Multilayer perceptron-based predictive model using wavelet transform for the reconstruction of missing rainfall data

• Published in: Stochastic environmental research and risk assessment 2023-07, Vol.37 (7), p.2791-2802
• Main Authors: Narimani, Roya, Jun, Changhyun, De Michele, Carlo, Gan, Thian Yew, Nezhad, Somayeh Moghimi, Byun, Jongyun
• Format: Article
• Language: English
• Subjects: Aquatic Pollution; Chemistry and Earth Sciences; Computational Intelligence; Computer Science; Correlation coefficient; Correlation coefficients; data collection; Deforestation; Earth and Environmental Science; Earth Sciences; Environment; Estimation; Hydrologic data; Math. Appl. in Environmental Science; Mathematical analysis; Mathematical models; meteorological data; Missing data; monsoon season; Monsoons; Multilayer perceptrons; Neural networks; Original Paper; Physics; prediction; Prediction models; Probability Theory and Stochastic Processes; rain; Rainfall; Reconstruction; risk; Seasonal variations; South Korea; Statistics; Statistics for Engineering; time series analysis; Waste Water Technology; Water Management; Water Pollution Control; wavelet; Wavelet transforms; Wind
• ISSN: 1436-3240; 1436-3259
• DOI: 10.1007/s00477-023-02471-8

The quality and completeness of rainfall data is a critical aspect in time series analysis and for the prediction of future water-related disasters. An accurate estimation of missing data is essential for better rainfall prediction results. This study suggests a novel approach for estimating missing rainfall data using Multilayer Perceptron (MLP) neural networks based on three configurations that are represented by the monsoon season (MS), non-monsoon season (NMS), and non-seasonal variation. For this purpose, first, the rainfall dataset was transformed by the wavelet transform method and then, a mathematical model was created to analyze and predict the transformed data in Seoul, South Korea. Missing rainfall data in three time periods from Seoul station were reconstructed using the transformed rainfall data of the other five stations (e.g., Guroguchung, Daegokgyo, Songjeongden, Dongmakgoljuchajang, and Wallgaegyo). The results showed that using the Coiflet wavelet transform with MLP model (named Coi_MLP) estimated missing data more accurately, which is obtained from the results of statistical criteria including root mean square error, mean absolute error, and correlation coefficient of 1.18, 0.49, and 0.99 for transformed MS data and 0.76, 0.18, and 0.99 for transformed NMS data, respectively. The Coi_MLP model can effectively perform rainfall data reconstruction and predict missing rainfall data accurately, especially when the length of the statistical period is limited to the MS and NMS with different volumes of rainfall.