Mapping Paddy Rice Using Multi-Temporal SPOT NDVI Images

• Main Authors: Szu-wei Huang, 黃思維
• Other Authors: Chi-Farn Chen
• Format: Others
• Language: zh-TW
• Published: 2010
• Online Access: http://ndltd.ncl.edu.tw/handle/16498585982416191585

碩士 === 國立中央大學 === 土木工程研究所 === 98 === The rice planting date and rice distribution are important information for the agriculture and food policy of the government. Traditionally, the investigation of the rice planting date and rice distribution cost a lot and is a time-consuming work. The use of SPOT NDVI time-series data for detecting the planting date and ditribution of the paddy rice is proposed in this study. However, the time-series satellite data are easily contaminated by noises such as cloud cover and atmospheric conditions. Conventionally, wavelet analysis is commonly used to reduce the high frequency noises in the time-series data. However, the selection of the proper mother wavelet and its parameters usually affects the success of reducing the noises of the time series signal. In this study, a method called Empirical Mode Decomposition (EMD) is proposed to reduce the high frequency noises in SPOT NDVI time-series data. Then by detecting and analyzing the local minimal points on the smooth NDVI time series profiles can provide the necessary information about the rice planting date. A classification method is also developed by using correlation and sign-test to discriminate double rice crops form NDVI filtering data. The method of this study basically includes four main parts: (1) constructing Normalized Difference Vegetation Index (NDVI) time-series data; (2) EMD is proposed to reduce the noise in a year-long SPOT images and retrieve the time variation of paddy rice; (3) detecting the local minimum points from NDVI time-series data of the rice pixels as rice planting dates; (4) using correlation and sign-test to discriminate rice crops. SPOT data in Chihshang, Taibao and Shinwu (2005) are used to test the proposed method. The experiment indicates EMD provides more stable results than wavelet analysis because EMD is insensitive to pre-determined parameters. Moreover, the results show that the time-series data filtered by using EMD method to estimate rice planting date and to detect double rice fields are more accuracy than that by using wavelet analysis. Keywords: SPOT, Paddy rice, NDVI, EMD