Empirical mode decomposition and wavelet analysis of time-series spot data for evaluating rice planting calendar

Time series of optical satellite data acquired from low-to-medium resolution satellite sensors have been proven as an important source of information for monitoring agricultural practices at the regional scale due to its wide coverage and high frequency of revisit on a regular basis. However, the mixed pixel problem usually effects the spectral characteristics of satellite imagery when applied for a sub-national scale. The objective of this study is to assess the rice planting calendar in Taiwan using SPOT data (acquired on an irregular basis).The data processing consists of three main procedures: (i) constructing time-series SPOT NDVI data; (ii) filtering time-series NDVI data by empirical mode decomposition (EMD) and wavelet transform; and (iii) detecting rice planting date from smooth time-series NDVI profile by using the local minimum method. As for the filtering with EMD, we used the EMD-based low-pass filter, which is designed based on the analysis of initial EMD-sifting results. In wavelet analysis, three types of mother wavelet (Coiflet 4, Symlet 6,Daubechies 13) were used. The results validated with the government rice planting statistics indicate that the case of using EMD-based filtered data for determination of rice planting date gave remarkably good results in comparison with other cases using wavelet transform. We thus proposed this EMD-based filtering method for noise reduction of time-series vegetation index of satellite data acquired at different time intervals for evaluating rice agricultural practices.