Predicting property damage from tornadoes with zero-inflated neural networks

Jeremy Diaz,Maxwell B Joseph

doi:10.1016/j.wace.2019.100216

Jeremy Diaz, Maxwell B Joseph

Open Access

https://doi.org/10.1016/j.wace.2019.100216

Copy DOI

Abstract

Tornadoes are the most violent of all atmospheric storms. In a typical year, the United States experiences hundreds of tornadoes with associated damages on the order of one billion dollars. Community preparation and resilience would benefit from accurate predictions of these economic losses, particularly as populations in tornado-prone areas increase in density and extent. Here, we use a zero-inflated modeling approach and artificial neural networks to predict tornado-induced property damage using publicly available data. We developed a neural network that predicts whether a tornado will cause property damage (out-of-sample accuracy = 0.821 and area under the receiver operating characteristic curve, AUROC, = 0.872). Conditional on a tornado causing damage, another neural network predicts the amount of damage (out-of-sample mean squared error = 0.0918 and R2 = 0.432). When used together, these two models function as a zero-inflated log-normal regression with hidden layers. From the best-performing models, we provide static and interactive gridded maps of monthly predicted probabilities of damage and property damages for the year 2019. Two primary weaknesses include (1) model fitting requires log-scale data which leads to large natural-scale residuals and (2) beginning tornado coordinates were utilized rather than tornado paths. Ultimately, this is the first known study to directly model tornado-induced property damages, and all data, code, and tools are publicly available. The predictive capacity of this model along with an interactive interface may provide an opportunity for science-informed tornado disaster planning.

Highlights

The United States experiences more tornadoes every year than any other country in the world, with the annual average of cumulative tornado-induced property damage at nearly one billion US dollars (Changnon, 2009)
The damages resulting from tornadoes is a function of the physical properties of storms and societal factors such as population density, property values, and quality of building materials (Kunkel et al, 1999; Changnon et al, 2000; American Meteorological Society)
In addition to in-print visualizations, we present prototype dashboards that communicate these predictions accessibly and in a way that could be used by planners without expertise in machine learning or spatial analysis (Andre and Smith, 2003)

Summary

Introduction

The United States experiences more tornadoes every year than any other country in the world, with the annual average of cumulative tornado-induced property damage at nearly one billion US dollars (Changnon, 2009). Independent of physical changes in the number, distribution, or intensity of tornadoes, increasing property values, population density, and manufactured home density may have contributed to increases in tornado damages in recent decades (Kunkel et al, 1999; Changnon et al, 2000; American Meteorological Society; Kellner and Dev, 2014; Ashley and Strader, 2016; Ashley et al, 2014) These societal factors may be useful for predicting future tornado damages under different scenarios, with applications to development planning (Godschalk, 2003), natural disaster asset prepositioning (Salmerón and Apte, 2010), refinement of public warning systems (Stensrud et al, 2009), the property-casualty insurance industry (Changnon, 2009), and disaster response coordination (McEntire, 2002). We use artificial neural networks to predict tornado-induced property damage over 22 of the most recent years (1997–2018) as a function of explanatory variables identified in the

Methods

Results

Discussion

Conclusion