Granata et al. (2025) The anatomy of drought in Italy: statistical signatures, spatiotemporal persistence, and forecasting potential

Identification

• Journal: Journal of Hydrology
• Year: 2025
• Date: 2025-10-17
• Authors: Francesco Granata, Fabio Di Nunno
• DOI: 10.1016/j.jhydrol.2025.134428

Research Groups

Department of Civil and Mechanical Engineering, University of Cassino and Southern Lazio, Cassino, Italy

Short Summary

This study comprehensively analyzes six-month Standardized Precipitation–Evapotranspiration Index (SPEI-6) time series across Italy using advanced statistical, persistence, clustering, and deep learning methods to characterize drought patterns and improve forecasting, revealing a tripartite drought structure and regional forecasting skill.

Objective

• To conduct a comprehensive analysis of six-month Standardized Precipitation–Evapotranspiration Index (SPEI-6) time series across Italy, integrating higher-order statistical descriptors, persistence diagnostics, advanced clustering algorithms, and deep learning forecasting to characterize drought and assess forecasting potential.

Study Configuration

• Spatial Scale: National (Italy)
• Temporal Scale: Multi-temporal (6-month SPEI index accumulation, 1-month lead and seasonal horizon forecasts)

Methodology and Data

• Models used: Kolmogorov–Arnold Fourier (KAF) networks, Long Short-Term Memory (LSTM) architectures, K-means, Agglomerative Hierarchical, Gaussian Mixture Models, Spectral Clustering, Hurst exponent (H), Detrended Fluctuation Analysis (DFA).
• Data sources: Six-month Standardized Precipitation–Evapotranspiration Index (SPEI-6) time series.

Main Results

• The analysis emphasized skewness and other higher-order moments to capture asymmetries in drought intensity and frequency, and employed scaling metrics to quantify long-range dependence and memory in hydroclimatic signals.
• Clustering approaches delineated a coherent tripartite drought structure across Italy: a persistent southern and insular regime with strong temporal memory, an intermediate northeastern corridor with moderate persistence, and a volatile northwestern Alpine domain with weak persistence and high variability.
• Kolmogorov–Arnold Fourier (KAF) networks showed substantial forecasting skill at a one-month lead, particularly in persistent southern and insular regions, outperforming LSTM architectures.
• Forecasting performance declined at seasonal horizons and in highly variable northern areas.

Contributions

• The study provides a comprehensive analysis of drought in Italy, moving beyond conventional mean-variance assessments by integrating higher-order statistical descriptors and persistence diagnostics.
• It delineates a coherent tripartite drought structure for Italy, offering a novel regionalization based on drought characteristics and persistence.
• It demonstrates the potential of Kolmogorov–Arnold Fourier (KAF) networks for drought forecasting, particularly at short leads in persistent regions.
• The presented methodological framework is modular and transferable, offering a rigorous template for drought diagnosis and early warning in other drought-prone regions.

Funding

• Not specified in the provided text.

Citation

@article{Granata2025anatomy,
  author = {Granata, Francesco and Nunno, Fabio Di},
  title = {The anatomy of drought in Italy: statistical signatures, spatiotemporal persistence, and forecasting potential},
  journal = {Journal of Hydrology},
  year = {2025},
  doi = {10.1016/j.jhydrol.2025.134428},
  url = {https://doi.org/10.1016/j.jhydrol.2025.134428}
}