Probabilistic rainfall generator for tropical cyclones affecting Louisiana

This study focuses on the development of a probabilistic rainfall generator for tropical cyclones (TCs) affecting Louisiana. We consider 12 storms making landfall along the Louisiana coast during 2002–2017 and generate ensembles of high‐resolution (~5 km and 20 min) TC‐rainfall fields for each storm. We develop a data‐driven multiplicative model, relating observed rainfall to the rainfall obtained from a parametric TC rainfall model (Interagency Performance Evaluation Task Force Rainfall Analysis [IPET]) through the product of a deterministic and a stochastic component; the former accounts for rain‐dependent biases, while the latter for the stochastic nature of the rainfall processes. As a preliminary step, we describe the overall bias of the IPET model as a function of total TC rainfall within the state and maximum wind speed at landfall. We then estimate the rain‐dependent bias using a cubic spline. Finally, we characterize the random errors in terms of their probability distribution and spatial correlation. We show that the marginal distribution of the logarithm of the random errors can be described by a mixture of four Gaussian distributions, and its spatial correlation is estimated based on the nonparametric Kendall's τ. We then present a methodology to generate ensembles of random fields with the specified statistical properties. Here, the generation of probabilistic rainfall comes from the statistical modelling of the uncertainties between IPET rainfall and observations. While these results are valid for Louisiana and the IPET model, the methodology can be generalized to other parametric rainfall models and regions, and it represents a viable tool to improve our quantification of the risk associated with TC rainfall. We have developed a probabilistic rainfall generator for tropical cyclones. It is a data‐driven model with a deterministic and a stochastic component. While it is applied to Louisiana, the methodology is general and can be applied widely.