Browsing by Subject "Geographically Weighted Regression"

Sort by: Order: Results:

Now showing items 1-1 of 1
  • Todorovic, Sara (Helsingin yliopisto, 2020)
    Fires in residential buildings can lead to significant personal injury and property damage, especially in cities. Fire incidence has been found to have a strong connection with the characteristics of neighbourhoods and their inhabitants, such as with socioeconomic status and the features of households and buildings. However, the influencing factors are complex and often interconnected, which has made it difficult to make accurate predictions. Risk modelling and spatial data analysis provide effective and practical means of studying the phenomenon, especially from the point of view of accident prevention and preparedness. To date, knowledge of the spatial risk factors affecting residential fire incidence is yet limited in Helsinki. Thus, this study has sought to bring new empirical evidence on the matter. This study analysed residential fires in Helsinki from 2014 to 2018 at a 250 x 250 m grid level. The spatial dependence of fires was investigated by observing statistically significant clusters of fires. In this study, a risk model was created that sought to identify the underlying structural, socioeconomic, and household characteristics of neighbourhoods that affect the likelihood of residential fire incidence. The methods used were linear regression and the Geographically Weighted Regression (GWR), which takes spatial heterogeneity into account. The results showed that residential fires are spatially clustered in Helsinki. A significant large concentration of fires was found in the inner-city area and smaller concentrations in eastern Helsinki. The results indicate that the structural features of the neighbourhoods, socioeconomic status, and household circumstances have an impact on the likelihood of residential fire incidence by both increasing and decreasing the risk of fire. At the neighbourhood level, statistically significant explanatory variables that increased fire risk were population density, low education, unemployment, occupancy rate of dwellings, and home ownership. A negative relationship with fire risk was found with residential building density, age of the buildings, high education, as well as home ownership. Overall, in the study area, these eight variables explained about half of the variance of residential fire incidence. In a comparison between the models, the explanatory power of the GWR was better than linear regression, and it was also able to identify significant local variations in the effects of explanatory variables on fire risk. A comprehensive understanding of the factors influencing residential fire risk at local levels is important for rescue services, especially in terms of planning response readiness and efficient allocation of resources. In the future, more precise models should be developed in order to achieve a more comprehensive understanding of fire risk and the factors affecting it. Particular attention should be paid to the use of more precise and diverse data and methods in modelling, as well as to the temporal dimension and the consequences of fires.