|
Fast access to satellite image time series data has received a leg up through a data storage concept developed by the remote sensing research unit (RSRU) at the Meraka Institute of the CSIR.
Satellite images of the earth's surface are represented as two dimensional grids where each grid cell is called a pixel. Repeated observations of the same area at regular intervals produce a sequence of satellite images. This sequence of images can be visualised as a three dimensional data set with the time axis as the third dimension. This is informally referred to as a 'cube'.
A pixel value might, for example, represent the percentage of vegetation cover at a fixed location on the land surface. A sequence of these pixel values through time could be used by remote sensing applications to detect changes in the environment. A change in land use, such as the founding of a new settlement, could be detected by observing a sudden decrease in the vegetation cover percentage.
The RSRU's HiTempo project aims to develop algorithms for performing automated change detection of this nature in a high performance computing environment such as C4.
Satellite image time series were traditionally extracted by opening each image individually and locating the relevant pixel values. Because this is a small amount of required data in a large image file, this approach incurs a significant input/output overhead even on high performance storage systems, and makes processing of large data sets time consuming.
Instead of grouping pixel values by their spatial relationships in a separate file for each observation date, the data cube format stores all the observations for a single pixel together. This allows a time series to be retrieved in a single step. By reducing the number of steps required to retrieve a time series, the data cube allows optimal use of available hardware resources and allows researchers to apply change detection methods to large geographical areas.
Performance benchmarks on the data cube show that a speedup of between 20 and 60 times has been obtained on the hardware available in the C4 facility.
In addition to the performance benefits, the data cube representation maintains compatibility with remote sensing agencies worldwide by using the hierarchical data format (.hdf).
Enquiries: CSIR Communication
|
