An inductive approach to simulating multispectral MODIS surface reflectance time series

Abstract

In this paper, a first order MODIS time series simulator, which uses a Colored Simple Harmonic Oscillator, is proposed. The simulated data can be used to augment data sets so that data intensive classification and change detection algorithms can be applied without enlarging the available ground truth data sets. The simulator’s validity is tested by simulating data sets of natural vegetation and human settlement areas and comparing it to the ground truth data in the Gauteng province located in South Africa. The difference found between the real and simulated data sets, which is reported in the experiments is negligent. The simulated and real world data sets are compared by using a wide selection of class and pixel metrics. In particular the average temporal Hellinger distance between the real and simulated data sets is 0.2364 and 0.2269 for the vegetation and settlement class respectively, while the average parameter Hellinger distance is 0.1835 and 0.2554 respectively.