A hybrid multivariate time series m odel for forecasting m eteorological data in peninsular malaysia
An extreme rainfall event, high temperature, haze, glacier melting, rises of sea level, and droughts are as a result of climate change. The impact of climate change may result to the devastation of the earth and life. For early preparations to face the challenges of climate change, a model that can...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2018
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/81410/1/SitiMariamNorrulashikinPFS2018.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | An extreme rainfall event, high temperature, haze, glacier melting, rises of sea level, and droughts are as a result of climate change. The impact of climate change may result to the devastation of the earth and life. For early preparations to face the challenges of climate change, a model that can forecast future weather variables is needed. There exist several weather models that forecast the future atmospheric data; however, the existing models which are not station-based models, hence will have an incomplete understanding of climate system of a particular case study area. To improve on the climatic modelling, this study developed a new model where the model used data collected from Alor Setar weather stations in Peninsular M alaysia by taking into consideration all the identified dynamic features of the variables. The model is an extension of multivariate time series method, namely vector autoregressive (VAR) model. Dynamic conditional correlation (DCC) model from generalised autoregressive conditional heteroscedasticity (GARCH) model was applied in this study since weather variable has high volatility and DCC model is able to capture the volatility of the model. However, because of the high persistence in the volatility, DCC model alone is not able to capture the structural changes in the volatility. To improve on the model, a joint model with hidden Markov model (HMM) is proposed whereby HM M method will consider the structural changes in the volatility that experienced high, moderate and low volatility. The findings presented that, due to neglected of structural change in volatility, the VAR multivariate time series with the hybrid of DCC model was not able to capture closely the volatility of the weather data. Nevertheless, the proposed joint model that uses the HM M to consider the structural changes in the volatility was able to capture the degree of persistence in the weather data. The out-sample forecasting accuracy gives less than ten percent of the mean absolute percentage error (MAPE) for the proposed joint model. Simulation study proves that the VAR-HMM- DCC proposed model has better result as compare to the hybrid of the conventional VAR-DCC model. The newly joint VAR-HMM-DCC model is the contribution that provides strategies for the future forecasting weather data. |
|---|