Quantifying Spatial Variability of Errors in Tropical Cyclone Rainfall Forecasts

• Activity Number: 84 - Contributed Poster Presentations: Section on Statistics and the Environment
• Type: Contributed
• Date/Time: Monday, August 3, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Stephen Walsh* and Marco Ferreira and Dave Higdon and Stephanie Zick
• Companies: Virginia Tech and Virginia Tech and Virginia Tech and Virginia Tech
• Keywords: spatial statistics; hierarchical modeling; hurricanes; uncertainty quantification; precipitation

Abstract

Hurricanes and tropical storms significantly impact coastal and inland communities. In this study, we examine the error fields of hurricane precipitation forecasts and use spatial analysis to quantify uncertainty in those forecasts with maximum likelihood estimation as well as Integrated Nested Laplace Approximations (INLA). The study includes 47 storms of tropical storm strength or greater that made landfall in the contiguous U.S. between 2004 and 2017. For observations, we use the Stage IV dataset (~4-km, hourly resolution), and for the forecast, we use the ~12km North American Mesoscale (NAM) model. Both datasets are available through NOAA/NCEP. After interpolation, we study the spatial correlation structure generated by all precipitation within a 700km buffer from the eye of the storm. We estimate the range and smoothness parameters, as well as variance terms for each error field. We explore relationships between parameter estimates and the corresponding storms’ locations and intensities. We will implement a hierarchical model to supplement future storms’ forecasts with uncertainty quantification prior to landfall, with validation performed on storms with landfall in 2018-2019.