The objective of this study is to find an efficient method of crop classification based solely on satellite microwave data. Microwave data can sometimes be the only non-contaminated satellite data available for a selected area, where frequent cloudiness makes optical data useless. Hence, this approach has to be studied at least as an alternative method for application when other data are missing. Although ASAR is not a fully polarimetric instrument, a selection of dual polarization modes and a selection of incidence angles are available when using ASAR alternating polarization product (APP). Larger incidence angle and cross-polarized data are better for crop discrimination, while lower incidence angle and co-polarized data contain more information on soil moisture. In the years 2003 and 2004 a sequence of ASAR IS2, IS4 and IS6 images acquired during a growth season was analyzed. Supervised classification of ASAR APP was performed with ground truth data collected for 700 plots covered by homogeneous crops. For growth seasons of 2003 and 2004 various combination of images were tested in order to find the best set of data providing the highest accuracy of crop classification.
An approach to classification of satellite images aimed at vegetation mapping in a wetland ecosystem has been presented. The wetlands of the Biebrza Valley located in the NE part of Poland has been chosen as a site of interest. The difficulty of using satellite images for the classification of a wetland land cover lies in the strong variability of the hydration state of such ecosystem in time. Satellite images acquired by optical or microwave sensors depend heavily on the current water level which often masks the most interesting long-time scale features of vegetation. Therefore the images have to be interpreted in the context of various ancillary data related to the investigated site. In the case of Biebrza Valley the most useful information was obtained from the soil and hydration maps as well as from the old vegetation maps. The object oriented classification approach applied in eCognition software enabled simultaneous use of satellite images together with the additional thematic data. Some supplementary knowledge concerning possible plant cover changes was also introduced into the process of classification. The accuracy of the classification was assessed versus ground-truth data and results of visual interpretation of aerial photos. The achieved accuracy depends on the type of vegetation community in question and is better for forest or shrubs than for meadows.