Quantifying temporal changes in gully erosion areas with object oriented analysis

Mapping gully systems and quantifying their changes over time are essential for implementing soil conservation measures. The aim of this study is to analyze the changes in the gully systems in the context of land management. Image analysis of remotely sensed data in combination with field verification has proven to be a good and practical approach for mapping gully features over large areas, given the variability in gully size, shape and occurrence. Recent work has shown object-based image analysis also termed as object oriented analysis (OOA) to be superior to pixel-based methods for the identification and characterization of gullies. This study presents an improved procedure to map gully features, one that has been extended to assess and quantify erosion area dynamics. In addition to the convention of using suitable topographic and spectral thresholds in the method, the use of contrast and edge filter information and process knowledge of erosion by overland flow was incorporated in the rule-set to help recognize potential gullies in Ikonos-2 and GeoEye-1 data. Changes in the area enclosed by the gully system boundary estimated for three different sub-watersheds (SW1, SW2 and SW3), observed over a period of eight years ranged from moderate (11% in SW2 and 21% in SW1) to a very high increase (91% in SW3). However, the changes seen in gully system density were much higher than the changes in the area enclosed by the gully system boundary (81%, 104% and 109% in the respective SWs). The complex gully systems were better identified semi-automatically than was possible by manual digitization; in particular the finer edges of the gully systems, which in turn led to better quantification of changes in the gully systems. The study also determined that the rule-set developed for gully feature extraction is more sensitive to the spatial resolution of the imagery used than the quality or resolution of the digital terrain model and its derivatives. Determining suitable segmentation parameters, selecting appropriate object features for the classification, the thresholds used retain the trial & error aspects in this analysis. This limits its application to a regional scale. The rule-set developed is generic and transferable to other region/imagery with adaptation. The rule-set will be made available on our website (www.itc.nl/OOA-group).