Comparison of approaches for water surface area segmentation using high resolution TerraSAR-X data for reservoir monitoring in a large semi-arid catchment in northeastern Brazil

The semi-arid Northeast of Brazil is characterized by distinct rainy and dry seasons. The water supply for the local population is based on surface reservoirs in which precipitation is collected. There are more than 150 reservoirs in the 933 km2 Benguê catchment, however, little is known about the temporal dynamics of the water storage in the reservoirs. In this study, we use TerraSAR-X imagery for a year-long monitoring of reservoir surface areas and their seasonal changes. The precise extraction of the reservoir surface areas forms the basis of the monitoring. Therefore, we evaluated the results of a pixel-based threshold classification and a feature-based segmentation (mean shift). The evaluation was based on in-situ GPS measurements and manual digitization. The results of the manual digitization and threshold classification were similar as both tended to underestimate the water surface area in comparison to GPS in-situ data. The mean shift segmentation, however, tended to spread over the shorelines into the surrounding areas. We used the threshold classification for the analysis of 47 TerraSAR-X images. The viewing direction of the TerraSAR-X sensor was also important for the distinction of the entire area of the reservoirs, since geometric effects at the shorelines shaded parts of the water surface area. For the monitoring of the reservoir area with only one viewing direction we derived an empirical geometry correction factor.