultivation areas in Guangdong province
in southern China. Information on the rice cultivation areas is of global economic and
environmental significance. A CCD camera sensor with 30 m spatial resolution onboard
China Environment Satellite HJ-1A and B has visible and near infrared bands and a revisit
period of four days; the temporal Normalized Difference Vegetation Index (NDVI) can
therefore be obtained from HJ-1A and B data. The characteristics of the temporal NDVI
derived from HJ-1A and B images of rice fields and other crops at rice growth stages
in the western part of Guangdong province of China with an area of about 67000 km2
were first analyzed in this research and an algorithm for mapping paddy rice fields was
developed based on the temporal changes of NDVI of rice fields from January to July, 2009.
The mapping result was evaluated by field survey and the data from China Ministry of
Agriculture and the promising accuracy was found with a Kappa factor of 0.71. The result of
this study suggests that the China Environment Satellite HJ-1A/B has great potential in the
development of an operational system for monitoring rice crop growth in southern China.
Crown Copyright © 2010 Published by Elsevier Ltd. All rights reserved.
1. Introduction
Rice is the most important primary food in Asia. It constitutes the foundation of the economy of many Asian countries
and provides staple food of the people. Rapid population growth puts increasing pressure on the already strained food
supply. The increasingly growing price of rice has put great negative impact on the life of the people in recent years. Rice
accounts for more than 42% of the crop yield in China. Its cultivation is strongly related to social stability and economic
sustainable development for China. Guangdong province is one of the most economically developed regions in China. With
the development of infrastructure and the change of agricultural management, the cropping system in this region has
changed rapidly with the tendency of the greater diversification (i.e., increasing the variety of crops planted) and more
frequent changes of crop cultivation, which reflects the farmers’ strategies in adapting their cropping practices to the change
of market demand. Regarding the global environment aspect, the knowledge of rice cultivation areas is important to estimate
the fluxes of methane (CH4 ) from irrigated rice fields to the atmosphere [1,2]. So the above socio-economic and global
environmental factors have put forward a strong demand on timely and effective operational system for monitoring rice
plantation areas and growth conditions.
∗
Corresponding author. Tel.: +86 13691565698.
E-mail address: [email protected] (J. Huang).
0895-7177/$ – see front matter Crown Copyright © 2010 Published by Elsevier Ltd. All rights reserved.
doi:10.1016/j.mcm.2010.11.033
1038
J. Chen et al. / Mathematical and Computer Modelling 54 (2011) 1037–1043
Table 1
HJ-1A/B main characteristics.
Satellite
Sensor
Band range
CCD camera
B01
B02
B03
B04
0.43–0.52
0.52–0.60
0.63–0.69
0.76–0.90
Hyperspectral imager
–
0.45–0.95
(110–128 bands)
CCD camera
B01
B02
B03
B04
0.43–0.52
0.52–0.60
0.63–0.69
0.76–0.9
Infrared multispectral camera
B05
B06
B07
B08
0.75–1.10
1.55–1.75
3.50–3.90
10.5–12.5
HJ-1A
HJ-1B
Spatial resolution (m)
Swath width (km)
Repetition cycle (day)
30
30
30
30
700
4
100
50
4
30
30
30
30
700
4
720
4
150
300 (10.5–2.5 µm)
Satellite remote sensing has been widely used in crop monitoring program for the past several decades because it can
provide effective and timely spatial and temporal information on crop planting areas and growth conditions. Some studies
have used optical satellite remote sensing data, such as NOAA Advanced Very High Resolution Radiometer (AVHRR), Landsat
Thematic Mapper (TM) data and SPOT data to identify paddy rice fields [3–5]. Those studies were conducted based on
the difference of temporal changes of the Normalized Difference Vegetation Index (NDVI) and spectral features of rice
fields from other crops. Moderate Resolution Imaging Spectroradiometer (MODIS) onboard the Terra and Aqua satellites
and VEGETATION (VGT) onboard the SPOT-4 satellite can provide additional shortwave infrared bands (SWIR, 1580–1750
nm) that are sensitive to vegetation moisture and soil water, which can provide an opportunity for developing improved
vegetation indices that are sensitive to equivalent water thickness (EWT, g H2 O/m2 ), such as the Land Surface Water Index
(LSWI) [6–8]. Some research have been carried out to map rice fields in China using temporal profiles of LSWI and NDVI
data derived from MODIS and VEGETATION data based on the sensitivity of LSWI to flooded rice fields at the transplanting
period [9,10].
China Environment Satellite HJ-1A and B were two remote sensing satellites with CCD camera, hyperspectral imaging
device and infrared multispectral camera. The two satellites have similar orbit and are part of future constellation of China
Environment Satellite. They were launched in September 2008, with the aim to monitor environment and mitigate disaster.
