|GeoImaging & GeoInformatics|
|Your location:||GIGI||>||Modules||>||Image Processing in the Temporal Domain|
|Introduction||Objectives||Getting Started||Essential Learning||Hungry Minds||Summary|
In this topic you will develop an understanding of reasons for, approaches to and techniques for processing multitemporal images. These skills are required to exploit the archives of EO data that are being constructed at regional and and global scales because some EO missions have acquired images for more than 30 years. Furthermore many newer missions are accompanied by initiatives to create online information systems to make life easier for analysts searching for useful data.
Before going any further let us clarify some terms. Multidate remote sensing refers to tasks supported by data acquired on different days whereas multitemporal remote sensing is a broader term that is used to describe not only those studies but also those that process images acquired on two or more occasions on the same day. This is an important distinction to make because when selecting multitemporal EO data it is important that an analyst appreciates the frequency of change in the targets s/he plans to monitor. To continue that theme, when selecting multitemporal EO data it is important to appreciate the nature of the change in the target - notably, the magnitude and rate of change, whether or not the change is reversible (cyclic) or permanent, and the scale of variation and spectral characteristics of the variable(s) of interest.
Unfortunately, there are several factors whose variations obscure meaningful trends in the data. These include illumination sources, atmospheric conditions, view geometry and resolution and the effects of georegistration and sensor calibration. Furthermore, we must remember that in any set of observations there is likely to be some natural variation in the measurements and thus it becomes the responsibility of an image analyst to select a method that isolates variation in those key variable(s) of interest.
In this topic you will discover, through case studies and practical demonstration, that the task of analysing temporal variation is not easy even though techniques for processing multitemporal images are many and varied. We must remember that the quality of geoinformation generated by these techniques should be assessed in terms of its accuracy and whether or not it is fit for the intended purpose - a purpsose defined by the intended user. Disturbing as it may seem when set in this context, there have been relatively few efforts to develop methods to assess the accuracy of multitemporal information. Before we get too disheartened I should add that we are still in the early stages of developing techniques to extract information in the temporal domain, certainly when compared to techniques to extract information in the spectral domain, and one can expect only substantial improvements as more studies devote resources to multitemporal analyses.