Spatiotemporal Generalized Additive Model: Investigating Weather-Related Road Crashes in Southern Finland

Road crashes pose a serious safety risk, particularly under adverse weather conditions. Having a deeper understanding of crash patterns and underpinning their connection to different meteorological factors is useful for targeted safety interventions. Many types of statistical and machine learning models seek to quantify the relationship between different meteorological parameters and accident risk. This thesis presents a spatiotemporal generalized additive model to explain which weather conditions increase the risk of a crash in the Finnish regions of Uusimaa and Varsinais-Suomi. The work also explores the spatial and temporal trends which are associated with a heightened probability of a car accident. The emphasis throughout this work was on carefully engineering the model by selecting an appropriate temporal and spatial granularity at which to perform the analysis. Incorporating a thoughtful study design and data aggregation procedure was paramount. Ultimately, the model assigns fitted probabilities for a combination of a smaller spatial unit at a specific hourly time, ranging from March 2017 to December 2021. The model employs MetCoOp Ensemble Prediction System (MEPS) data which was obtained from the Finnish Meteorological Institute. The constructed model explained 33.8% of the deviance and had good fit as per diagnostic plots of the randomized quantile residuals. The model indicates that snow and sleet increase the log-odds of a crash. Other factors such as rush hour and the fact that a crash happened nearby in the last two hours also added explanatory power to the model. The highest probability of a car crash happens around the Helsinki and Turku regions.

Grantor dc:publisher

Helsingin yliopisto

Year dc:date.issued

2024

Author dc:creator

• Sebag, Etienne

Contributors dc:contributor

• Helsingin yliopisto, Matemaattis-luonnontieteellinen tiedekunta
• University of Helsinki, Faculty of Science
• Helsingfors universitet, Matematisk-naturvetenskapliga fakulteten

• Generalized Additive Model
• Spatiotemporal Analysis
• Road Crashes
• MEPS forecast data

Language dc:language.iso

eng

Identifier URI

URN:NBN:fi:hulib-202410104267

OAI identifier oai:identifier

oai:helda.helsinki.fi:10138/586818