A frailty-contagion model for multi-site hourly precipitation driven by atmospheric covariates

TL;DR

This paper introduces a new multi-site hourly precipitation model inspired by frailty-contagion methods, capable of capturing dry spells and heavy rainfall using atmospheric covariates, with demonstrated effectiveness on real French data.

Contribution

It presents a novel, simple dynamical model for hourly precipitation that incorporates atmospheric variables and handles complex precipitation patterns.

Findings

Model effectively captures dry and heavy rainfall periods.

Inference approach validated on simulated data.

Applied successfully to French Brittany measurements.

Abstract

Accurate stochastic simulations of hourly precipitation are needed for impact studies at local spatial scales. Statistically, hourly precipitation data represent a difficult challenge. They are non-negative, skewed, heavy tailed, contain a lot of zeros (dry hours) and they have complex temporal structures (e.g., long persistence of dry episodes). Inspired by frailty-contagion approaches used in finance and insurance, we propose a multi-site precipitation simulator that, given appropriate regional atmospheric variables, can simultaneously handle dry events and heavy rainfall periods. One advantage of our model is its conceptual simplicity in its dynamical structure. In particular, the temporal variability is represented by a common factor based on a few classical atmospheric covariates like temperatures, pressures and others. Our inference approach is tested on simulated data and applied on measurements made in the northern part of French Brittany.