Long Satellite Image Time Series: Characterization and Information Content

Kurzfassung

Since the development of remote sensing satellites in the latter half of the 20th century huge quantities of remotely sensed data have been gathered in the archives. Such amount of data poses a difficult challenge in extracting valuable information from the data, being estimated that only a small percentage of the data is currently analyzed. Furthermore, the current approach of assuring the long-term preservation of the archived data only emphasizes the problem of data analysis in order to extract useful information. In this paper we focus on a different approach, that of long-term data exploitation, by defining a framework for characterization and extraction of long-term satellite image time series (SITS), allowing us to fully exploit the informational content of data archives. The proposed framework aims towards creating automated methods for SITS extraction, allowing us to overcome the issues related to working with historical imagery, such as the co registration of images with missing or erroneous metadata, and by proposing methods that automate the evaluation of data quality, in support of enhancing the informational content of the extracted SITS. Considering the future remote sensing missions such as the Landsat 8 and the Sentinel programme, we focus on developing methods that allow us to enhance the temporal resolution of the extracted SITS by adding new data, enabling us to combine older and new information for a better understanding of the spatio-temporal evolution patterns. Considering the vast amount of spatial, spectral and temporal information embedded within SITS, we aim to identify possible applications of SITS analysis. In support of describing the SITS informational content we define spatio temporal feature descriptors and their potential of use for specific applications. The proposed methods are applied on a set of 106 Landsat TM and ETM+ datasets covering the area of Bucharest and Ilfov county in Romania, covering a timespan of over 27 years, resulting in the extraction of one of the longest satellite image time series.