Modified singular spectrum analysis in identifying temporal torrential rainfall patterns over Peninsular Malaysia

Shazlyn Milleana Shaharudin

Abstract

Identifying local time scale to determine when the torrential rainfall events occur at a particular location is critically important. This can be detected by observing the trend which is characterized by the shape of time series data in order to detect the abnormally heavy rainfall that can cause torrential rainfall events. One of the methods in identifying the range of local time scale according to the trend is based on Singular Spectrum Analysis (SSA).The variations in the time series observations can be decomposed and reconstructed to locate the time period in which the extreme rainfall events occur when using SSA approach. However, in torrential rainfall time series data, the daily amount of rainfall is approximately similar over a period of time. This situation leads to a problem when using SSA since there are coinciding singular values. It might cause disjoint sets of singular values and different series components to mix up with each other. Besides, using SSA, leading component from the eigen time series are usually assessed subjectively by graphical inference of the eigenvector plots, lacking a specific statistical measure. In consequence of both issues, the extracted trend from SSA tends to flatten out and does not show any distinct pattern. Thus, the aim of this study is to propose (i) the adjustments on the coinciding singular values obtained from decomposed time series matrix based on a restricted singular value decomposition (ii) a guided clustering method in discriminating the eigenvectors from this iterative procedure. For the results, the extracted trend using modification of SSA is expected to appear fit to the original time series and looks more flexible compared to SSA. The outputs can be beneficial to hydrologists to recommend actions in mitigating the extreme rainfall events and taking necessary precautions when they happen.

References

Githui F.W., Opere A., Bauwens, W. (2003). Statistical and trend analysis of rainfall and river discharge: Yala River basin, Kenya. In Proc. International Conference of UNESCO. Netherlands.

Khaleghi, M., Zeinivand, H., Moradipour, S. (2014). Rainfall and river discharge trend analysis: a case study of Jajrood watershed, Iran. International Bulletin of Water Resources and Development. 2(3), 8-15.

Mondal, R.A., Kundu, S., Mukhopadhyay, A. (2012). Rainfall trend analysis by Mann-Kendall test: a case study of North-Eastern part of Cuttack district, Orissa. International Journal of Geology, Earth and Environmental Sciences. 2(1), 700-78.

Rodrigues, P.C., Tuy, P.G.S.E., Mahmoudvand, R. (2018). Randomized singular spectrum analysis for long time series. Journal of Statistical Computation and Simulation. 88(10), 1921-1935.

Leles, M.C.R., Cardoso, A.S.V., Moreira, M.G., Sbruzzi, E.F., Nascimento, C.L., Guimaraes, H.N. (2018). A singular spectrum analysis based trend-following trading system. In Proc. 2018 Annual IEEE International Systems Conference. Canada.

Leles, M.C.R., Sansao, J.P.H., Mozalli, L.A., Guimaraes, H.N. (2018). Improving reconstruction of time-series based in singular spectrum analysis: A segmentation approach. SoftwareX. 77, 63-76. Rainfall and river discharge trend analysis: a case study of Jajrood watershed, Iran.

Alexandrov, T., Golyandina, N.E., Spirov, A. (2008). Singular spectrum analysis of gene expression profiles of early drosophila embryo: exponential-in-distance patterns.Research Letters in Signal Processing. 2008(2008),1-5.

Zvonarev, N., Golyandina, N.E. (2017). Iterative algorithms for weighted and unweighted finite-rank time eries approximations. Statistics its interface. DOI: 10.4310/SII.2017.v10.n1.a1 • Source: arXiv,

Golyandina, N., Shlemov, A. (2014). Variations of singular spectrum analysis for separability improvement: non-orthogonal decompositions of time series. Statistics and Its Interface. 1-15.

Kondo, Y., Salibian-Barrera, M., Zamar, R. A robust and sparse k-means clustering algorithm. arXiv:1201.6082v1.