Jordan Graesser, Radost Stanimirova, Mark A Friedl

Time series reconstruction methods---used to generate gap-free time series of satellite observations---were historically designed for sensors with frequent image acquisitions. Since 2008, interest in leveraging time series methods has shifted from sensors such as AVHRR and MODIS to Landsat because of free, higher-resolution data availability and improved access to high-performance compute systems. Existing methods are typically designed for specific applications such as land cover classification or for estimating the timing of phenology events.Moreover, approaches developed for specific ecological systems, such as tropical forests or temperate agriculture, often do not generalize well across land cover, vegetation, and climate types. In this study, we introduce a dynamic temporal smoothing (DTS) method to reconstruct sparse, noisy signals into dense time series at regular intervals. The DTS is a weighted smoother with dynamic parameters that is applied over a signal. The smoother is intended to have wide applicability, with particular focus on applications in vegetation remote sensing. In this paper we present and illustrate the DTS over short- and long-term Landsat (TM, ETM+, and OLI) time series and demonstrate the effectiveness of robust gap-filling over a range of landscapes in the South American Southern Cone region.