At the Hradec Králové branch of the Czech Hydrometeorological Institute, an applied climatology research group has been operating for several years now. Throughout this time, Statistica has been the essential tool for processing climatological data here and has proven extremely valuable.
Areas where the software is used at the Czech Hydrometeorological Institute in Hradec Králové
RNDr. Ladislav Metelka PhD., Head of the Hradec Králové branch, comments on the use of the software at the Hradec Králové branch:
Since our group focuses primarily on nonlinear empirical modelling in meteorology and climatology, one of the most frequently used tools in the entire Statistica product suite is the Neural Networks module. It proves very effective at modelling complex processes in nonlinear systems, in which the system's behaviour is influenced by a whole range of hidden positive and negative feedback loops. It is precisely the nonlinearity and complexity of the climate system that often limit the use of classical statistical methods or cause problems with identifying the optimal model or interpreting the results. Neural networks, by contrast, can very well model even these complex and nonlinear processes, generally deliver better results than classical empirical modelling methods, and are often able to bring an entirely new perspective on the problem at hand.
RNDr. Ladislav Metelka PhD., Head of the Czech Hydrometeorological Institute branch in Hradec Králové
Tasks addressed
RNDr. Metelka continues: „Today, the programs in the Statistica suite are routinely used to address a range of research tasks supported by the Czech Science Foundation, projects supporting science and research, and research projects supported by the European Commission, as well as in the course of everyday data processing. Neural networks have been used, for example, to address the following tasks:
- interpolation and regionalization of selected climatological characteristics across the Czech Republic
- correction of radar estimates of precipitation totals in situations with heavy torrential rainfall
- analysis of the nonlinear variability of the pressure field in the North Atlantic region (nonlinear principal component analysis using an autoassociative neural network, classification of circulation types using a Kohonen network)
- analysis of the dynamic and chemical influences acting on the ozone layer (within the European project CANDIDOZ, probably the very first use of neural networks for this purpose)
- reconstruction of doses of ultraviolet solar radiation reaching the Earth's surface in the period before direct measurement of this characteristic began
- analyses of the outputs of regional climate models
The results of a number of studies have been published in specialist journals (in the Czech Republic mostly in the journal Meteorologické zprávy) or presented at prestigious international conferences (the annual meetings of the American Meteorological Society, conferences on statistical climatology, and the conferences and workshops of individual research projects, e.g. the CLIVAR project).
Thanks to the interconnection of the Statistica modules, what we find outstanding is above all the ability to solve entire large-scale problems in a truly comprehensive way within a single software package. From the seamless selection of the required data from climatological databases via ODBC, through basic data analysis and preprocessing and the actual development of the model, all the way to high-quality numerical or graphical presentation of the results and the calculation of a range of related statistical characteristics. Also excellent is the ability to work with macros and to program practically every step using the Statistica Visual Basic programming language. This makes it possible both to automate repetitive calculations and to build very complex „hybrid“ models, in which, for example, classical statistical methods are combined with neural networks.“
RNDr. Ladislav Metelka PhD., Head of the Czech Hydrometeorological Institute branch in Hradec Králové