Data-Driven Modelling. Investigation of data-driven flood forecasting models

      Аннотация: The advantage of statistical models of input-output type is that they can be relatively easily constructed and applied, but on the other hand the disadvantage of such models is that that they don’t reveal the inner nature of observed phenomenon. Conceptual models, which have advantage of transparent functioning, but are sometimes hard to be proven correct. Artificial intelligence offers methods of machine learning from examples, which eliminate the disadvantages of statistical as well as conceptual approaches and integrate the advantages. A comprehensive data driven modelling experiment based on regression trees is presented in this book. Regression trees have been employed on practical problem of constructing a data driven model for runoff prediction from known present and past runoff at water-level-gauges and rainfall at rain gauges within the catchment. Results based on approximation and prediction accuracy obtained from regression trees are then compared with other DDM techniques namely, artificial neural networks, Gaussian process, support vector regressions and multiple linear regressions. Book is a must read for the researchers working in the field of data-driven modelling.

Ключевые слова: artificial neural networks, data-driven modelling, Gaussian process, hydrological modelling, Multiple Linear Regression, regression trees, support vector regression, flood forecasting models, rainfall-runoff models