Please use this identifier to cite or link to this item:
Title: The Investigation of Trends, Analysis and Modelling of Daily Rainfall in Australia
Authors: Don, McNeil
Bright Emmanuel Owusu
Faculty of Sciecnce and Technology (Mathematics and Computer Science)
คณะวิทยาศาสตร์และเทคโนโลยี ภาควิชาคณิตศาสตร์และวิทยาการคอมพิวเตอร์
Keywords: Analysis and Modelling
Issue Date: 2018
Publisher: Prince of Songkla University, Pattani Campus
Abstract: Daily accumulated rainfall data obtained from observational stations in Australia over a period of 64 years were explored using statistical methods to determine the patterns. The thesis composed three studies. In the first study, daily rainfall during 1950-2013 for 92 observational stations was obtained from Australian Bureau of Meteorology. Rainfall variability over successive 5-day periods from 1950-2013 was examined. The first model uses factor analysis to classify the 92 stations of factors, which represent distinct geographical regions. Gamma generalised linear models were then fitted to describe the patterns in the 5-day non-zero rainfall amount in each factor region with period and year as the predictors. The factor analysis revealed eight factors in the data, which represent eight geographical regions. These regions comprise north, northeast, central east, northwest, central south, north southeast, southwest and south-southeast. The rainfall from each region exhibits different overall mean with various seasonal variations. The second study also uses daily accumulated rainfall from the 92 observational stations acquired from the Australian Bureau of Meteorology. Multiple linear regression and Gamma GLM models fitted to the 5-day rainfall observations were compared. A simple linear regression model was also fitted to explore the trends in the annual rainfall. The 5-day rainfall observations in successive periods were serially correlated, and this was minimised by using the AR(1) technique in fitting the models. The Gamma GLM and the multiple regression models fitted the data quite well in all the regions, particularly in delineating the periodic seasonal patterns. The patterns of the models were well marked mainly in the tropical regions (lower latitudes) where the internal atmospheric variability is small relative to the forced change. However, the multiple regression models did better than the Gamma generalised linear model in the north, northeast and the southwest regions. These models could be used to infill data in areas where rainfall records are inadequate. The linear regression model revealed substantial decreasing annual rainfall trends in the southwest and the north southeast regions. In contrast, significant increasing annual rainfall trends were found in the north and the northwest regions. The final study used daily accumulated rainfall from 105 observational stations acquired from the Australian Bureau of Meteorology. The 64 years data also span between 1950 and 2013. The stations were randomly selected to cover the whole of Australia. Logistic regression was then applied to develop a model to predict 5-day rainfall probability of occurrence or non-occurrence in each station, but nine stations were carefully chosen as a case study. The nine stations were carefully selected to capture the patterns observed in the data from all the stations and to capture most of the six climate classification by the Australian Bureau of Meteorology. The fitted logistic regression models predicted the occurrence and non-occurrence rainfall events quite well with good accuracies. The predictors significantly affected the models in all stations. Analysis of the levels of each of the predictors showed that the parameters of the 5-day period factors were more influential in all the models, particularly during the rainy season in most stations relative to that of the annual factors.
Description: Thesis (Ph.D.(Research Methodology))--Prince of Songkla University, 2018
Appears in Collections:746 Thesis

Files in This Item:
File Description SizeFormat 
TC1548.pdf2.71 MBAdobe PDFView/Open

Items in PSU Knowledge Bank are protected by copyright, with all rights reserved, unless otherwise indicated.