摘要
Crop phenology retrieval in the double-crop- ping area of China is of great significance in crop yield estimation and water management under the influences of global change. In this study, rice phenology in Jiangsu Province, China was extracted from multi-temporal MODIS NDVI using frequency-based analysis. Pure MODIS pixels of rice were selected with the help of TM images. Discrete Fourier Transformation （DFT）, Discrete Wavelet Transformation （DWT）, and Empirical Mode Decomposition （EMD） were performed to decompose time series into components of different frequencies. Rice phenology in the double-cropping area is mainly located on the last 2 IMFs of EMD and the first 2-3 frequencies of DFT and DWT. Compared with DFT and DWT, EMD is limited to fewer frequencies. Multi-temporal MODIS NDVI data combined with frequency-based analysis can retrieve rice phenology dates with on average 79% valid estimates. The sorting result for effective estimations from different methods is DWT （85%） 〉 EMD （80%） 〉 DFT （74%）. Planting date （88%） is easier to estimate than harvesting date （70%）. Rice planting date is easily affected by the former cropping mode within the same year in a double-cropping region. This study sheds light on under- standing crop phenology dynamics in the frequency domain of multi-temporal MODIS data.
Crop phenology retrieval in the double-crop- ping area of China is of great significance in crop yield estimation and water management under the influences of global change. In this study, rice phenology in Jiangsu Province, China was extracted from multi-temporal MODIS NDVI using frequency-based analysis. Pure MODIS pixels of rice were selected with the help of TM images. Discrete Fourier Transformation （DFT）, Discrete Wavelet Transformation （DWT）, and Empirical Mode Decomposition （EMD） were performed to decompose time series into components of different frequencies. Rice phenology in the double-cropping area is mainly located on the last 2 IMFs of EMD and the first 2-3 frequencies of DFT and DWT. Compared with DFT and DWT, EMD is limited to fewer frequencies. Multi-temporal MODIS NDVI data combined with frequency-based analysis can retrieve rice phenology dates with on average 79% valid estimates. The sorting result for effective estimations from different methods is DWT （85%） 〉 EMD （80%） 〉 DFT （74%）. Planting date （88%） is easier to estimate than harvesting date （70%）. Rice planting date is easily affected by the former cropping mode within the same year in a double-cropping region. This study sheds light on under- standing crop phenology dynamics in the frequency domain of multi-temporal MODIS data.
