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

Abstract

Abstract-A method for detecting land cover change using NDVI time series data derived from 500m MODIS satellite data is proposed. The algorithm acts as a per pixel change alarm and takes as input the NDVI time series of a 3x3 grid of MODIS pixels. The NDVI time series for each of these pixels was modeled as a triply (mean, phase and amplitude) modulated cosine function, and an Extended Kalman Filter was used to estimate the parameters of the modulated cosine function through time. A spatial comparison between the center pixel of the the 3x3 grid and each of its neighboring pixel's mean and amplitude parameter sequence was done to calculate a change metric which yields a change or no-change decision after thresholding. Although the development of new settlements is the most prevalent form of land cover change in South Africa, it is rarely mapped and known examples amounts to a limited number of changed MODIS pixels. Therefore simulated change data was generated and used for preliminary optimization of the change detection method. After optimization the method was evaluated on examples of known land cover change in the study area and experimental results indicate a 89% change detection accuracy, while a traditional annual NDVI differencing method could only achieve a 63% change detection accuracy.

Detecting Anomaly Regions in Satellite Image Time Series Based on Sesaonal Autocorrelation Analysis

“…The anomaly regions in satellite image time series are detected by comparing the parameters of the fitted model for different parts of the time series data. The second group consists of the methods that monitor anomalies in satellite time series data using some forecasting model, such as Extended Kalman Filter (Kleynhans et al, 2011), Gaussian Process (Chandola and Vatsavai, 2011), harmonic model (Verbesselt et al, 2012;Zhou et al, 2014;Zhu et al, 2012), nonlinear least square or finite impulse response filter (Anees and Aryal, 2014a), and Martingale theory and martingale central limit theorem (Anees and Aryal, 2014b), etc. In general, these monitoring methods consist of two main steps, i.e., model-fitting of historical data and anomaly detection by comparing the new observations to the predictions from the fitted model.…”

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

“…The specific land product that was used was the MCD43A4 product [13]. This specific product have been used in various land cover change detection applications [7,8,9] and is produced using data acquired from the MODIS sensor on-board the Aqua and Terra satellites and provides one composited sample (consisting of 16 days of acquisition) every 8 days. This product has a spatial resolution of 500m and is BRDF-corrected.…”

“…In [7], a Neural network based post classification change detection approach was used to detect when land cover conversion takes place from natural vegetation to settlement classes. In [8], MODIS time-series data was modeled as a triply modulated cosine function and an Extended Kalman filter was used to track the parameters of the model and declare change based on parameter behavior. In [9], the use of Page's cumulative sum (CUSUM) test was proposed as a method to detect new settlement.…”

Multi-sensor forest vegetation height mapping methods for Tanzania

“…However, by using the same approach in Tanzania, the seasonal fluctuations of the optical signal may falsely be interpreted as state changes in a hidden Markov model. Kleynhans et al (2011) used extended Kalman filtering on an NDVI time series for land cover change detection.…”