Prediction of meteorological time series using nonlinear analysis methods

Doctoral Dissertation uoadl:2879989 80 Read counter

Unit:
Department of Physics
Library of the School of Science
Deposit date:
2019-08-02
Year:
2019
Author:
Kalamaras Nikolaos
Dissertation committee:
Δέσποινα Δεληγιώργη, Αναπληρώτρια Καθηγήτρια (Κύρια Επιβλέπουσα),
Τμήμα Φυσικής, Εθνικό και Καποδιστριακό Πανεπιστήμιο Αθηνών

Κωνσταντίνος Βαρώτσος, Καθηγητής
Τμήμα Φυσικής, Εθνικό και Καποδιστριακό Πανεπιστήμιο Αθηνών

Χρήστος Τζάνης, Επίκουρος Καθηγητής
Τμήμα Φυσικής, Εθνικό και Καποδιστριακό Πανεπιστήμιο Αθηνών

Κωνσταντίνος Καρτάλης, Καθηγητής
Τμήμα Φυσικής, Εθνικό και Καποδιστριακό Πανεπιστήμιο Αθηνών

Ματθαίος Σανταμούρης, Καθηγητής
University of New South Wales, Sydney, Australia

Νικόλαος Σαρλής, Αναπληρωτής Καθηγητής
Τμήμα Φυσικής, Εθνικό και Καποδιστριακό Πανεπιστήμιο Αθηνών

Ευθύμιος Σκορδάς, Επίκουρος Καθηγητής
Τμήμα Φυσικής, Εθνικό και Καποδιστριακό Πανεπιστήμιο Αθηνών
Original Title:
Πρόβλεψη μετεωρολογικών χρονοσειρών με τη χρήση μη γραμμικών μεθόδων
Languages:
Greek
Translated title:
Prediction of meteorological time series using nonlinear analysis methods
Summary:
The subject of this doctoral dissertation is the prediction of meteorological time series using nonlinear analysis methods. The methods used here are the Detrended Fluctuation Analysis (DFA) and, mainly, the Multifractal DFA (MF-DFA). This dissertation focuses on the study of those properties of temperature and dew point meteorological time series which cannot be detected by linear statistical methods and on the possibility of predicting the time series behavior in the future using autocorrelation at a wide time range. The way of the correlation between those properties and climatic conditions is also studied.
The originality of that dissertation is based mainly on the application of MF-DFA method using daily values of air temperature and dew point coming from many Greek weather stations observations. It also must be stressed the remarkably wide climatic range at a region which covers a relatively small area like the case of Greece. In particular, the reasons for the large climatic variety over Greece are its geographical location, the complex topography and the continual alternation of land and sea. In some cases, locations that are very close together have such a climatic difference that rarely can be found on our planet.
At first, DFA method was applied on daily mean, maximum and minimum air temperature time series and on daily dew point time series as well. These time series came from a number of weather stations of the Hellenic National Meteorological Service (HNMS) network and most of them cover the period from 1973 up to 2014. The basic conclusion is that for all the time series the scaling behavior is characterized by long-range positive correlations, that is the time series appear to have “memory”, which means that their pattern of fluctuations, even magnified, will be observed in the future.
MF-DFA method, which then applied on the same time series, verified the existence of the same scaling behavior as is the case for DFA method. Moreover, multifractal spectrum revealed the multifractal structure of the time series, which is mostly sensitive to local fluctuations with small magnitudes. The spatial distribution of the main multifractal spectrum parameters revealed the dependency of those parameters on the climatic conditions and local topography. The basic deseasonalization method that used in this study, involves the subtraction of daily mean values from the corresponding values of the time series for each year. The same conclusions found when another deseasonalization method, STL, was used and when MF-DFA was applied on homogenized temperature time series coming mostly from the same stations. In addition, the multifractal properties of the time series are found to come mainly from the different long-range correlations for fluctuations having different magnitudes. By the examination of three weather stations located on Greek regions with different climatic characteristics, time series were found to have relatively less persistent behavior in the winter period.
The use of temperature and dew point time series coming from ECMWF reanalysis data which taken from grid points which cover all the Greek area, also revealed the existence of long-range positive correlations at the time series. From the multifractal spectrum was also found the multifractal structure of the time series and the sensitivity of that structure on local fluctuations which mainly have small magnitude. But the most striking finding is that the sea and land distribution affects significantly the multifractal spectrum parameters. More specifically, time series over sea appear to have more persistent behavior (that is, greater positive long-range correlations) and a greater degree of multifractality.
MF-DFA method gives the opportunity of the study of meteorological time series (that come from the complex interaction among various processes in the atmosphere which obey to nonlinear laws), whose features cannot be detected by conventional linear methods. Moreover, the fact that meteorological time series appear to have memory can be used for the prediction of the future behavior of the climate variability. In addition, this analysis could help on the evaluation of climatic models in terms of their ability to reproduce the nonlinear dynamics of temperature variations.
Main subject category:
Science
Keywords:
Nonlinear analysis, Multifractal Detrended Fluctuation Analysis, Air Temperature, Dew point, Climatology
Index:
No
Number of index pages:
0
Contains images:
Yes
Number of references:
233
Number of pages:
220

Kalamaras PhD.pdf
49 MB
File access is restricted until 2022-08-07.