Identifying regime switches through Bayesian wavelet estimation: evidence from flood detection in the Taquari River Valley

TL;DR

This paper introduces a Bayesian wavelet-based method for detecting regime switches in time series data, specifically applied to flood detection in the Taquari River Valley, by modeling dynamic mixture weights.

Contribution

It develops a novel Bayesian wavelet estimation approach with dynamic mixture weights, enhancing the detection of regime changes in heterogeneous time series data.

Findings

Effective in identifying flood regime shifts in the Taquari River

Demonstrates robustness through Monte Carlo simulations

Provides a flexible model for heterogeneous time series analysis

Abstract

Two-component mixture models have proved to be a powerful tool for modeling heterogeneity in several cluster analysis contexts. However, most methods based on these models assume a constant behavior for the mixture weights, which can be restrictive and unsuitable for some applications. In this paper, we relax this assumption and allow the mixture weights to vary according to the index (e.g., time) to make the model more adaptive to a broader range of data sets. We propose an efficient MCMC algorithm to jointly estimate both component parameters and dynamic weights from their posterior samples. We evaluate the method's performance by running Monte Carlo simulation studies under different scenarios for the dynamic weights. In addition, we apply the algorithm to a time series that records the level reached by a river in southern Brazil. The Taquari River is a water body whose frequent flood inundations have caused various damage to riverside communities. Implementing a dynamic mixture model allows us to properly describe the flood regimes for the areas most affected by these phenomena.