Time-Series Processing for Hyper-Temporal Optical Data Analysis
Figure: Application of TIMESAT to land-NDVI measurements; (top): Raw land-NDVI imagery with poor pixels removed, and the output modelled data after applying TIMESAT, and (bottom): a 3-year single-pixel time-series of raw and modelled data processed through the TIMESAT algorithm. The curve allows for the extraction of model values to substitute the raw values.
Satellites provide near-daily repeat times over fixed points, and optical datasets currently extending over one decade (e.g. Terra/Aqua MODIS, Envisat MERIS), up to more than three decades (e.g. NOAA AVHRR). Such “hyper-temporal” (Piowar & LeDrew 1995; de Bie et al. 2008) time series datasets act as radiometrically and geometrically calibrated historical records of the Earth’s surface environmental characteristics over time, allowing analysts to track fluctuations corresponding to seasonal cycles in various ecological processes, including biological activity. For example, previous studies on land surface phenology (Dethier et al. 1973; Knapp & Dethier 1976) indicate it is possible to track seasonal fluctuations in land surface vegetation cover over extensive areas using satellite-based sensor systems. The creation of such hyper-temporal time series datasets has prompted calls for the development of new techniques, and the adaptation of existing image processing techniques to extract information from long-term satellite archives.
The application developed here aims to analyse and explore hyper-temporal time series of satellite products, namely ocean colour and sea surface temperature (SST), in order to extract spatiotemporal SST gradients, identify temperature fronts, assess spatiotemporal correlations between SST and biological phenologies, etc. Using the methods developed for this application, standard image processing and time series analysis techniques are applied in a novel application area, and in a new dimension (time).Posted on: 2nd June 2016, by : admin