Author(s): Bruno Oliveira; Rodrigo Maia
Linked Author(s):
Keywords: Stochastic Processes; Streamflow Generation; Markov chain model; Derivate Decomposition; Synthetic Data
Abstract: In nearly all projects involving the study of river flow and its variability (e. g. on Water Resource Management or Environmental Protection) ,the selection/definition of the flow’s time series is of great importance. Commonly, available time series suffer from either being too short or from having too many missing values for the desired applications, thereby limiting its interest, relevance and applicability. In the present paper we study the possibility of generating a large number of flow time series by using a non-parametric methodology, or, conversely, of creating a long, realistic, flow series from a comparatively small amount of observed data. Overall, this procedure involves the separation of the available observed time series into its multiple components (i. e. periodicity, random variation, etc. ), the assessment and disentanglement of their respective relevance, as well as the generation/sampling of time series values, all within a univariate framework. The proposed methodology produced good results with only a minimal amount of user intervention (the method only requires the definition of one parameter) .As is, the method is capable of reproducing common and complex streamflow structures (including streamflow recession) ,displaying only a relatively small amount of noise-like variation resulting from its stochastic nature. Conversely, it should be noted that this method makes little assumption in regard to the series’behavior and is therefore limited to reproducing results with the same quality as that of the input data (the observed streamflow series) .Ergo, care should be taken when applying this method using very short observed streamflow series as references.
Year: 2015