Abstract Flood risk to agriculture is strongly influenced by the timing of inundation relative to crop development stages, making flood seasonality a critical but often overlooked component in damage estimation. This study introduces a generalizable regionalization framework that combines hydrological clustering and machine learning to incorporate seasonal flood probability into agricultural risk assessment. The approach involves identifying clusters of gauged catchments with similar patterns of intra-annual flood occurrence and using supervised classification to extrapolate these seasonal regimes to ungauged catchments based on their physical attributes. The resulting spatially distributed maps of monthly flood probability can be then integrated with a flood damage model to calculate expected annual losses and support risk estimates across entire river districts. The proposed framework, applied in this study to the Po River District (Italy) for illustrative purposes, is scalable and adaptable to different regions, contributing to more robust and context-sensitive adaptation planning in agriculture. Results highlight the importance of accounting for flood seasonality in cost-benefit analyses within agricultural contexts, as neglecting intra-annual variability can lead to overestimated damage projections and suboptimal mitigation strategies.

Abstract The study of record-breaking values is of significant interest in environmental sciences. Studying records implies analyzing both their occurrence and their magnitude. Further, the study of this phenomenon within a spatio-temporal framework is vital for evaluating seasonal behaviors, identifying spatial patterns, and quantifying the effect of climate change on it. With interest in record-breaking temperatures, we specify models for these observations rather than models for the entire daily temperature stream. Models specifically designed for record-breaking events must consider two random components: the occurrence and the magnitude of each record. With primary interest in the magnitudes, we model the magnitude data given the occurrence data, with the goal of making inference about their evolution within a spatio-temporal framework. We employ a set of 40 geo-referenced time series of daily temperatures across peninsular Spain. From these, we extract the series of occurrences and values of record-breaking events during the summer months, June, July, and August, spanning from 1960 to 2021. The results reveal that the behavior of the increments is neither spatially nor temporally homogeneous, and that there is significant dependence on the previous day: the occurrence of a record increases the posterior mean of the next day’s increment by between 0.3 and 0.6 °C. It is also found that the posterior mean of the average increment on a record-breaking day during the decade 2012–2021 is approximately 1 °C inland, increasing to around 2°C in some coastal areas. After 30 years, mean increments stabilize near 1°C with a mild downward trend.