Herrera-Marín et al. (2026) Characterizing Climate-Driven Shifts in Chilean Rainfall Regimes with a Hybrid Hidden Markov–Copula Framework

Identification

• Journal: Hydrology and earth system sciences
• Year: 2026
• Date: 2026-10-23
• Authors: Mauricio Herrera-Marín, Francisca Kleisinger, Diego Andres Rivera, Andrés Wilson, Álex Godoy‐Faúndez
• DOI: 10.5194/hess-30-6531-2026

Research Groups

• Research Center on Sustainability and Strategic Resource Management (CISGER), Faculty of Engineering, Universidad del Desarrollo (UDD), Santiago, Chile

Short Summary

This study introduces a hierarchical framework combining a non-homogeneous Hidden Markov Model, Dynamic Time Warping clustering, and vine copulas to characterize climate-driven shifts in Chilean rainfall regimes and provide uncertainty-aware probabilistic predictions. It reveals a significant decline in wet-state self-persistence over 1980–2022, consistent with observed drying trends, and demonstrates improved out-of-sample probabilistic forecasting skill in rainfall-active regions.

Objective

• To develop a hierarchical framework that explicitly separates large-scale rainfall regime dynamics from local spatial dependence to enable regime-aware, probabilistic rainfall prediction in Chile.

Study Configuration

• Spatial Scale: Continental Chile, 462 grid points, 0.05° resolution. Divided into 7 precipitation-coherent climatic zones.
• Temporal Scale: Daily observations for austral winter (May–August, MJJA) from 1980 to 2021 for training and 2022 for out-of-sample evaluation.

Methodology and Data

• Models used:
  • Covariate-driven non-homogeneous Hidden Markov Model (nHMM) for synoptic precipitation regimes and climate-sensitive transitions.
  • Dynamic Time Warping (DTW) clustering for delineating precipitation-coherent climatic zones.
  • State-conditional Regular Vine copulas with Generalized Pareto (GPD) tails to model residual spatial dependence and extremes.
  • Generalized Linear Mixed Models (GLMMs) for analyzing trends in annual transition proportions.
• Data sources:
  • Precipitation: CR2MET dataset (0.05° daily gridded product).
  • Atmospheric Covariates: ERA5 reanalysis (0.25° × 0.25°) for zonal and meridional wind components (850 hPa), geopotential height (500 hPa), sea-level pressure, sea surface temperature, vertically integrated moisture flux and its divergence.
  • ENSO Indices: Southern Oscillation Index (SOI), Oceanic Niño Index (ONI), and Global Mean Sea Surface Temperature (GMSST).

Main Results

• Five physically consistent rainfall regimes were identified, ranging from an anticyclonic dry state to a cyclonic wet state, confirmed by composites of mean sea-level pressure, 850 hPa winds, and 500 hPa geopotential height.
• A statistically significant decline in wet-state self-persistence of approximately 0.34 % per year (95 % CI: [0.21, 0.47] % per year; approximately 14.5 % over 1980–2022) was observed, coincident with rising GMSST.
• Out-of-sample ensembles for MJJA 2022 demonstrated good calibration across all seven DTW zones, with central 90 % prediction intervals achieving near-nominal coverage (88 %–98 %), asymmetry degrees ≤0.07, and widths increasing monotonically southward with climatological variance.
• GPD shape parameters ranged from ˆξ < 0 in anticyclonic dry states (bounded tails) to ˆξ > 0 in episodic frontal states (heavy tails), confirming state-specific extreme behavior.
• The vine-copula layer yielded statistically significant improvements over a conditionally independent baseline (Diebold–Mariano test, p < 0.001) in the rainfall-active southern zones, reducing mean areal Continuous Ranked Probability Score (CRPS) by 18 % (Zone 5) and 26 % (Zone 6). Skill scores of CRPS reached 0.51 (Zone 5) and 0.35 (Zone 6) relative to a climatological benchmark, and the area under the receiver operating characteristic curve exceeded 0.87 for the 25 mm exceedance threshold.

Contributions

• This study introduces a novel hybrid nHMM–DTW–vine–GPD framework that explicitly integrates covariate-driven regime transitions, DTW-based climatic zoning, and state-conditioned high-dimensional R-vines with GPD tail augmentation for daily precipitation totals.
• It overcomes the conditional-independence limitation of classical HMM applications by disentangling large-scale regime dynamics from residual spatial dependence, while preserving physical interpretability and achieving multivariate coherence and extreme-aware behavior.
• The framework provides meteorologically coherent states, quantifies climate-induced shifts in regime persistence (e.g., decline in wet-state self-persistence), and delivers quantitatively validated, uncertainty-aware probabilistic forecasts, offering a transferable template for hydroclimate risk management in complex regions.

Funding

• Agencia Nacional de Investigación y Desarrollo (ANID), Chile, through the ANILLO project ATE250004 (grant no. ANID/ANILLO/ATE250004).
• FONDECYT project No. 1230520 (D. Rivera).
• CISFECh (ANID/CIN250005) (D. Rivera).

Citation

@article{HerreraMarín2026Characterizing,
  author = {Herrera-Marín, Mauricio and Kleisinger, Francisca and Rivera, Diego Andres and Wilson, Andrés and Godoy‐Faúndez, Álex},
  title = {Characterizing Climate-Driven Shifts in Chilean Rainfall Regimes with a Hybrid Hidden Markov–Copula Framework},
  journal = {Hydrology and earth system sciences},
  year = {2026},
  doi = {10.5194/hess-30-6531-2026},
  url = {https://doi.org/10.5194/hess-30-6531-2026}
}