A new method of MCI extraction with multi- temporal MODIS EVI data

With Huabei Plain of China as the study area, this research proposed a new method for pixel-level MCI (multiple cropping index) extraction with multi-temporal MODIS EVI data based on crop phenology and DT (decision tree). The method could be divided into two steps. First, based on the local crop phenology, several features were identified to discriminate pixel-level MCI, which includes three types: fallow, single cropped and double cropped. Jeffries-Matusita (JM) distance was employed to measure the features´ MCI discriminating ability and determine which feature to be used. Second, the threshold of each feature was calculated with the CART algorithm of Tree model in SPSS. Then two types of data (statistics and visual interpretation results gained from TM images) were used to validate the accuracy of the method. Furthermore, the new method was compared with the HANTS method commonly used in MCI extraction in previous researches. The results showed that the new method was superior in both data requirement and processing speed with no loss in performance.