Estimating missing daily rainfall data via artificial neural network over peninsular Malaysia

• Main Author: Loh, Wing Son
• Format: Final Year Project / Dissertation / Thesis
• Published: 2021
• Subjects: QA Mathematics
• Online Access: http://eprints.utar.edu.my/4980/1/MEME15110_Project_LohWingSon_20UEM00089.pdf; http://eprints.utar.edu.my/4980/

The presence of missing rainfall data has always known to be an obstacle for rain gauge stations to preserve a serially complete real time rainfall database. Various techniques were implemented in dealing with missing rainfall data in the past but artificial neural network (ANN) models have also gradually earned much renown due to its promising estimation results. The Self-Organising Feature Map (SOFM), a type of ANN was proposed in this research to account for the missing daily rainfall values and the complex dynamics of rainfall over Peninsular Malaysia. SOFM was applied in two stages for which the first stage was to train the SOFM model using the complete daily rainfall data and the second stage was to apply the trained SOFM to estimate the missing daily rainfall data. The estimated results were then compared and contrast by setting up different proportion of 10%, 20%, and 30% for the missing daily rainfall data. Ten different rainfall stations distributed over the Peninsular Malaysia were studied. The daily rainfall data for the North-East monsoon (NEM) season from the rainfall stations were obtained to assess the performance of the SOFM in describing the spatial relationship of the rainfall events as well as in estimating the missing daily rainfall data. The mean error (ME) and root mean square error (RMSE) were computed to evaluate the missing daily rainfall data estimated by the SOFM model. The analysis for all of the three different proportion of missing daily rainfall data suggested that each of the rainfall stations possess distinctive rainfall patterns. The SOFM has also provided reasonable estimates for the missing daily rainfall data.