An alternative representation of coarse-resolution remote sensing images for time-series processing

This paper presents a solution to increase the accuracy of time-series processing of coarse-resolution Earth observation imagery (such as MODIS). It is based on two main points. First, the processing of imagery is based on a vector data model that enables more accurate representation of the actual observation footprints than the original raster model. Second, time-series composition is carried out at the level of units of analysis relevant to the problem to be handled (e.g. agricultural parcels for the monitoring of agricultural activities), defined a priori from ancillary data sources, independent from predefined grids used for “traditional” time-series processing. Compared to the raster/grid-based approach, this methods gives more control over time-series composition and analysis by providing details on the geometric configuration and hence the “purity” of each observation (pixel) with respect to the objects to be observed. Testing is done over an agricultural area in Hungary.