Using Bayes’ Theorem to Understand Uncertainty in the North American Mesoscale (NAM) Model: A Spatial Analysis of Rainfall Forecast Error for Hurricane Barry

Tropical cyclones present grave risks to coastal communities. Strong winds, storm surge, flooding, and heavy rainfall result in billions of dollars of damage to homes, infrastructure, and agricultural resources, as well as injuries and fatalities. Meteorologists use forecast models to predict the hazardous impacts of hurricanes such that these hazards can be communicated to vulnerable communities. Under some unknown conditions, there are large errors in tropical cyclones’ forecasted rainfall; because of this, it is difficult to predict rainfall impacts from landfalling hurricanes. To build more resilience in these communities, it is imperative that scientists gain insight into model uncertainty patterns, determine causes of model errors, and use those uncertainties to improve warnings to the public. This study investigates forecast model uncertainty by comparing the North American Mesoscale (NAM) model forecasted rainfall with the observed rainfall from Hurricane Barry (2018). In addition, NAM model forecast errors from 47 tropical storms are used as inputs to a Bayesian model to evaluate how the forecasted precipitation error varies spatially in 1000 hypothetical rainfall scenarios. The analysis, of the rainfall forecast error from these scenarios, consisted of three methods: 1) an analysis of the probability density function of error magnitudes forecasted by the Bayesian model, 2) a spatial, comparative analysis of mapped rainfall from the NAM model, Stage IV observations, and Bayesian model scenarios, and 3) a comparison of descriptive statistics for different vulnerable communities with boxplots. The results demonstrated that the modeled error values mostly fell within a small range; however, many simulations had outlier error values that stretched past the extreme values. Forecasted precipitation scenarios calculated from the modeled error values and the observed rainfall revealed a large range of potential rainfall amounts, as well as spatial variability of the rainfall distribution.