Universität Stuttgart
Permanent URI for this communityhttps://elib.uni-stuttgart.de/handle/11682/1
Browse
1 results
Search Results
Item Open Access Dynamic water masks from optical satellite imagery(München : Verlag der Bayerischen Akademie der Wissenschaften, 2019) Elmi, Omid; Sneeuw, Nico (Prof. Dr.-Ing.)Investigation of the global freshwater system has a vital role in critical issues e.g. sustainable development of water resources, acceleration of the hydrological cycle, variability of global sea level. Measurement of river streamflow is vital for such investigations as it gives a reliable estimate of freshwater fluxes over the continents. Despite such importance, the number of river discharge gauging station has been decreasing. At the same time, information on the global freshwater system has been increasing because of various types of ground observations, water-use information and spaceborne geodetic observations. Nevertheless, we cannot answer properly crucial questions about the amount of freshwater available on a certain river basin, or the spatial and temporal dynamics of freshwater variations and discharge, or the distribution of world’s freshwater resources in the future. The lack of comprehensive measurements of surface water storage and river discharge is a major impediment for a realistic understanding of the hydrological water cycle, which is a must for answering the aforementioned questions. This thesis aims to improve the methods for monitoring the surface extent of inland water bodies using satellite images. Satellite imaging systems capture the Earth surface in a wide variety of spectral and spatial resolution repeatedly. Therefore satellite imagery provides the opportunity to monitor the spatial change in shorelines, which can serve as a way to determine the water extent. Each band of a multispectral image reveals a unique characteristic of the Earth surface features like surface water extent. However selecting the spectral bands which provide the relevant information is a challenging task. In this thesis, we analyse the potential of multispectral transformations like Principal Component Analysis (PCA) and Canonical Correlation Analysis (CCA) to tackle this issue by condensing the information available in all spectral bands in just a few uncorrelated variables. Moreover, we investigate how the change between multispectral images at different epochs can be highlighted by using the transformations. This study proposes an automatic algorithm for extracting the lake water extent from MODIS images and generating dynamics lake masks. For improving the accuracy of the lake masks and computational efficiency of the algorithm, two masks are defined for limiting the search area. The restricting masks are developed according to DEM of the surrounding area together with a map of the long-term variation of pixel values. Subsequently, an unsupervised pixel-based classification algorithm is applied for defining the lake coastline. The algorithm particularly deals with the challenges of generating long time series of lake masks. We apply the algorithm on five lakes in Africa and Asia, each of which demonstrates a challenge for lake area monitoring. However in the validation section, we demonstrate that the algorithm can generate accurate dynamic lake masks. Rivers show diverse behaviour along their path due to the contribution of different parameters like gradient of the elevation, river slope, tributaries and river bed morphology. Therefore for generating accurate river reach mask, we need to consider additional sources of information apart from pixel intensity. The region-based classification algorithm that we propose in this study takes advantages of all types of available information including pixel intensity and spatial and temporal interactions. Markov Random Fields provide a flexible frame for interaction between different sources of data and constraint. To find the most probable configuration of the field, the Maximum A Posteriori solution for the MRF must be found. To this end, the problem is reshaped as an energy minimization. The energy function is minimized applying graph cuts as a powerful optimization technique. The uncertainty in the graph cuts solution is also measured by calculating the minimum marginal energies. The proposed method is applied to four rivers reaches with different hydrological characteristics. We validate the obtained river area time series by comparing with in situ river discharge and satellite altimetric water level time series. Moreover, in this study, we present river discharge estimation models using the generated river reach masks. Our aim is to find an empirical relationship between the average river reach width and river discharge. The statistics in the validation periods support the idea of using river width-discharge prediction models as a complementary technique to the other spaceborne geodetic river discharge prediction approaches.