tceic.com
学霸学习网 这下你爽了
当前位置:首页 >> 文学研究 >>

Assessment+of+rice+fields+by+GISGPS-supported+classification+of+MODIS+data

412

Cheng et al. / J Zhejiang Univ SCI 2004 5(4) :412-417

Journal of Zhejiang University SCIENCE ISSN 1009-3095 http://www.zju.edu.cn/jzus E-mail: jzus@zju.edu.cn

Assessment of rice fields by GIS/GPS-supported classification of MODIS data*
CHENG Qian (程 乾)?, HUANG Jing-feng (黄敬峰), WANG Ren-chao (王人潮)
(Institute of Agricultural Remote Sensing & Information Application, Zhejiang University, Hangzhou 310029, China)
?

E-mail: qiancheng525@yahoo.com.cn

Received Apr. 23, 2003; revision accepted May 28, 2003

Abstract: The new Moderate-Resolution Imaging Spectroradiometer (MODIS) satellite image offers a large choice of opportunities for operational applications. The 1-km Advanced Very High Resolution Radiometer (AVHRR) image is not suitable for retrieval of field level parameter and Landsat data are not frequent enough for monitoring changes in crop parameters during the critical crop growth periods. A methodology to map areas of paddy fields using MODIS, geographic information system (GIS) and global position system (GPS) is introduced in this paper. Training samples are selected and located with the help of GPS to provide maximal accuracy. A concept of assessing areas of potential cultivation of rice is suggested by means of GIS integration. By integration of MODIS with GIS and GPS technologies the actual areas of rice fields in 2002 have been mapped. The classification accuracy was 95.7% percent compared with the statistical data of the Agricultural Bureau of Zhejiang Province.
Key words: Areas of rice fields, Moderate-Resolution Imaging Spectroradiometer (MODIS), Geographic information system (GIS), Global position system (GPS) Document code: A CLC number: TP7.5; S511

INTRODUCTION In the last decade, remote sensing has been increasingly identified as an objective, standardized, possibly cheaper and faster methodology for crop production surveys than conventional field investigations (Bauman, 1992). Much progress has been made through intensive studies on inventory and production forecasting for major crops, such as wheat, rice, cotton and maize (Tennakoon et al., 1992; Ray et al., 1994; Fang et al., 1998). China, a developing country with 1.2 billion people, has been paying close attention to its ex-

*

Project supported by the National Natural Science Foundation of China (No. 40171065) and the Hi-Tech Research and Development Program (863) of China (No. 2002AA243011)

isting and future food production. Crop production forecasting using remote sensing techniques for large areas is already one of the key projects in China. One of the key aims of crop production forecasting is to estimate areas of paddy fields. Satellite-based remote sensing techniques can be effectively applied for estimating areas of rice fields and for accurate assessment of the dynamic crop characteristics. The launch of Terra MODIS was an important milestone in moderate resolution remote sensing, providing a marked increase in observation capabilities. Many workers have analyzed vegetation indices derived from MODIS data for comparison with NOAA/AVHRR data (Ferreira et al., 2000; Fensholt et al., 2002). The two 250 m bands provide the moderate spatial resolution from MODIS and

万方数据

Cheng et al. / J Zhejiang Univ SCI 2004 5(4):412-417

413

unique daily observations of the land surface. The use of these data for land cover change detection was explored by Zhan et al.(2002). Introducing multi-temporal and multi-thematic spatial data supported the development of agricultural information systems based on GIS technologies. Janssen et al. (1990) observed an improvement of more than 12% in overall classification accuracy of agricultural fields when integrating topographical data from GIS with the classification of Landsat TM (Thematic Mapper) satellite data. In this paper a method for estimation of the extension of areas of rice fields in Zhejiang Province in 2002 is presented. The integration of MODIS, GIS and GPS increased the classification accuracy of areas of paddy fields. The location of paddy fields was defined by GIS and after training samples were selected by GPS, the areas of rice fields were then extracted by the classification of MODIS data.

