Improving Land Cover Class Separation Using an Extended Kalman Filter on MODIS NDVI Time-Series Data

Abstract

Abstract-It is proposed that the normalized difference vegetation index time series derived from Moderate Resolution Imaging Spectroradiometer satellite data can be modeled as a triply (mean, phase, and amplitude) modulated cosine function. Second, a nonlinear extended Kalman filter is developed to estimate the parameters of the modulated cosine function as a function of time. It is shown that the maximum separability of the parameters for natural vegetation and settlement land cover types is better than that of methods based on the fast Fourier transform using data from two study areas in South Africa.

Detecting Land Cover Change Using an Extended Kalman Filter on MODIS NDVI Time-Series Data

“…The exact date of change does not affect the method as long as the change does not occur within the first two years of the NDVI time series. The reason for this is that an initial two year period should be allowed for the EKF to effectively track the parameters of the triply modulated cosine model for each NDVI time series [18]. As will be described in section III, the algorithm uses a 3x3 pixel grid with the center pixel being compared to all neighboring pixels.…”

“…It was proposed in [18] that the MODIS 8-day NDVI time series be modeled as a single, but triply modulated cosine function, where the mean µ, amplitude α and the phase φ values are a function of time. The parameters of the triply modulated cosine function were estimated using a non-linear extended Kalman filter (EKF) and the consequent EKF derived parameter sequences of the mean and amplitude were found to be highly separable for natural vegetation and settlement land cover types [18].…”

“…The parameters of the triply modulated cosine function were estimated using a non-linear extended Kalman filter (EKF) and the consequent EKF derived parameter sequences of the mean and amplitude were found to be highly separable for natural vegetation and settlement land cover types [18]. This paper extended this methodology to the change detection case by calculating a change metric by means of spatial comparison of the EKF derived mean and amplitude parameter sequences of any given pixel with that of its neighboring pixels (hereafter referred to as the EKF change detection method).…”

Rapid detection of new and expanding human settlements in the Limpopo province of South Africa using a spatio-temporal change detection method

“…This method uses MODIS time-series data, which have previously been shown to be separable (distinguishable) for the natural vegetation and settlement land cover classes considered in this study [11]. The method uses the ACF of a MODIS time-series to provide an indication of the level of time-series stationarity (by considering the stability of the time-series mean and variance over time) which is then consequently used as a measure of land cover change.…”

The Global Land Surface Satellite (GLASS) Remote Sensing Data Processing System and Products

“…A series of operating satellites from different countries are producing tremendous volumes of data at significantly higher levels of measurement precision [1,2]. Along with the application of the Advanced Very High Resolution Radiometer (AVHRR) [3] and Moderate Resolution Imaging Spectroradiometer (MODIS) [4] data, time series remote sensing data are now available in several spatial and temporal resolutions and have been used for various purposes [5][6][7][8][9][10]. The effective utilization of time series of remote sensing data is an important area in current research, concurrent with the development of domestic and overseas satellite technology [11][12][13].…”