The frequency and intensity of flooding are increasing due to climate change [1
]. According to data from EM-DAT, floods accounted for 49% of global disasters and 68% of the affected population in 2019 [3
]. Urbanization has increased the risk of urban flooding by changing topography and geological conditions, both of which influence hydrological processes [4
]. Therefore, floods have become the main natural disaster in many cities, which are characterized by a large population and rapid urbanization, posing serious threats to human life, production, and social and economic activities [6
]. As such, accurate flood risk assessments for cities are vital for formulating local disaster prevention policies.
Population is the most important flood-hit object; thus, accurate population-based spatial distribution information is an important basis for disaster prevention and mitigation [7
]. The population affected by urban floods is typically calculated by superimposing population data sets on flood information. Therefore, population mapping is key for disaster assessment calculation, with higher-resolution population datasets more accurately estimating the size and spatial distribution of affected populations [8
], which greatly improves the rationality of flood-related decision-making [9
]. Common population datasets include census data provided by the government, with administrative divisions as statistical units; however, census data have a long update cycle and suffer from the modifiable areal unit problem (MAUP) when calculations are based on GIS [10
]. In addition, census data based on administrative units ignore the spatial heterogeneity of population distributions, resulting in an inaccurate risk assessment. Therefore, it is difficult for traditional census data to mine microscopic dynamic laws, discover new phenomena [12
], or meet the increasing requirements for accurate urban computing. To solve this problem, dynamic mapping is applied to demographic data spatialization [13
], which maps demographic data according to the unit of the administrative region to a geographic grid of a certain size. Common methods include the average distribution method, spatial interpolation method, population distribution model, correlation analysis method, and remote sensing estimation. Population distribution data based on a spatial grid correlate statistical data with their spatial location, with high-resolution grids better reflecting the spatial distribution characteristics of the population [11
]. After years of development, some global population distribution grid datasets have been developed, including the Gridded Population of the World (GPW), the Global Rural and Urban Mapping Project (GRUMP), LandScan, and WorldPop [14
], with some countries or regions boasting population distribution grids with resolutions of higher than 100 m. Furthermore, various types of population distribution grid datasets have been used in disaster risk analysis and management, which has provided effective support for emergency government responses and disaster reduction and preparedness [7
In contrast to large-scale population spatialization studies, urban population distributions are significantly related to urban infrastructure such as educational resources and transportation [21
]. Urban surface conditions can be characterized by multi-source geographic data, such as social media data, volunteered geographic information, vehicle trajectory, and mobile phone data [20
], which provide a foundation for obtaining more refined urban population data [23
]. Some studies have suggested that buildings are a more suitable scale for high-resolution simulations of population distribution [25
]. For example, Bakillah constructed a population distribution map at the building scale in Hamburg, Germany [27
] and Yao integrated multi-source spatial data such as points of interest (POIs) and real-time Tencent user densities (RTUD) to construct a refined population distribution at the building scale in Shenzhen, China [21
]. Thus, populations can be estimated at the building scale using multi-source geodata such as POI to determine the relationship between buildings and populations.
Building evaluation represents a fine-scale spatial analytics method of urban flood risk assessment [19
]. Moreover, the spatialization of population data provided by building maps is highly suitable for urban flooding disaster assessments, as they effectively combine population and buildings. With the aim of more accurately assessing the population affected by urban floods, this study employs the population redistribution method based on the grid dataset. Lishui City in Zhejiang Province, China, is used as the study area, where the population affected by urban flooding is estimated using the building-scale population distribution map, with flooding simulated by the LISFLOOD-FP hydrodynamic model.
Considering that the population distribution is highly related to the building distribution during urban flooding, this study assessed the population affected by urban floods using population mapping at the building scale. The building-scale population was mapped by calculating the correlation coefficient between the POI type and WorldPop population grid to establish the relationship between building function and population distribution. The urban area of Lishui City in Zhejiang Province, China, was used as the experimental area, where the affected population was estimated under different rainstorm scenarios. The results showed that the population affected by urban flooding was significantly reduced under different rainstorm scenarios when using the building-scale population instead of WorldPop. This is because the WorldPop grids of the inundated area may cover open areas such as parks and green spaces, which have a negligible population during rainstorms. Moreover, the population neglected by WorldPop could be identified by the building-scale population in some areas, especially when the building-scale population was larger than the population in the affected grids.
The aim of this study was to explore the impact of different dataset spatial scales on the assessment of flood-affected populations. The results indicate that building-scale assessments can improve the relevance and effectiveness of emergency risk management in cities. Although the WorldPop dataset was used in this study, the proposed method could be applied to other datasets to calculate the building-scale population, which would inevitably increase the accuracy of the spatialization results of the population. Furthermore, as the urban population is highly dynamic and the spatialization of demographic data is a top-down analysis method, future work should consider people’s dynamic characteristics when evaluating the population affected by urban flooding.