Leveraging Automation in Data Enhancement and Quality Control Protocols for Post-Windstorm Reconnaissance Data

Abstract

Extreme windstorms cause loss of life, major financial losses, and disrupt community well-being. Post-windstorm field reconnaissance is one of the traditional methods used to collect data about these natural hazards. The analysis of this data helps us better understand these disasters and improve the performance of buildings and other infrastructure during such extreme events. The raw data collected from reconnaissance missions is often fragmented and nonuniform, which necessitates formalizing data enhancement and quality control protocols that increase the completeness and accuracy of these datasets. This thesis presents data enhancement and quality control protocols for ensuring that the building performance datasets are accurate, complete, and standardized, making them suitable for analysis and further use by the natural hazards engineering community. However, this data enhancement and quality control process can take months to complete, delaying data analysis. This thesis also demonstrates a preliminary framework for automating key components of the data enhancement and quality control process to reduce the time required. The automation framework uses modern technologies that include web scraping and the use of established machine learning models from past works to classify damage based on imagery. The results of comparing the human approach for the data enhancement and quality control process to the automation framework shows promise in incorporating automation in post-windstorm field reconnaissance. While the accuracy of the framework is not as high as the human approach, continuous improvements can be made to the individual components of the framework to increase the accuracy. Ultimately, this approach aims to ensure high quality, standardized post-windstorm reconnaissance datasets can be generated and published rapidly after extreme windstorms and used to strengthen the resilience of at-risk communities.