Title: Mixed-frequency extreme value regression: Estimating the effect of MCS on extreme rainfall in the Midwest
Authors: Luca Trapin - University of Bologna (Italy) [presenting]
Debbie Dupuis - HEC Montreal (Canada)
Abstract: More frequent and longer-lasting mesoscale convective systems (MCS) are the principal driver of observed increases in springtime extreme rainfall in the Central United States. We develop new models that integrate and exploit hourly MCS information in analyses of extremes of maximum hourly rainfall over a much longer time period, e.g. one month, and gain some insight into the increases to these extremes and how MCS may have driven the changes. This requires extreme value regression models handling observations sampled at different frequencies. We borrow some elements from the MIxed-DAta Sampling (MIDAS) regression literature and propose a flexible, data-driven aggregation scheme to face this challenge. We study the monthly maximum hourly precipitation in five US midwest cities from 1979 to 2014. We model these maxima with a Generalized Extreme Value (GEV) distribution, and let the location parameter of this model vary as a function of the monthly number of MCS occurring in each of the 24 hours covering a day. Our mixed-frequency GEV model confirms that the occurrence of an MCS is a good predictor of the extreme rainfall, also reveals that MCS occurring in different parts of the day contribute differently to explain the increased rainfall intensity.