RSRU – Projects

Time-series Analysis of Hyper-temporal satellite data, using High Performance Computing

The purpose of the research is to unlock the wealth of information in the large data sets of hyper-temporal (daily, 10-daily), coarse resolution satellite images using advanced time-series analysis in a High Performance Computing (HPC) environment. Sustainable development requires spatially-explicit information on environmental change and long-term satellite data archives (e.g. MODIS, AVHRR) are the only sources of such information at regional scales.

Existing methods for detecting environmental change from satellite data sets are insufficient, since data volumes are prohibitively large (>10 Terabytes), conventional image data structures cause a massive Input/Output bottleneck which impedes effective time-series analysis and its computational intensity far exceeds the capability of single workstations. As a result a number of potentially very useful signal processing methods have not yet been applied to this problem.

Objectives: (i) establish the CSIR as a world leader in time series analysis of satellite data, (ii) develop and optimize an HPC time-series analysis system (the “HiTempo” system), including a novel per-pixel, time-sequential data structure, (iii) compare and refine various time-series analysis methods (e.g. Hidden Markov Model, Singular Spectrum analysis and wavelet analysis) for characterizing patterns, trends and changes in earth surface properties by assessing the accuracy of their outputs (change or trend maps) with independent data (e.g. Landsat images). The HiTempo system will provide a platform for innovative cross-disciplinary R&D and thus promote wider application of digital and mathematical sciences.

The research will furthermore be supporting the South African Earth Observation System (SAEOS) and Global Earth Observation System of Systems (GEOSS). It is envisaged that the HiTempo system will deliver information (e.g. maps depicting environmental changes) to inform government decision making and policy development. This research will ultimately allow us to interpret the present in the context of the past, in near-real time.

View graphical representation of time series analysis of satellite data