Abstract
Accurate, real-time and fine-spatial population distribution is crucial for urban planning, government management, and advertisement promotion. Limited by technics and tools, we rely on the census to obtain this information in the past, which is coarse and costly. The popularity of mobile phones gives us a new opportunity to investigate population estimation. However, real-time and accurate population estimation is still a challenging problem because of the coarse localization and complicated user behaviors. With the help of the passively collected human mobility and locations from the mobile networks including call detail records and mobility management signals, we develop a bimodal model beyond the prior work to better estimate real-time population distribution at metropolitan scales. We discuss how the estimation interval, space granularity, and data type will influence the estimation accuracy, and find the data collected from the mobility management signals with the 30 min estimation interval performs better which reduces the population estimation error by 30% in terms of Root Mean Square Error (RMSE). These results show us the great potential of using bimodal model and mobile phone data to estimate real-time population distribution.
Highlights
Knowing real-time population distribution around the city is important for many fields, such as urban planning, business location, transportation schedule and especially emergency management.If the government knew the real-time population distribution in Shanghai Bund on 31 December 2014 and took action on the crowd, many innocent tourists may have escaped the tragedy
When we scan the whole city, we find that the road network is import for human mobility and regions enclosed by it is a natural segmentation of the city
We focus on three factors, which may play an important role in the evaluation, include data type, time span, and space granularity
Summary
If the government knew the real-time population distribution in Shanghai Bund on 31 December 2014 and took action on the crowd, many innocent tourists may have escaped the tragedy. The census, with great cost in money and clerks, is still the main method to obtain population distribution. Traditional methods usually use the satellite images with dasymetric modeling tools to estimate population distribution [1]. As the state of the art works [2], researchers can obtain the population distribution by using random forest model to estimate population from satellite images. Even though, they can only update the results once a month which is far from the real-time application
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
