AbstractsBusiness Management & Administration

Monitoring land use dynamics with optical and radar remote sensing data in western Ukraine

by Jan Stefanski

Institution: Freie Universität Berlin
Department: FB Geowissenschaften
Degree: PhD
Year: 2015
Record ID: 1102682
Full text PDF: http://edocs.fu-berlin.de/diss/receive/FUDISS_thesis_000000098517


Remote sensing is a key technology for systematic and broad-scale observations of the Earth's surface and provides the basis for a large body of research and applications. However, region wide land use intensity mapping as well as monitoring of changes of land management based on remote sensing data has not yet been studied thoroughly. The main goal of this thesis was to develop and apply a framework for monitoring land management regimes that differ in land use intensity in order to advance the mapping and understanding of broad-scale land use changes based on remote sensing, and to assess the spatio-temporal patterns of land management regimes. The land management regimes defined in this thesis are large-scale cropland with a high management intensity, small-scale cropland with a potentially low management intensity, and farmland abandonment that implies no active land management. Eastern Europe is a prime example for drastic broad-scale land use changes due to the momentous political and socio-economic changes after the collapse of the Soviet Union in 1991. In order to monitor land management regimes in western Ukraine, a combination of different strategies was used. First, a semi-automatic parameter selection was developed to optimize and economize image segmentation, which is a prerequisite for object-based analysis. To select the optimal parameters of the Superpixel Contour segmentation algorithm, a predefined range of parameters is selected by the user and the best image segmentation is subsequently assessed by the fast internal accuracy assessment of the Random Forest classifier, and, optionally, by using additional validation data. Second, by integrating optical and radar data into the object-based image analysis, the synergistic and complementary effects of both data types were used to improve the mapping and monitoring approach. For example, the radar data with its high temporal resolution provided elementary information to distinguish pasture from abandonment, as both classes have a similar grassland cover but different phenological stages. Moreover, the relatively weather independent radar data was a reliable alternative to fill gaps of optical time series that can occur, for example, due to cloud cover. Third, by using a change trajectory analysis approach, land use and land use intensity changes were monitored for western Ukraine between 1986 and 2010. The results clearly showed substantial abandonment of the large collectivized farmland in the 1990s and 2000s. With Ukraine's integration in world markets and the emerge of agri-business at the end of the 2000s, many abandoned fields were recultivated. Since the beginning of the 1990s, small-scale cropland as subsistence agriculture emerged for the greatest part directly from the conversion from large-scale cropland. Nevertheless, large-scale cropland was the dominant class in the study area at any time during the study period. To further explore the spatial patterns of land management regimes, the final classification results were overlayed by a…