Date of Graduation

5-2025

Document Type

Thesis

Degree Name

Master of Science in Geology (MS)

Degree Level

Graduate

Department

Geosciences

Advisor/Mentor

Cheng, Linyin

Committee Member

Peter, Brad G.

Second Committee Member

He, Yaqian

Keywords

Geospatial modeling; GIS; Hazards; Meteorology; Property Damage; Winter storms

Abstract

Storm Data, a monthly publication by the National Weather Service with records of weather events, reports on property damage caused by natural hazards. Since 1996, the database has included winter-related events. While the property damage data is considered incomplete, no other unique, public datasets report winter-related property damage at the county level. From 1996 to 2024, Arkansas ranked #1 in winter-related property damage. To understand the factors that are driving property damage and potentially reduce Arkansas’ vulnerability to it, three groups separated by temporal periods, i.e., “2000” (01/1996 - 06/2005), “2010” (07/2005 - 06/2015), and “2020” (07/2015 - 06/2024) are examined with geospatial modeling tools. The dependent variable is calculated as the total winter-related property damage during the period divided by the decade’s population, referred to as “damage per capita”. Sixteen potentially influential variables, including storm characteristics, demographic information, environmental data, and climatological data, are identified. Then, using the most influential variables for each group, the performance of ordinary least squares, spatial-lag, spatial-error, geographically weighted regression, and multi-scale geographically weighted regression models are compared. Lastly, using the most appropriate model, the variables’ influence on damage per capita is quantified by analyzing model coefficients for each group. Results show that different factors contributed to winter-related property damage per capita over space and time in Arkansas. Accounting bias, calculated as the percentage of reports with non-zero property damages, positively influenced all three models, serving as the most influential factor for the 2020 model. The 2010 model was most significantly influenced positively by the number of hours and the presence of deciduous forests and negatively by the median income, mean precipitation, and mean snowfall. The 2000 model was the best performing group, but its coefficients were only significant over small portions of the state. In southwest Arkansas, coefficients were negative and positive for the number of heavy snow reports and mean snowfall, respectively. Coefficients were negative for the number of hours in west-central Arkansas and positive for accounting bias along the western third of the state. These results are used to make recommendations to policymakers and emergency managers.

Download

Included in

Geography Commons

Share

COinS