@phdthesis{Li2015, author = {Li, Chi-Yu}, title = {Time series scenario composition framework in Hydroinformatics Systems}, url = {http://nbn-resolving.de/urn:nbn:de:kobv:co1-opus4-33650}, school = {BTU Cottbus - Senftenberg}, year = {2015}, abstract = {Since Z3, the first automatic, programmable and operational computer, emerged in 1941, computers have become an unshakable tool in varieties of engineering researches, studies and applications. In the field of hydroinformatics, there exist a number of tools focusing on data collection and management, data analysis, numerical simulations, model coupling, post-processing, etc. in different time and space scales. However, one crucial process is still missing — filling the gap between available mass raw data and simulation tools. In this research work, a general software framework for time series scenario composition is proposed to improve this issue. The design of this framework is aimed at facilitating simulation tasks by providing input data sets, e.g. Boundary Conditions (BCs), generated for user-specified what-if scenarios. These scenarios are based on the available raw data of different sources, such as field and laboratory measurements and simulation results. In addition, the framework also monitors the workflow by keeping track of the related metadata to ensure its traceability. This framework is data-driven and semi-automatic. It contains four basic modules: data pre-processing, event identification, process identification, and scenario composition. These modules mainly involve Time Series Knowledge Mining (TSKM), fuzzy logic and Multivariate Adaptive Regression Splines (MARS) to extract features from the collected data and interconnect themselves. The extracted features together with other statistical information form the most fundamental elements, MetaEvents, for scenario composition and further time series generation. The MetaEvents are extracted through semi-automatic steps forming Aspects, Primitive Patterns, Successions, and Events from a set of time series raw data. Furthermore, different state variables are interconnected by the physical relationships derived from process identification. These MetaEvents represent the complementary features and consider identified physical relationships among different state variables from the available time series data of different sources rather than the isolated ones. The composed scenarios can be further converted into a set of time series data as, for example, BCs, to facilitate numerical simulations. A software prototype of this framework was designed and implemented on top of the Java and R software technologies. The prototype together with four prototype application examples containing mathematical function-generated data, artificial model-synthetic hydrological data, and measured hydrological and hydrodynamic data, are used to demonstrate the concept. The results from the application examples present the capability of reproducing similar time series patterns from specific scenarios compared to the original ones as well as the capability of generating artificial time series data from composed scenarios based on the interest of users, such as numerical modelers. In this respect, it demonstrates the concept's capability of answering the impacts from what-if scenarios together with simulation tools. The semi-automatic concept of the prototype also prevents from inappropriate black-box applications and allows the consideration of the knowledge and experiences of domain experts. Overall, the framework is a valuable and progressive step towards holistic hydroinformatics systems in reducing the gap between raw data and simulation tools in an engineering suitable manner.}, subject = {TSKM; Event identification; Time series analysis; Scenario composition; Hydroinformatics; TSKM; Ereigniserkennung; Zeitreihenanalyse; Szenariokomposition; Hydroinformatik; Hydrologie; Datenbank; Zeitreihe; Algorithmus}, language = {en} } @phdthesis{AlegueFeugo2008, author = {Alegue Feugo, Jean Duclos}, title = {Investigating ecological indicators of freshwater ecosystems using signal analysis methods}, url = {http://nbn-resolving.de/urn:nbn:de:kobv:co1-opus-4396}, school = {BTU Cottbus - Senftenberg}, year = {2008}, abstract = {Human life depends on water of good quality. Freshwater ecosystems are being degraded as a result of anthropogenic activities. Managing freshwater resources require a good understanding of the dynamics of the processes affecting water quality and the interrelationship existing between them. Signal analysis methods are used to extract information from water quality time series. The information obtained from such analysis is required for developing dynamic models as an aid in investigating the outcome of different management scenarios. They are equally useful in determining the appropriate sampling frequency required for monitoring ecological indicators. Modern methods developed in other field such as mathematical statistics need to be investigated in the fields of water quality management so as to enhance the knowledge of the functioning and the structure of water bodies. Classical methods consisting of time domain and frequency domain methods in combination with a modern method, wavelet analysis, were used to extract information from water quality indicators from the River Havel in the State of Brandenburg in Germany. The indicators were dissolved oxygen, chlorophyll-a and water temperature which are respectively chemical, biological and physical indicators. The time domain methods revealed the behavior of the signal across time as well as the relationship between them. The frequency domain methods proved quite inadequate because the signals from water quality are non-stationary with changing variance across time. The wavelet methods were quite good in unraveling the behavior of these signals at different time-scales. This analysis revealed that the high frequency changes have no significant effect on the long term dynamics of water quality signals. Given that only the low frequency components influence the long term behavior of these signals, it was found that it makes more sense to sample most of the signals at a weekly or two weekly intervals so as to avoid noisy or redundant information. In addition, the time-scale decomposition allows for noisy or redundant information that blurs the long term dynamics thereby negatively affecting the quality of models to be detected and kicked out. Moreover, it was also found that the use of a time delay in a dynamic model should be based on the delay from the time-scale that influence the long term dynamics the most in the freshwater body by the help of the wavelet cross-correlation rather than the classical cross-correlation. Finally, the same indicator from different water bodies of the same watershed portrayed different time and frequency dependent behavior implying that each freshwater body needs to be uniquely investigated. Applying the same management techniques to different water bodies without prior investigation will not be sound. Freshwater ecosystems are more and more threatened by phenomena such as global warming, pharmaceuticals in water bodies, invasive species requiring more investigation on the applicability of tools developed in other fields of science in the field of water resource management. Enhanced knowledge will improve the existing knowledge on the structure and functioning of freshwater bodies, the quality of models developed as management decision aid and the quality of the data used for decision making.}, subject = {S{\"u}ßwasser; {\"O}kosystem; Verschmutzung; Indikatoren; Zeitreihenanalyse; wavelet; Pollution; Indicators; Time series; Modeling; Wavelet}, language = {en} }