Sensitivity of wind residual correlation models to residual decomposition and semivariogram assumptions for U.S. landfalling tropical cyclones

• Investigated the impacts of modeling assumptions on characterizing tropical cyclone wind speed residual spatial correlation. • Found substantial variability in spatial correlation structure across tropical cyclones. • Spatial correlation models are sensitive to the selection of statistical methods and model forms. • Developed mean and intensity-based correlation models for wind risk assessment of distributed systems. Risk assessment for spatially distributed infrastructure systems under tropical cyclone-induced strong winds requires models of the spatial correlation of wind speed residuals. This study characterizes this correlation by calculating residuals and semivariograms for sixteen tropical cyclones landfalling in the United States. We use the H*wind dataset as observational wind speeds and estimates from the HAZUS hurricane model as predicted wind speeds. To analyze the impact of modeling assumptions on the correlation structure, we considered two methods for wind speed residual decomposition and used classic and robust semivariogram estimators. We further considered three correlation model functional forms, along with weighted and non-weighted methods for fitting correlation models to the semivariogram estimates. The results show substantial variability among residual correlation structures across tropical cyclones, with varying sensitivity to modeling assumptions. Finally, we developed generalized mean and intensity-based correlation models to characterize the spatial correlation of wind speed residuals, both in terms of overall structure and the influence of storm intensity.