The CCD camera onboard them can provide remote sensing data with four bands from visible to near infrared. Its main
characteristics can be seen in Table 1. Compared with other often used optical remote sensing data, CCD camera onboard
HJ-1A and B has a better spatial resolution of 30 m than MODIS and SPOT VEGETATION, which make it more suitable to
map rice fields with relatively small areas like in Guangdong province. HJ-1A and B has a higher temporal resolution of four
days and bigger imaging swath of 700 km than TM and SPOT. These features make it possible to obtain optical more remote
sensing data during key periods of rice growth and the less number of images to cover the test site.
The objective of this study is to explore the potential of HJ-1A and B data for the development of an operational system for
mapping rice fields. The western part of Guangdong province in southern China was selected as the case study area because
rice has been planted for more than forty years and it is one of the key bases of rice monitoring by the Agriculture Ministry
of China. In this study, the temporal NDVI of rice fields and other crops were first analyzed and an algorithm was developed
to identify rice fields using threshold segment instead of classification. The result was evaluated using field survey and data
from Ministry of Agriculture.
2. Study area and remote sensing data
2.1. Study and rice calendar
Guangdong province is situated in the southern part of China. The province belongs to eastern Asia monsoon region with
an annual average temperature of 22.3° and annual precipitation of 1777 mm. The province has a total area of 177,316 km2 ,
and cropland area of 44,933 km2 , among which rice fields account for 19,873 km2 according to 2008 statistical data of local
agriculture department. The study area in this research is located in the western part of Guangdong province and with an
area of about 65,780 km2 . Fig. 1 shows the location of the study area. The site belongs to tropical climate annual precipitation
of about 2100 mm and an average elevation of 180 m [10]. The reason for selecting it as the study area is that the crop land
distribution has been updated in 2008 by local agriculture department and the mapping result can be evaluated using the
data.
Paddy rice has been planted in the area for more than forty years. The rice is cultivated with rotation system of early
season rice and late season rice per year. There are five major growth periods in the life cycles. (1) Transplanting period:
J. Chen et al. / Mathematical and Computer Modelling 54 (2011) 1037–1043
1039
Fig. 1. The location of study area.
Fig. 2. HJ-1A image of the study area on March 12 (red: band 3, green: band 4; blue: band 1).
rice plant seedlings are transplanted from the seedbed to the paddy field. The transplanting date depends on the weather,
especially on the temperature. (2) Seedling developing period: the seedling splits up and begins to develop a root system.
(3) Ear differentiation period. (4) Heading period: headings begin to form. (5) Mature period: the rice plants mature and are
ready to be harvested. Temporally, these five periods for early season rice are usually February 20–March 15, April 15–30,
May 10–30, June 10–25 and July 5–31 per year. For late season rice, the growth stages occur as follows: July 20–August
5, August 10–20, September 1–30, October 1–20 and November 1–25, respectively. These dates are subject to adjustments
according to weather conditions. The other major ground categories within the study area are building, mountain forest, fish
pond, sugar cane, banana, fruit trees and other kinds of economic crop. This year early rice in the study area was transplanted
from February 21 to March 20. So March 10 was set as the beginning date of rice growth in this research.
2.2. Remote sensing data
Usually late rice is harvested from late December to mid-January and then the rice field would be laid farrow till the
beginning of the transplanting period of early rice. Some parts of rice fields may be changed to plant other crops or fruit
tree to adapt to market demand. This research focus on mapping early rice fields in the study area. Based on early rice
calendar, HJ-1A/B data from February 4 to August 10 over the study area were collected at an interval of four days. This
period covers the whole growth stage of early rice from before transplanting to after harvest. The images were geo-rectified
and registered. The calibration and cloud removal were made by Satellite Environment Center. In this study 21 HJ-1A/B
images with relatively less cloud were selected and were processed to reduce cloud effect and these time series of images
were used for studying the spectral characteristics of rice. Fig. 2 shows HJ-1A image of the study area on March 12.
