Land-cover Monitoring Using Time Series of Satellite Images Based on Harmonic Analysis

Abstract

Remote Sensing is a useful tool to monitor land covers observing the earth's surface. Multi-temporal approach based on the temporal profile allows us to get information about the earth and human activities from a long-term perspective. The goal of this study is to develop an efficient method for land-cover monitoring using time series of satellite images. A harmonic model can characterize the temporal variability with four components: mean level, frequency, phase and amplitude. The components of the harmonic function inherently contain temporal information about seasonal changes. Seasonal periodicity can be incorporates into multi-temporal classification. In this research, a classification method using harmonic model is proposed. The method is tested and evaluated with multi-temporal Moderate Resolution Imaging Spectroradiometer (MODIS) data, one of the most useful remotely-sensed imagery. Satellite data inevitably contain disturbances caused by atmospheric effects and surface anisotropy scattering, which impede the analysis of time series data. The quality of data streams is important for the sequential data analysis. A filtering method was applied to reconstruct high-quality data stream. Then the proposed classification method was performed over the Korean peninsula from 2012 to 2016, where land-cover types were classified both with the estimated harmonic components using an unsupervised classification approach. The results of the classification show that the new approach has a great potential for land-cover monitoring