Stochastic modeling of sunshine number data

0398524 - ÚI 2014 RIV US eng C - Conference Paper (international conference)
Brabec, Marek - Paulescu, M. - Badescu, V.
Stochastic modeling of sunshine number data.
TIM 2012 Physics Conference. New York: AIP Publishing LLC, 2013 - (Bunoiu, M.; Biris, C.; Avram, N.), s. 178-187. AIP Conference Proceedings, 1564. ISBN 978-0-7354-1192-0. ISSN 1551-7616.
[TIM 2012 Physics Conference. Timisoara (RO), 27.10.2012-30.10.2012]
R&D Projects: GA MŠMT LD12009
Grant - others:European Cooperation in Science and Technology(XE) COST ES1002
Institutional support: RVO:67985807
Keywords : sunshine number * Markov chain * logistic regression model
Subject RIV: BB - Applied Statistics, Operational Research

We present a unified statistical modeling framework for estimation and forecasting sunshine number (SSN) data. Sunshine number has been proposed earlier to describe sunshine time series in qualitative terms (Theor Appl Climatol 72 (2002) 127-136) and it was shown to be useful both for theoretical and practical purposes, e.g. those related to the photovoltaic energy production. Statistical modeling and prediction of SSN as a binary time series has been challenging problem, however. Our statistical model for SSN time series is based on an underlying stochastic process formulation of Markov chain type. We will show how its transition probabilities can be efficiently estimated within logistic regression framework. In fact our logistic Markovian model can be fitted via maximum likelihood approach. This is optimal in many respects and it also enables us to use formalized statistical inference theory to obtain not only the point estimates of transition probabilities and their functions of interest but also related uncertainties as well as to test of various hypotheses. It is straightforward to deal with non-homogeneous transition probabilities in this framework. Very importantly, logistic Markov model class allows us to test hypotheses about how SSN dependents on various external covariates (e.g. elevation angle solar time etc.) and about details of the dynamic model (order and functional shape of the Markov kernel etc.). Therefore using generalized additive model approach (GAM), we can fit and compare models of various complexity which insist on keeping physical interpretation of the statistical model and its parts. After introducing the Markovian model and general approach for identification of its parameters we will illustrate its use and performance on high resolution SSN data from the Solar Radiation Monitoring Station of the West University of Timisoara.
Permanent Link: http://hdl.handle.net/11104/0226007