3. Mapping rice fields in the study area
3.1. Analysis of vegetation index of rice fields and other crops
A number of studies have shown that vegetation index (VI) can be used as an indicator of the status of vegetation and
crops [11]. There are many vegetation indices, such as NDVI, EVI (enhanced vegetation index) and SAVI (Soil-Adjusted
Vegetation Index), which have been used to discriminate crops and monitor crop growth conditions [12–14]. Among these
vegetation indices, the NDVI is the most widely used VI and other indices are its refined form. There is a consensus that
the NDVI can be used as an effective measure of photosynthetically active biomass and chlorophyll activity of vegetation
and crop. In addition to serve as indication of the ‘greenness’ of the vegetation, the NDVI is also able to offer valuable
1040
J. Chen et al. / Mathematical and Computer Modelling 54 (2011) 1037–1043
Fig. 3. Water content of rice canopy in growth stages.
information on the dynamic changes of specific crop species. Therefore, the NDVI can also be employed to reflect phenology
and periodically dynamic changes of crop groups and estimate crop yield [15,16]. The principle of applying NDVI in crop
mapping is that crop is highly reflective in the near infrared and highly absorptive in the visible red. The contrast between
these channels can be used as an indicator of the status of the vegetation and crop. Consider the other vegetation indices
have different calculation based on certain crop and soil conditions, the NDVI is employed to analyze temporal spectral
dynamics of rice and other crops in the study area. The NDVI can be calculated as following:
NDVI =
Rnir − Rred
Rnir + Rred
(1)
where Rnir and Rred represent reflectance of near infrared and red bands, respectively, which correspond to fourth and third
bands for HJ-1A/B images.
After the period of transplanting rice from twenty plots of rice fields with an area of 100 m × 100 m in the study area with
the transplanting date difference of 8–15 days, some samples of other ground types including fish ponds, forests, shrubs,
grasslands and buildings, and other crops with similar planting dates such as early rice, including sugarcane and some
vegetables, were selected in this study. The locations of these samples were obtained by local agriculture department using
GPS and their corresponding positions in temporal series of HJ-1A/B data can also be located. The height, water content and
leaf area index (LAI) of rice were measured and averaged around dates of acquisition of remote sensing data. Figs. 3 and
4 shows the temporal changes of water content and the height of rice plants. The temporal dynamics of NDVI of samples
of rice fields and other types from selected HJ-1A/B images were calculated by Eq. (1) and the average NDVI values of the
samples at different rice growth stages are showed in Fig. 5.
As shown in Fig. 5, the temporal NDVI profiles from HJ-1A/B reflect the emergence, growth and senescence of rice
plants and growth conditions of other crops. Before transplanting, NDVI has low values of 0.11 for bare soil. During the
transplanting period, rice fields are flooded, which makes the NDVI value decrease to about 0.02 on March 20 about 10
days after transplanting. After transplanting, the NDVI value increases rapidly with the growth of rice and reaches a peak of
0.73 during the ear differentiation period on May 12 about 63 days after transplanting. The NDVI value decreases after the
heading period with decrease in water content and photosynthesis activities. After harvest in August, rice fields turned into
bare soil with a NDVI value of 0.09 on August 10. For other ground types, the NDVI of water and fish pond varied little over
the whole period of rice growth with values of about −0.43 and −0.26 because of the absorption in near infrared. Evergreen
forest and banana canopy have a high NDVI value of around 0.77 and 0.59. Some sugarcane and fruits was planted at the
similar period of rice transplanting, but they have a longer life cycle than rice so that their NDVI values remain after rice
harvest.
3.2. Rice fields mapping algorithms
The key to mapping rice fields is to find proper remote sensing data combinations to maximize the temporal variation
of rice fields and other types. There are many classification methods such as MLC and decision tree which have been used
in mapping crop lands. The above temporal NDVI profiles can suggest that the NDVI can be used as a possible factor to
discriminate rice fields from other ground types in the study area through their unique phenology and dynamic profiles of
NDVI. The analysis of the NDVI profiles shows that the important time for early rice monitoring is at the beginning of the
transplanting period in March and the heading period in May, and the heading period and just after the harvest period in
August. The NDVI values at these dates can give the biggest difference between rice fields and other ground types. Based on
the above analysis, an algorithm for mapping rice fields was proposed as following.
1. First, HJ-1A/B images are processed for removing cloudy areas. Because the precipitation from April to September
accounts for over 80% of annual precipitation in the study area, and each HJ-1 A/B image has a large imaging swath,
J. Chen et al. / Mathematical and Computer Modelling 54 (2011) 1037–1043
1041
Fig. 4. Heights of rice plants in growth stages.
Fig. 5. Temporal variation of NDVI of rice fields and other ground types in the study area.
2.
3.
4.
5.
6.
many images collected were contaminated by a large amount of clouds. There are some cloud removal methods which
have been used for TM, MODIS and SPOT images. The information on clouds in HJ-1A/B can be extracted based on its high
reflectance in blue band (first band of HJ-1A/B data). The threshold ranges from 0.19 to 0.23 based on the distribution
of ground types in the images. After detecting cloudy areas, masks can be made to exclude cloudy areas from the next
analysis or replace the DN values of these areas with average values of ground types.
