Statistical Clustering of Heavy Precipitation Radar Images in Surabaya Using Gaussian Mixture Model

Ferawati, Kiki (2018) Statistical Clustering of Heavy Precipitation Radar Images in Surabaya Using Gaussian Mixture Model. Masters thesis, Institut Teknologi Sepuluh Nopember.

[thumbnail of 06211650010023-Master_Thesis.pdf]
06211650010023-Master_Thesis.pdf - Accepted Version

Download (3MB) | Preview


Precipitation in Indonesia is affected by a wide range of weather variability. Understanding the characteristics of precipitation in the area is essential in order to predict heavy precipitation event. Characteristics of precipitation, e.g. its shape and pattern are important feature to predict extreme rainfall events obtained from radar images. This study applied the Gaussian Mixture Model (GMM) for high dimensional data clustering (hereafter denoted as HDDC) to cluster the shapes appearing in the radar images associated with heavy precipitation events in Surabaya. Another method used for this analysis is K-means clustering with principal component analysis (PCA). Using ITS precipitation data, the Hill Plot and Mean Residual Life Plot (MRLP) suggested that the extreme event is characterized with the precipitation above 1.5 mm per ten minutes. According to the Bayesian Information Criterion (BIC), the HDDC suggested 10 clusters to characterize the heavy precipitation patterns. Another clustering method, K-means with PCA is also applied to the data. However, out of the 10 clusters, several clusters show similar pattern, suggesting that 10 clusters are too many for the data. Reviewing the value of Pseudo-F and Silhouette of K-means and the BIC value of HDDC, 2 clusters are deemed best for radar images data. The analysis for both K-means and HDDC shows some inconsistency in terms of the cluster members, due to the small sample size. Hence, ensemble-based HDDC is proposed to overcome the problem. This method generated better results with robust cluster. It resulted in two clusters representing the pattern of precipitation system in Surabaya.

Item Type: Thesis (Masters)
Uncontrolled Keywords: Radar image, Heavy precipitation, Cluster, Gaussian Mixture Model, K-means
Subjects: Q Science > QA Mathematics > QA278.55 Cluster analysis
Divisions: Faculty of Mathematics, Computation, and Data Science > Statistics > 49101-(S2) Master Thesis
Depositing User: Ferawati Kiki
Date Deposited: 19 Jun 2021 12:08
Last Modified: 19 Jun 2021 12:08

Actions (login required)

View Item View Item