1. UKM Learning and Research Repository
2. PHD Thesis / Tesis PHD
3. Faculty of Science and Technology / Fakulti Sains dan Teknologi

Title:	Pemodelan Bayesan untuk kebasahan dan kekeringan cuaca Semenanjung Malaysia
Authors:	Wahidah Sanusi (P54226)
Supervisor:	Abdul Aziz Jemain, Prof. Dr.
Keywords:	Pemodelan Bayesan Keadaan basah Keadaaan kering Semenanjung Malaysia Dissertations, Academic -- Malaysia
Issue Date:	29-Apr-2015
Description:	Kajian mengenai ciri keadaan basah dan kering di sesuatu kawasan semakin mendapat perhatian di antara penyelidik, kerana maklumat yang diberikan sangat bermanfaat kepada pelbagai pihak dalam usaha untuk mencegah dan mengurangkan kesan yang mungkin diakibatkan oleh kedua-dua fenomena alam itu. Kajian ini bertumpu kepada pencirian kebarangkalian dan peramalan jangka pendek bagi keadaan basah dan kering di Semenanjung Malaysia berdasar pada pemodelan Bayesan. Data amaun hujan mingguan (dalam mm) bagi 39 stesen hujan yang terpilih di Semenanjung Malaysia bagi tempoh 1978 hingga 2007 digunakan dalam kajian ini. Empat kaedah rawatan data lenyap telah diuji untuk menentukan kaedah yang sesuai bagi setiap stesen. Gambaran tentang keadaan basah (kering) diperoleh melalui indeks kerpasan piawai. Indeks ini mendapati bahawa sepanjang tempoh kajian, pada umumnya, Semenanjung Malaysia pernah mengalami keadaan ekstrim sama ada basah atau kering. Selain itu, kekerapan bagi ekstrim kering didapati lebih tinggi berbanding dengan ekstrim basah. Indeks ini turut menunjukkan bahawa tempoh bagi keadaan basah (kering), umumnya berlaku selama satu minggu dan semakin berlarutan, kekerapannya didapati semakin menurun. Analisis trend memperlihatkan bahawa majoriti stesen menunjukkan trend menaik bagi penunjuk kekerapan dan tempoh keadaan basah dan sebaliknya trend menurun bagi kedua-dua penunjuk keadaan kering. Sepanjang tempoh kajian, titik perubahan bererti bagi penunjuk kekerapan dan tempoh keadaan basah telah terjadi pada tahun 1994 dan 1996, manakala bagi keadaan kering telah terjadi pada tahun 1992-1993, 1997 dan 2001. Hasil pengujian bagi sifat model rantai Markov menunjukkan bahawa majoriti stesen kajian memenuhi model rantai Markov tertib satu dan pegun sama ada semasa musim monsun timur laut atau monsun barat daya. Hasil daripada model rantai Markov tertib satu dan pegun menunjukkan bahawa majoriti stesen mempunyai nilai kebarangkalian bagi berlakunya keadaan basah (kering) mengalami penurunan mengikuti darjah keadaan tersebut. Bahagian timur Semenanjung Malaysia didapati mengalami keadaan sangat basah yang lebih lama berbanding dengan keadaan sederhana basah pada musim monsun timur laut, manakala kawasan barat mengalami keadaan sangat kering lebih lama berbanding dengan keadaan sederhana kering pada musim monsun barat daya. Berdasarkan pengelompokan kabur Gustafson Kessel, didapati bahawa wilayah Semenanjung Malaysia boleh dibahagikan ke dalam enam rantau yang homogen. Model loglinear terbaik bagi keseluruhan rantau tersebut juga diperolehi dalam kajian ini melalui kaedah klasik dan Bayes. Selain pendekatan klasik bagi pemodelan rantai Markov dan loglinear untuk menggambarkan keadaan basah dan kering, pemodelan Bayesan juga digunakan. Tiga jenis taburan prior yang berbeza diambil kira, termasuk prior seragam, prior Jeffrey dan prior Bayes empirik. Hasil kajian ini menunjukkan bahawa kaedah Bayes empirik memberikan penganggar varians yang lebih kecil berbanding dengan jenis prior yang lain, kerana maklumat daripada stesen jiran telah digunakan. Berdasarkan hasil yang diperoleh melalui kaedah Bayes empirik bagi kedua-dua model tersebut, peralihan daripada keadaan basah (kering) kepada tahap yang sama lebih berkemungkinan berbanding dengan peralihan kepada tahap yang lain.,Research on characteristics of wet and dry conditions at any particular area has garnered much interest among researchers as the information provided is very useful for various parties in the effort to mitigate and prevent the impact that may be caused by both conditions. This study is focused on the probabilistic characterization and short-term forecasting of wet and dry conditions in Peninsular Malaysia based on Bayesian modelling. A weekly rainfall amount data (in mm) for the 39 selected rainfall stations in the peninsula from the period of 1978 to 2007 is used in this study. Four missing data treatments are tested to find the most suitable method for each station. Description of wet (dry) condition is obtained using Standardized Precipitation Index. The result showed that during the period of study, in general, the Peninsular Malaysia had experienced extreme weather for both wet and dry conditions. In addition to the frequency of extreme dry was found to be higher than extreme wet. The result also showed that in general, the most common length of duration of wet (dry) condition is one week, and if the duration is longest than one week, the frequency decreases. Trend analysis indicates that the majority of stations showed an upward trend for the frequency and duration indicators of wet condition, however, results showed a downward trend for both indicators of dry condition. During the study period, the significant change points for the frequency and duration indicators of wet condition were 1994 and 1996, while for dry condition in 1992- 1993, 1997 and 2001. The test result of the Markov chain properties showed that the majority of stations satisfy the first-order homogeneous Markov chain model for both the northeast and the southwest monsoon seasons. Analysis of the first-order homogeneous Markov chain models reveal that the decrease in the probability of wet (dry) for majority of stations with the degree of severity of wet (dry) condition. It is also found that the eastern part of peninsula experienced severe wet much longer than moderate wet during the northeast monsoon, while the western part experienced severe dry much longer than moderate dry during the southwest monsoon season. Based on the Gustafson-Kessel Fuzzy clustering obtained, Peninsular Malaysia could be divided into six homogeneous regions. In this study, the best loglinear model for all regions are also obtained using the classical and Bayes methods, respectively. In addition to the classical approach for Markov chain and loglinear modelling for explaining the wet and dry conditions, Bayesian modelling is also applied. Three different types of prior distributions are considered, including uniform prior, Jeffrey's prior and empirical Bayes prior. It is found that the results based on the empirical Bayes method provide the estimator with a lower variance as compared to those found based on the other types of priors, since information from neighbouring station has been utilized. Based on the results obtained using the empirical Bayes method for both models, the transition from wet (dry) condition to the same class is more probable than to the other class.,Ph.D
Pages:	302
Call Number:	QC20.7.B38 W337 2015
Publisher:	UKM, Bangi
Appears in Collections:	Faculty of Science and Technology / Fakulti Sains dan Teknologi

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.