Geographia Technica, Vol 22, Issue 1, 2027, pp. 44-63
UNRAVELING URBAN RAINFALL DYNAMICS: A 161-YEAR BAYESIAN DECOMPOSITION OF ABRUPT CHANGES, SEASONALITY, AND TRENDS IN JAKARTA, INDONESIA
Fadilah Karamun Nisaa NADIYAH 
, Sri NURDIATI , I Wayan MANGKU , Mohamad Khoirun NAJIB , Syukri Arif RAFHIDA
ABSTRACT: Rainfall variability in tropical urban regions remains challenging to characterize due to its strong spatiotemporal complexity and non-stationary behavior. This study investigates long-term rainfall dynamics in Jakarta, Indonesia, using the Bayesian Estimator of Abrupt Change, Seasonality, and Trend (BEAST) to analyze a 161-year monthly rainfall time series (1864–2024). The BEAST framework was used to decompose the rainfall series into trend, seasonal, and outlier components, allowing persistent structural changes to be distinguished from isolated extreme observations. The results indicate substantial temporal variability in Jakarta’s rainfall regime. The most strongly supported candidate trend changepoint occurred around January 2020, with a posterior probability of 83.56% and an estimated abrupt change of 121.02 mm. The seasonal component exhibited its strongest candidate changepoints around January 2014 and January 2016, with posterior probabilities of 29.3% and 21.16%, respectively. The outlier component identified several isolated extreme observations, including January 1979, with an outlier changepoint probability of 92.49%, followed by January 1963 (82.72%), January 2002 (68.11%), January 1872 (67.92%), January 1965 (61.83%), and February 2020 (56.63%). Sensitivity analysis showed that explicitly accounting for isolated extreme observations increased the posterior support for the January 2020 trend changepoint from 71.04% to 83.56%, while its timing remained stable. These findings demonstrate that extreme observations can influence the strength of inferred structural changes and highlight the importance of probabilistic, non-stationary approaches for characterizing long-term rainfall dynamics in rapidly urbanizing tropical regions.
Keywords: Bayesian Estimator of Abrupt Change, Seasonality, and Trend (BEAST); Changepoint detection; Rainfall variability; Outlier detection; Jakarta.
Full article here