1628?1652, 2105?2155 nm, respectively). Orbital parameters of the Terra satellite and MODIS viewing geometry produce daily global coverage with a revisiting frequency of approximately 1.2 days. MODIS data have higher spatial resolution than NOAA/AVHRR with similar temporal resolution. For this study band 1 and band 2 of MODIS imagery were used to map areas of rice fields. According to the farming practice in the study area MODIS data of July 24th and September 1st, 2002 covering the whole of Zhejiang Province was acquired with the support of the Remote Sensing Application Centre at the Chinese Academy of Science. This date correspond with the late rice planting season (transplant stage and elongation stage). The software packages used for image and geo-information analyses were ENVI 3.4 and ARC/ INFO 8.1. Topographic maps, 1:250000 scale land maps, recently produced maps of vegetation types, land cover / land use maps and other ancillary geoinformation were also used in the study. GPS was applied to register the training samples. Image processing Data processing was performed by using ENVI 3.4 software. Twenty-six ground control points (GCPs) extracted from a 1:250000 digital topographic map were used for geocoding the image data. Radiometric correction was carried out. Co-registration accuracy of the image was better than 0.5 pixel in both x and y directions. The 1:250000 land use map of Zhejiang Province was digitized and a land use mask was created. Afterwards the potential rice cultivation areas or the land use class rice fields were extracted. The MODIS image of Zhejiang Province was superimposed on the mask of potential areas of rice fields. According to investigations in the rice cultivation regions, the training samples were selected by using GPS. Classification of the specific subsets of MODIS data allowed for the assessment of the actual area of paddy fields. Fig.1 illustrates the whole process of information extraction.

MATERIALS AND METHODS Study area The study area is located in China’s Zhejiang Province (118°01′ E to 123°10′ E, 27°06′ N to 31°11′ N) with area of 10 530 000 ha and 71.6% hilly and mountainous regions, 22.0% plains and 6.4% rivers and lakes. The mean annual precipitation is 1320.9 mm and mean annual temperature is 16.2 °C. The soils are mostly red soils and paddy soils. Paddy fields cover most of the irrigated farm land. Agricultural production is comprised mainly of cereal crops such as rice, wheat and maize. Acquisition of image data The MODIS instrument onboard NASA’s Terra satellite is a scanning radiometer system with 36 spectral bands extending from the visible to thermal infrared wavelengths (Running et al., 1994). The first seven bands are primarily designed for remote sensing of land surfaces providing spatial resolutions of 250 m for band 1 (red, 620?670 nm) and band 2 (near infrared, 841?876 nm), and 500 m for bands 3 to 7 (459?479, 545?565, 1230?1250,

RESULTS AND DISCUSSION

万方数据

414

Cheng et al. / J Zhejiang Univ SCI 2004 5(4) :412-417

Pre-processing Scan(*.tif) Registration of grid image Layer of vector data Error check Revision ARC/INFO Build topographic Input code of data

Error check Revision

Assessment of rice fields areas using MODIS data The training samples were selected according to the land use map and the features in the MODIS image, additionally taking into account the variations of geographical and climatic conditions. Nineteen training samples of rice, dry land, water, forest and town were selected. The supervised approach of Minimum Distance Classification was conducted. Table 1 highlights the classification results. Because areas of rice fields are distributed all over the province and were thus hardly to be identified and separated from other sorts of vegetation coverage, the accuracy of results was poor. Assessment of areas of potential rice fields by integrating GIS Although the quantitative results of the Minimum Distance Classification using MODIS data seemed good in terms of extracting areas of rice fields, it is difficult to guarantee that the classification results obtained really belong to early rice and not to other vegetation types. The objective of this study was to extract areas of paddy fields using MODIS data. It was not aimed to classify all land use categories. Thus the potential areas of rice cultivation were extracted to avoid misclassification of rice fields versus forest, water and other land use classes. The potential areas of rice fields indicate the locations of areas of cultivation of rice as well as of rice rotation. Thus the potential areas of rice fields were usually larger than the actual areas. According to the regulations of the Zhejiang land use map, the cultivated land is divided into irrigated land, land of rainfed agriculture and dry land. The irrigated land and the land of rainfed agri-

