The effect of EVI time series density on crop classification accuracy

Time series remote sensing data have been found very useful in discriminating crops due to its temporal character that map the whole stages of crops. In order to analyze their performances, a range of different time series i.e. 16-day, 32-day, 48-day and 64-day interval was built from MODIS (Moderate Resolution Imaging Spectroradiometer) 250-m Enhanced Vegetation Index (EVI) data. These times series were used to discriminate five crops i-e alfalfa, corn, sorghum, soybean and winter wheat in the United State, Kansas in 2010. The time series data were used to test the discriminating ability of different classifiers like Maximum Likelihood Classifier (MLC), Minimum Distance (MD), Support Vector Machine (SVM), Neural Network (NN) and Random Forest (RF) for crop classification. The results showed that the high temporal resolution time series returned high classification accuracy and vice versa. The results comparison showed that RF classifier returned high accuracy (overall accuracy 92.61%) followed by SVM and Neural Network. However, minimum distance and MLC returned low accuracy showing their less adoptability towards different time series data.