Multi-temporal Composite Trend Classification using DMSP-OLS Images

Abstract

The DMSP-OLS constellation of satellites constructs images consisting of visible light present on the Earth’s surface at night. The DMSP-OLS satellite data has been utilized for many different studies involving physical, social and economic indicators. It has been shown at varying scales that the DMSP-OLS satellite dataset correlates especially well with economic statistics. The method proposed in this study seeks to define a fast and accurate supervised classification of changes occurring over time using a multi-temporal composite of the DMSP-OLS images. A method is sought to identify trending in composited satellite-based remote sensing data representing multiple coverages over time by identifying and classifying linear trends using spectral matching. For the years 2005 to 2009, DMSP-OLS images were composited and co-registered. A time change profile library of idealized trends was created for input to the Spectral Angle Mapper in order to match and classify multi-temporal rasters. A map of trends was created. While an initial analysis of the results was performed, further analysis is pending development of an accuracy assessment for the method.