Pages: 273-282

Authors: Dimitrije M. Maletić, Radomir M. Banjanac, Dejan R. Joković, Aleksandar L. Dragić, Nikola B. Veselinović, Mihailo R. Savić, Zoran R. Mijić, Vladimir I. Udovičić, Svetlana D. Živković-Radeta, and Jelena V. Udovičić

Abstract

Multivariate classification and regression analysis of multiple meteorological variables and indoor radon activity concentration in Ground Level Laboratory in the Institute of Physics Belgrade, was performed and discussed. Meteorological variables used in this analysis were from radon active device, nearby meteorological station and finally from Global Data Assimilation System. Single variate analysis has identified variables with greatest value of Pearson's correlation coefficient with radon activity concentration and also, variables with greatest separation of events with increased radon activity concentration of over 200 Bqm^-3 and of events with radon level below this value. This initial analysis is showing the expected behavior of radon concentration with meteorological variables, with emphasis on data periods with or without air conditioning and with emphasis on indoor water vapor pressure, which was, in our previous research, identified as important variable in analysis of radon variability. This single variate analysis, including all data, proved that Global Data Assimilation System data could be used as a good enough approximate replacement for meteorological data from nearby meteorological station for multivariate analysis. Variable importance of Boosted Decision Trees with Gradient boosting multivariate analysis method are shown for all three periods and most important variables were discussed. Multivariate regression analysis gave good results, and can be useful to better tune the multivariate analysis methods.

Key words: continuous radon monitoring, multivariate analysis, Global Data Assimilation System, meteorological station

FULL PAPER IN PDF FORMAT (1,65 MB)