Statistical downscaling of projecting rainfall amount based on SVC-RVM model
<p>The objective of this study is to evaluate and compare the proposed statistical</p><p>downscaling model in Kelantan and Terengganu states. The study also investigates</p><p>the most accurate imputation methods in handling the m...
Saved in:
| Main Author: | |
|---|---|
| Format: | thesis |
| Language: | eng |
| Published: |
2022
|
| Subjects: | |
| Online Access: | https://ir.upsi.edu.my/detailsg.php?det=8788 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | <p>The objective of this study is to evaluate and compare the proposed statistical</p><p>downscaling model in Kelantan and Terengganu states. The study also investigates</p><p>the most accurate imputation methods in handling the missing atmospheric data and</p><p>the important predictors for a statistical downscaling method by reducing the</p><p>dimensionality data. The data used in this study include atmospheric data (predictors)</p><p>and daily rainfall data (predictand) from 1998 until 2007. As part of its methodology,</p><p>this study had used an imputation method for handling missing data. Then, Principal</p><p>Component Analysis (PCA) was applied to rectify the issue of high-dimensional data</p><p>and select predictors for a two-phase model. The two-phase machine learning</p><p>techniques were introduced as a precise statistical downscaling method in Kelantan and</p><p>Terengganu states. The first phase is a classification using the Support Vector</p><p>Classification (SVC) that determines dry and wet days. Subsequently, a regression</p><p>estimates the amount of rainfall based on the frequency of wet days using the Support</p><p>Vector Regression (SVR), Artificial Neural Network (ANN), and Relevant Vector</p><p>Machine (RVM). The proposed model was analysed by using the performance</p><p>measures that are Root Mean Square Error (RMSE) and Nash-Sutcliffe Efficiency</p><p>(NSE). The result of imputation methods shows Random Forest (RF) is having the</p><p>lowest RMSE value and the highest NSE value. The analysis of PCA results indicates</p><p>two selected Principal Components cut-off eigenvalues at 1.6 and 70.29% cumulative</p><p>percentage of the total variance. In the conclusion of this study, the comparison of</p><p>results from the SVC and RVM hybridizations reveals that the hybrid reproduces the</p><p>most reasonable daily rainfall projection and supports the high rainfall extremes,</p><p>making it a perfect candidate for rainfall prediction research. The implication of this</p><p>study is to establish the relationship between predictand variables and predictors in</p><p>order to improve predicting accuracy in climate change projections by using a</p><p>hybridization model.</p> |
|---|