Grid format Mask of potential rice Registration MODIS image

Band math ENVI Image of potential rice planting area

Training samples by GPS

Maximum likelihood classification

Rice planting area Fig.1 Flow chart of GIS/GPS-supported classification of MODIS data

Table 1 Classification results based on Minimum Distance Classification of MODIS imagery Name of type Rice area Dry land Forest Water area Town area Minimum distance classification Statistical area (ha) 957 293 278 713 5 587 535 912 357 559 293 Pixels 183 827 72 608 546 781 14 076 30 112 Area (ha) 1 148 918 453 800 3 417 381 87 975 188 200 Relative error (%) 20.0% 62.8% ?38.8% ?90.4% ?66.4%

万方数据

Cheng et al. / J Zhejiang Univ SCI 2004 5(4):412-417

415

culture are potential areas of rice cultivation. Thus these two classes were extracted from the digital land use map (Fig.2).

Therefore, the normalized difference vegetation index (NDVI) was used in this paper, and the difference of two times NDVI was calculated as follows:
k k k ?VI ij = VI ij (t2 ) ? VI ij (t1 )

(1)

Where VIij is vegetable indices (VI) of two times, k is wavelength, i and j are samples and lines of pixels, t1 and t2 are the first and second times. For the image of potential rice cultivation area,
k ?VI ij corresponding to indices of two times pixels

Fig.2 Potential rice cultivation areas and training samples

GPS-supported selection of training samples and subsequent extraction of areas of rice fields Although the extraction of potential areas of rice fields allowed for the removal of land cover classes such as dry land, forest, water and settlements, the effects of embedded land use features like roads, villages and field ridge could not be removed. Therefore the MODIS image and the potential areas of rice fields were overlaid in order to extract the corresponding subsets of imagery. Afterwards the actual late rice area were distinguished from non-rice area and extracted. In this paper, specifically, two MODIS 250 m scenes were analyzed; since, at this spatial resolution, only the Red and NIR bands are available. Different studies proved that after transplanting, the tillering number of rice increase with rice growth (Wang, 1996; Shen et al., 1996). With the increase of rice canopy, the reflectance in NIR wavelength increases. When the effects of background weaken, the reflectance in Red wavelength in rice field decreases. So, vegetable indices in the rice field increase until the heading stage of rice.

must be greater than zero. According to the farming practice in the study area, MODIS data of July 24th correspond to the transplant stage of rice growth in Zhejiang Province and MODIS data of September 1st correspond to elongation stage. In the transplant stage, the reflectance of rice which is young is weak, but in the elongation stage the reflectance of rice which is in active growth is stronger. Eq.(1) was used to calculate the difference of MODIS NDVI of July 24th and September 1st, 2002. The values of 23 520 pixels were less than zero and these pixels must be deleted from this image. The classification map was created using two bands at 250 m of MODIS data of
k September 1st and the ?VI ij image.

The supervised classification approach of Maximum Likelihood Classification was applied to extract the rice fields. In a supervised classification the spectral characteristics of the selected training samples are used to sharpen the classification algorithm towards mapping of specific land use/land cover classes based on the image data (Jensen, 1986). Once a representative number of training samples is selected, multivariate statistical parameters calculated from each training sample are used to classify each pixel. Each pixel is assigned to a land use class following its likelihood of membership. The quality of training samples is of utmost importance for achieving accurate classification results. According to investigations in the rice cultivation regions, 26 training samples were selected and registered by using GPS (Fig.2).