After cloud removal, the temporal series NDVI images derived from HJ-1A/B from February 4 to August 10 are obtained
by using Eq. (1).
Masks are made to exclude water areas and fish pond with NDVI values below 0 in the NDVI images.
Buildings and bare soil are detected and excluded using masks with the NDVI values about 0.12 in the NDVI images.
After above steps, two NDVI difference images between NDVI image on March 12 and May 10 and between May 10 and
August 10 are made. Mask with value less than 0.12 in the two difference images is obtained to exclude evergreen forest
and banana area.
A threshold with the value of 0.62 is decided based on the assumption of normal distribution of DN values of HJ-1A/B
images and the spectral features of rice fields over the whole life cycle. The threshold is set to separate rice fields from
sugarcane and other vegetation types and fruit fields. The pixels with value greater than 0.65 in the two different images
are identified as rice fields. After this step, early rice fields will be obtained.
4. Results and evaluation
Fig. 6 shows the mapping result of paddy rice in the study area derived from HJ-1A/B data in 2009. Rice fields are found
distributed throughout the study area with an area of about 2400 km2 , with the exception of central part with average
altitude over 300 m. Rice fields were concentrated in coastal regions with lower elevation less than 100 m. The spatial
pattern of paddy rice from HJ-1A/B is generally similar to that from MODIS and NLCD-2000 reference. Accuracy assessment
of the mapping result is a challenging task for many reasons. With the development of infrastructure and the change of
1042
J. Chen et al. / Mathematical and Computer Modelling 54 (2011) 1037–1043
Fig. 6. The distribution of rice fields in the study area from HJ-1A/B data.
agricultural management, the cropping system in this region has changed rapidly. It is difficult to conduct labor intensive
field surveys to validate the mapping result in such a large area because of limitations of human forces and budget. In this
study, rice field distribution map edited by local agriculture departments in 2007 is employed as ancillary data to evaluate
the mapping result from HJ-1A/B as an alternative method. The assessment consisted of three components: (1) a sampling
design for selecting accuracy assessment samples; (2) an interpretation of ancillary map for collecting reference data; and
(3) an analysis for obtaining accuracy estimates. A stratified random sampling was adopted for collecting all samples. A
total of 920 independent plots were sampled using stratified method throughout the study area. The traditional error
matrix approach is used to provide information on the accuracy of the mapping result as applied to an independent set of
observations. Information in the error matrix may be evaluated using simple descriptive statistics, such as overall accuracy,
producer’s accuracy, and user’s accuracy, or multivariate analytical statistical techniques. A better index to determine map
accuracy from confusion matrices is the Kappa index, which compares the agreement against that which might be expected
by chance [17]. Kappa index is calculated from error matrix to assess the accuracy of the mapping result from HJ_1A/B. The
results showed that the mapping results of HJ-1A/B have the Kappa index of 0.69, which can be considered to be a promising
agreement with ancillary data. This confirms that HJ-1A/B data has great potential in mapping rice fields.
5. Discussion
This paper shows the preliminary result of mapping rice cultivation areas in the western part of Guangdong province
of China using HJ-1A/B data of higher temporal resolution and wider imaging swath based on temporal profile analysis of
vegetation indices retrieved from HJ-1A/B data. The experiment takes advantage of the difference of spectral features of rice
fields from other ground types to identify rice fields. A method for mapping early rice fields is proposed using the data set
of HJ-1A/B in March at the beginning of the growth of early rice, in May during the ear differentiation period and in August
after harvest with an accuracy of 69%. The study also indicates that some factors could affect the efficiency and accuracy
of mapping paddy rice fields using HJ-1A/B data, among which the most important is atmospheric condition. The study
area is located in rainy and cloudy areas. Residual clouds can seriously affect the quality of HJ-1A/B data. Development of
effective cloud removal methods and synergistic use of synthetic aperture radar (SAR) data which is independent of weather
conditions and has all day imaging capability, may be a solution to this problem. So far, some SAR data can be obtained with
low cost and some experiments have been carried out to map rice fields using SAR data [18]. HJ-1A/B data with higher
temporal resolution could help to design a nearly synchronic image acquisition plan of SAR images. Another important
factor is the drastic variation of crop cultivation in the study area. Some non-rice fields have the similar temporal change of
NDVI as rice fields, which can affect the accuracy of the mapping result. Robust algorithms need to be developed in future
study. In summary, the result suggests that HJ-1A/B appear as a promising remote sensing data in developing an operational
system for monitoring rice crop growth in southern China. The data can potentially be used to monitor paddy rice growth
in Asia, where paddy rice is cultivated on a large scale.
