Scorzini et al. (2026) An integrated regionalization framework for incorporating flood seasonality into agricultural flood risk assessments

Identification

• Journal: Stochastic Environmental Research and Risk Assessment
• Year: 2026
• Date: 2026-01-01
• Authors: Anna Rita Scorzini, Charlie Dayane Paz Idarraga, Daniela Molinari
• DOI: 10.1007/s00477-025-03137-3

Research Groups

• Department of Civil, Environmental and Architectural Engineering, University of L’Aquila, Italy
• Department of Civil and Environmental Engineering, Politecnico di Milano, Italy

Short Summary

This study introduces a generalizable regionalization framework combining hydrological clustering and machine learning to incorporate seasonal flood probability into agricultural flood risk assessments, demonstrating its critical importance for accurate damage estimation and cost-benefit analyses in the Po River District, Italy.

Objective

• To develop and test a generalizable regionalization framework that integrates unsupervised clustering and supervised machine learning to infer and incorporate seasonally-resolved flood probabilities into agricultural flood risk assessments, particularly for ungauged catchments.

Study Configuration

• Spatial Scale: Po River District, Northern Italy (over 70,000 km²), with an illustrative case study in the Panaro-Reno Area of Potential Significant Flood Risk (APSFRs).
• Temporal Scale: Monthly flood probability profiles derived from a minimum of 20 years of hydrometric data; expected annual losses (EAL) calculated over a 50-year service life for mitigation measures.

Methodology and Data

• Models used:
  • Unsupervised hierarchical clustering (Ward linkage, Euclidean distance) for grouping gauged catchments into seasonal flood regimes.
  • Supervised machine learning classifiers: Support Vector Machine (SVM), Random Forest (RF), and K-Nearest Neighbors (KNN), with an ensemble-consensus strategy for regionalization to ungauged catchments.
  • Crop damage model: AGRIDE-c (Molinari et al. 2019), which accounts for time-varying crop vulnerability based on phenological stages.
• Data sources:
  • Monthly flow data (discharge or hydrometric level) from 120 gauged stations in the Po River District.
  • Physical catchment attributes: Italy’s 10-meter resolution Digital Elevation Model (DEM), CORINE Land Cover dataset, and national dam register.
  • Spatially detailed crop data from the agricultural cadaster of the Emilia-Romagna region.
  • High-resolution inundation maps (10 m for Reno, 5 m for Panaro) for 25, 100, and 500-year return period flood scenarios, provided by the Po River District Authority.

Main Results

• Four distinct seasonal flood regime clusters were identified: autumn peaks (C1), winter peaks (C2), summer peaks (C3), and spring/spring-autumn peaks (C4).
• The Support Vector Machine (SVM) classifier achieved the best performance for regionalizing flood seasonality, with an average classification accuracy of 0.64 and precision of 0.58 on the test set.
• Spatially distributed maps of flood seasonality for the Po River District revealed clear zonation, with autumn peaks in Apennine areas, winter peaks in the central Po Plain and inner Alpine slopes, and spring peaks in high-altitude northern Alps.
• Variable importance analysis consistently highlighted elevation-related descriptors (mean altitude, relief) and land cover (forest cover) as key predictors of flood seasonality.
• Accounting for flood seasonality (Lave) significantly reduced Expected Annual Loss (EAL) compared to using maximum monthly loss (Lmax). For the Panaro-Reno system, EAL was approximately €435,000 (Lave) versus €750,000 (Lmax) for the considered scenarios.
• This difference substantially impacted cost-benefit analyses: economically feasible investments for flood protection (BCR > 1) ranged from €7 million to €12 million using Lave, compared to €13 million to €20 million using Lmax (for a 50-year service life and 3-5% discount rates).
• Uncertainties from the regionalization step, such as misclassification between adjacent seasonal regimes, were shown to propagate and slightly affect economic indicators like the Benefit-Cost Ratio (BCR).

Contributions

• Introduces a generalizable, scalable, and adaptable regionalization framework for incorporating flood seasonality into agricultural risk assessments, particularly in data-scarce regions.
• Integrates unsupervised clustering of hydrological signatures with supervised machine learning to infer and spatially transfer monthly-resolved flood probabilities to ungauged catchments.
• Links traditionally separate domains of seasonal hazard information, impact assessment, and economic evaluations, leading to more physically consistent and policy-relevant flood risk estimates.
• Quantitatively demonstrates the critical importance of accounting for flood seasonality and crop phenology in agricultural flood damage modeling and cost-benefit analyses, preventing overestimated damage projections and suboptimal mitigation strategies.

Funding

• Università degli Studi dell’Aquila (CRUI-CARE Agreement)
• Autorità di Bacino Distrettuale del Fiume Po (MOVIDA agreement)

Citation

@article{Scorzini2026integrated,
  author = {Scorzini, Anna Rita and Idarraga, Charlie Dayane Paz and Molinari, Daniela},
  title = {An integrated regionalization framework for incorporating flood seasonality into agricultural flood risk assessments},
  journal = {Stochastic Environmental Research and Risk Assessment},
  year = {2026},
  doi = {10.1007/s00477-025-03137-3},
  url = {https://doi.org/10.1007/s00477-025-03137-3}
}