The wavelet transform also can be used to remove noise. Its strength is the feasibility of identifying the timing of events such as localized objective signals in the presence of noise. The wavelet transform has been used for estimating forest LAI and for canopy closure mapping from EO-1 hyper-spectral data (Koger et al., 2003), and for detecting the interannual variability of NOAA/AVHRR NDVI and its relationship with El Niño/Southern Oscillation Index (Li & Kafatos, 2000). However few studies have used the wavelet transform for smoothing temporal VI data and for detecting crop phenological stages. The wavelet transform retains time components when transforming time-series data, and so can reproduce seasonal changes of vegetation without losing the temporal characteristics. In contrast, the trigonometric functions used in the Fourier transform are not localized in the time domain, and therefore the time component of the input data is averaged after the transforming process. Thus, we assumed that the wavelet transform could divide the noise components and reconstruct the seasonal VI time profile better than the Fourier transform. We used statistics of phenological stages of paddy rice to compare the performance of our method using the wavelet and Fourier transforms.