万方数据

416

Cheng et al. / J Zhejiang Univ SCI 2004 5(4) :412-417

Maximum Likelihood Classification achieved the best results in assessing areas of rice fields in hilly regions of Southern China compared to other classification methods (Lai and Yang, 1998). Therefore this study also applied the Maximum Likelihood Classifier for image analysis. The results showed that the late rice cultivation areas in Zhejiang Province in 2002 extended over 998 456 ha (Fig.3). The relative error of estimation of area is 4.3% compared to the data of the Agricultural Bureau of Zhejiang.

exactly extracted. The concept of GIS was employed to support spatial information management, and analysis of spatial data combined with related MODIS data was used to support the rice area extraction process. Using GPS for registering the training samples allowed for accurate calculation of spectral parameters of land use classes and especially of areas of rice fields. GIS was a very helpful tool to pinpoint the location of potential areas of rice fields. The classification accuracy of areas of late rice cultivation by means of remote sensing techniques was thus significantly increased. The two 250 m bands of MODIS provide moderate spatial resolution and daily observation of the land surface. It could be verified that these data are well suited for mapping areas of rice fields and for assessing growth parameters of rice. Using the integration of GIS, GPS and MODIS-based image analysis enhances the quantitative and qualitative accuracy of classification of rice fields. The methodology proved to be feasible, repeatable, and fast, since the GIS database and GPS were available. The range of relative error of classification was small. The methodology can be effectively used for the accurate classification of rice fields based on daily available MODIS earth observation data.

Fig.3 Area of late rice cultivation in Zhejiang Province

ACKNOWLEDGMENTS The quantitative accuracy of this method was enhanced up to 95.7%, and the spatial accuracy of rice location was also enhanced. Since late rice is planted in paddy fields and the results of this study were derived from MODIS imagery of these paddy fields, the probability of detecting areas of late rice was highest. The authors are very grateful to Prof. Habil. Elmar Csaplovics, Department of Geosciences, University of Dresden, Germany, who helped review this paper and gave valuable comments. References
Bauman, B.A.M., 1992. Radiometric measurements and crop yield forecasting – some observations over Millet and Sorghum experimental plots in Mali. International Journal of Remote Sensing, 9(10):1539-1552. Fang, H.L., Wu, B.F., Liu, H.Y., Huang, X., 1998. Using NOAA AVHRR and Landsat TM to estimate rice area year-by-year. International Journal of Remote Sensing, 19(3):521-525. Fensholt, R., Sandholt, I., Rasmussen, M.S., 2002. Earth Observation of Vegetation Status in A Semi-arid Envi-

CONCLUSION The introduced methodology is promising. It enhanced the quantitative accuracy from 80.0% for MODIS-based image classification to 95.7%. It ensured that the late rice cultivation areas were

万方数据

Cheng et al. / J Zhejiang Univ SCI 2004 5(4):412-417

417

ronment: Comparison of Terra MODIS and NOAA AVHRR Satellite Data. In: Proceedings of IEEE International Geoscience and Remote Sensing Symposium. Toronto, Canada, (4):2205-2207. Ferreira, L.G., Huete, A., Yoshioka, H., Sano, E., 2000. Preliminary Analysis of MODIS Vegetation Indices over the LBA Sites in the Cerrado Region, Brazil. In: Proceedings of IEEE International Geoscience and Remote Sensing Symposium. HI, USA, (2):524-526. Jensen, J.R., 1986. Introductory Digital Image Processing. Englewood Cliffs. Prentice Hall, New Jersey, p.177230. Janssen, L.F., Janorsma, M.N., Vander Linden, E.T.M., 1990. Integration topographic data with remote sensing for land-cover classification. Photogrammetric Engineering and Remote Sensing, 56(3):1503-1506. Lai, G.Y., Yang, X.W., 1998. Analysis of feasibility of professional operating on estimating rice planting area of hilly region in southern China using remote sensing technique. Remote Sensing Technology and Application, 13(3):1-7 (in Chinese). Ray, S.S., Pokharna, S.S., Ajar, E., 1994. Cotton production estimation using IRS-1B and meteorological data. International Journal of Remote Sensing, 15(5):10851090.

