Accurate knowledge and identification of the spatial distribution of fire incidents are vital for effective urban fire safety management. While the complexity of the physical environment poses challenges in obtaining precise fire risk distribution. To address this, we propose a novel approach that leverages the species distribution prediction model, focus on examining the spatial association between fires and physical environment, as well as the application of niche-based model for fire risk prediction. By utilizing the local co-localization quotient (LCLQ), we found that 8 out of 9 types of point of interest (POI) in Beijing, China, exhibit a spatial association with fire incident locations. Based on this relationship, we incorporate the impactful POI and fire data into MaxEnt, a species distribution prediction model, to predict the distribution of fire hazards. Additionally, we perform a comprehensive fire risk assessment by incorporating diurnal population data. Our experimental results demonstrate that the MaxEnt model achieved a maximum AUC value of 0.78 when employing a POI density search radius of 500 m. Furthermore, the high-resolution results show good integration with other geographical data. The utilization of LCLQ and niche-based model improves the efficiency and resolution of fire risk assessment, enhancing fire risk assessment practices and management.