ANALYSIS OF MULTIVARIATE STOCHASTIC HYDROLOGICAL SYSTEMS USING TRANSFER FUNCTION-NOISE MODELS

In recent years, hydrologists have attempted, with encouraging results, to synthesize the accomplishments of univariate time series analysis and deterministic systems methods into the more general framework of stochastic dynamic systems. Nevertheless, many problems remain unsolved, others still unformulated.;The premise is that our prior knowledge of the physical system we want to study, as well as our objectives, can considerably simplify our inquiry.;The theme is the proper identification of model structure. This is important both for proper estimation of the model, as well as for its proper interpretation in terms of the hydrological processes involved.;The main objective of this study is to develop a procedure for identification and estimation of a model for analysis of multivariate stochastic systems. A general stochastic systems model is presented and from it is derived a multiple input-single output transfer function-noise model that is well suited for modeling of hydrological systems. The model is then further developed to account for correlated inputs, which is often the case for hydrological systems. A general model building strategy is then developed and applied to real watershed systems.;The aim of this thesis is to resolve some of these problems, using as guidelines a balanced view of the physical reality we want to understand and the mathematical methods we use as our tools.;A multiple input transfer function-noise model for riverflow using correlated input series of precipitation, groundwater levels and temperature, was identified, estimated and checked using data for the Ellidaar River Basin in Iceland, demonstrating the validity of the proposed procedures. Its performance in terms of residual and prediction variances compares favorably with the performance of a univariate model for the riverflow. Its interpretation in terms of the geophysical processes involved is easier and more illuminating than is the case with the univariate model.;The transfer function-noise model exploits our knowledge about the causal relations of the processes present in a watershed system. This leads to improved analysis and furthers our understanding of the hydrological system itself and its processes. It thus brings us back to inquiries about nature itself instead of approaching the problem as a mere statistical analysis.