High-Dimensional Pixel Composites From Earth Observation Time Series

High-quality and large-scale image composites are increasingly important for a variety of applications. Yet a number of challenges still exist in the generation of composites with certain desirable qualities such as maintaining the spectral relationship between bands, reduced spatial noise, and consistency across scene boundaries so that large mosaics can be generated. We present a new method for generating pixel-based composite mosaics that achieves these goals. The method, based on a high-dimensional statistic called the `geometric median,' effectively trades a temporal stack of poor quality observations for a single high-quality pixel composite with reduced spatial noise. The method requires no parameters or expert-defined rules. We quantitatively assess its strengths by benchmarking it against two other pixel-based compositing approaches over Tasmania, which is one of the most challenging locations in Australia for obtaining cloud-free imagery.