Acknowledgements
The work described in this paper is funded by the Hong Kong Research Grants Council CUHK461907, the Social Science
and Education Panel Direct Grant of Chinese university of Hong Kong and the National Natural Science Foundation Project
J. Chen et al. / Mathematical and Computer Modelling 54 (2011) 1037–1043
1043
of China (No. 40901161). The authors wish to take this opportunity to express their sincere acknowledgment to Satellite
Environment Center, Ministry of Environmental Protection of China.
References
[1] H.A.C. Denier van der Gon, Changes in CH4 emission from rice fields from 1960s to 1990s: 1. Impacts of modern rice technology, Global Biogeochemical
Cycles 14 (1) (2000) 61–72.
[2] I. Aselmann, P.J. Crutzen, Global distribution of natural freshwater wetlands and rice paddies, their net primary productivity, seasonality and possible
methane emissions, Journal of Atmospheric Chemistry 8 (4) (1989) 307–358.
[3] H. Fang, Rice crop area estimation of an administrative division in China using remote sensing data, International Journal of Remote Sensing 17 (1998)
3411–3419.
[4] K. Okamoto, H. Kawashima, Estimating of rice-planted area in the tropical zone using a combination of optical and microwave satellite sensor data,
International Journal of Remote Sensing 5 (1999) 1045–1048.
[5] T.G. Van Niel, T.R. McVicar, H. Fang, S. Liang, Calculating environmental moisture for per-field discrimination of rice crops, International Journal of
Remote Sensing 24 (4) (2003) 885–890.
[6] M. Makia, M. lshiahrab, M. Tamura, Estimation of leaf water status to monitor the risk of forest fires by using remotely sensed data, Remote Sensing
of Environment 90 (2004) 441–450.
[7] X. Xiao, S. Boles, S. Frolking, W. Salas, B. Moore, C. Li, et al., Landscape-scale characterization of cropland in China using vegetation and landsat TM
images, International Journal of Remote Sensing 23 (2002) 3579–3594.
[8] X. Xiao, S. Boles, S. Frolking, W. Salas, B. Moore, C. Li, et al., Observation of flooding and rice transplanting of paddy rice fields at the site to landscape
scales in China using VEGETATION sensor data, International Journal of Remote Sensing 23 (2002) 3009–3022.
[9] D. Bachelet, Rice paddy inventory in a few provinces of China using AVHRR data, Geocarto International 10 (1995) 23–38.
[10] X. Xiao, T. Boles, J. Liu, D. Zhuang, S. Frolking, C. Li, W. Salas, B. Moore, Mapping paddy rice agriculture in southern China using multi-temporal MODIS
images, Remote Sensing of Environment 90 (4) (2005) 480–492.
[11] R.B. Myneni, F.G. Hall, P.J. Sellers, et al., The interpretation of spectral vegetation indexes, IEEE Transactions on Geoscience and Remote Sensing 33
(1995) 481–486.
[12] S.N. Goward, B. Markham, D. Dye, W. Dulaney, J. Yang, Normalized difference vegetation index measurements from the advances very high resolution
radiometer, Remote Sensing of Environment 35 (1991) 257–277.
[13] D. Kamthonkiat, K. Honda, H. Turral, N.K. Tripathi, V. Wuwongse, Discrimination of irrigated and rainfed rice in a tropical agricultural system using
SPOT VEGETATION NDVI and rainfall data, International Journal of Remote Sensing 26 (2005) 2527–2547.
[14] M.A. Friedl, D.K. McIver, J.C.F. Hodges, et al., Global land cover mapping from MODIS: algorithms and early results, Remote Sensing of Environment
83 (1/2) (2002) 287–302.
[15] M. Pax Lenney, C.E. Woodcock, J.C. Collins, H. Hamdi, The status of agricultural lands in Egypt: the use of multitemporal NDVI features derived from
landsat TM, Remote Sensing of Environment 56 (1) (1996) 8–20.
[16] S.B. Tennakoon, V.V.N. Murty, A. Etumnoh, Estimation of cropped area and grain yield of rice using remote sensing data, International, Journal of
Remote Sensing 13 (3) (1992) 427–439.
[17] M. Ozdogan, G. Gutman, A new methodology to map irrigated areas using multi-temporal MODIS and ancillary data: an application example in the
continental US, Remote Sensing of Environment 112 (2008) 3520–3537.
[18] J. Chen, H. Lin, Z. Pei, Mapping rice crop growth using ENVISAT ASAR data in southern China, IEEE Transactions on Geosciences and Remote Sensing
Letters 3 (6) (2007) 431–436.