Running, S.W., Justice, C.O., Salomonson, V.V., Hall, D., Barker., J., Kaufman, Y.J., Strahler, A.H., Huete, A.R., Muller, J.P., Vanderbit, V., Wan, Z.M., Teillet, P., Carneggie, D., 1994. Terrestrial remote sensing science and algorithms planned for EOS/MODIS. International Journal of Remote Sensing, 15(17):427-439. Shen, Z.Q., Wang, K., Wang, R.C., 1996. A study on dynamic changes of rice reflective characteristics in growth period. Remote Sensing for Land & Resources, 30(4):40-45 (in Chinese). Tennakoon, S.B., Murty, V.V.N., Eiumnoh, A., 1992. Estimation of cropped area and grain yield of rice using remote sensing data. International Journal of Remote Sensing, 13(3):427-439. Wang, Y.Y., 1996. The relationship between vegetation index and rice growth and rice yield components. Remote Sensing for Land & Resources, 30(1):56-59 (in Chinese). Zhan, X., Sohlberg, R.A., Townshend, J.R.G., DiMiceli, C., Carroll, M.L., Eastman, J.C., Hansen, M.C., DeFries, R.S., 2002. Detection of land cover changes using MODIS 250 m data. Remote Sensing of Environment, 83(1-2):336-350.

万方数据

Assessment of rice fields by GIS/GPS-supported classification of MODIS data
作者: 作者单位: 刊名: 英文刊名: 年,卷(期): 被引用次数: 程乾, 黄敬峰, 王人潮 Institute of Agricultural Remote Sensing & Information Application, Zhejiang University, Hangzhou 310029, China 浙江大学学报(英文版) JOURNAL OF ZHEJIANG UNIVERSITY SCIENCE 2004,5(4) 2次

参考文献(13条) 1.Ferreira L G;Huete A;Yoshioka H Preliminary Analysis of MODIS Vegetation Indices over the LBA Sites in the Cerrado Region, Brazil[外文会议] 2000 2.Fensholt R;Sandholt I;Rasmussen M S Earth Observation of Vegetation Status in A Semi-arid Envi-ronment: Comparison of Terra MODIS and NOAA AVHRR Satellite Data[外文会议] 2002 3.Fang H L;Wu B F;Liu H Y;Huang X Using NOAA AVHRR and Landsat TM to estimate rice area year-by-year[外文期刊] 1998(03) 4.Bauman B A M Radiometric measurements and crop yield forecasting - some observations over Millet and Sorghum experimental plots in Mali 1992(10) 5.Zhan X;Sohlberg R A;Townshend J R G;DiMiceli C,Carroll M L,Eastman J C,Hansen M C,DeFries R S Detection of land cover changes using MODIS 250 m data[外文期刊] 2002(1-2) 6.Wang Y Y The relationship between vegetation index and rice growth and rice yield components 1996(01) 7.Tennakoon S B;Murty V V N;Eiumnoh A Estimation of cropped area and grain yield of rice using remote sensing data 1992(03) 8.Shen Z Q;Wang K;Wang R C A study on dynamic changes of rice reflective characteristics in growth period 1996(04) 9.Running S W;Justice C O;Salomonson V V;Hall D,Barker,J,Kaufman Y J,Strahler A H,Huete A R,Muller J P,Vanderbit V,Wan Z M,Teillet P,Carneggie D Terrestrial remote sensing science and algorithms planned for EOS/MODIS 1994(17) 10.Ray S S;Pokharna S S;Ajar E Cotton production estimation using IRS-1B and meteorological data 1994(05) 11.Lai G Y;Yang X W Analysis of feasibility of professional operating on estimating rice planting area of hilly region in southern China using remote sensing technique[期刊论文]-遥感技术与应用 1998(03) 12.Janssen L F;Janorsma M N;Vander Linden E T M Integration topographic data with remote sensing for land-cover classification 1990(03) 13.Jensen J R Introductory Digital Image Processing 1986

引证文献(2条) 1.程乾 基于MOD13产品水稻遥感估产模型研究[期刊论文]-农业工程学报 2006(3) 2.程乾.王人潮 数字高程模型和多时相MODIS数据复合的水稻种植面积遥感估算方法研究[期刊论文]-农业工程学报 2005(5)

本文链接:http://d.g.wanfangdata.com.cn/Periodical_zjdxxb-e200404006.aspx


网站首页 | 网站地图 | 学霸百科
All rights reserved Powered by 学霸学习网 www.tceic.com
copyright ©right 2010-2021。
文档资料库内容来自网络,如有侵犯请联系客服。zhit325@126.com