Principal Component Analysis-Based Climatology Data for Rainfall Clustering

Abstract Global climate change has made rainfall patterns increasingly complex and difficult to predict, so a more adaptive analytical approach is needed to support climatological studies. This study aims to apply Principal Component Analysis (PCA) in forming unsupervised rainfall labels, which are then used as the basis for the rainfall pattern clustering process. Research data was obtained from the East Java Meteorology, Climatology, and Geophysics Agency (BMKG) at the Sumberpucung and Karangploso rainfall stations during the January-December 2023 period, with a total of 713 daily data. The research stages include data cleaning to overcome inconsistencies, normalization using the Standard Scaler so that all variables are on a comparable scale, dimension reduction through PCA to highlight dominant components, formation of labels based on principal component values, and grouping rainfall categories referring to BMKG classification standards. The results of the analysis showed that PCA–Clustering was able to form two main groups, namely light rain (0–20 mm/day) and moderate rain (20–50 mm/day), while the heavy rain category (> 50 mm/day) did not appear because no data were found that met this range during the observation period. Visualization of the data distribution in the main component space shows consistent cluster separation, with clearly grouped data points at the PC1 and PC2 coordinates. Quantitative evaluation using the Silhouette Score yielded a value of 0.55, which indicates that the quality of the grouping is quite good with tight intra-cluster spacing and separate interclusters. These findings show that PCA not only serves for dimension reduction, but can also be used as a basis for labeling and grouping rainfall, thus making a methodological contribution to the analysis of climatological data and potentially supporting further research in data-driven modeling, prediction, and mitigation of hydrometeorological disasters.