Multi-source sensor based urban habitat and resident health sensing: A case study of Wuhan, China

Abstract

The COVID-19 pandemic undoubtedly has a great impact on the world economy, especially the urban economy. It is urgent to study the environmental pathogenic factors and transmission route of it. We want to discuss the relationship between the urban living environment and the number of confirmed cases at the community scale, and examine the driving forces of community infection (e.g., environment, ecology, convenience, livability, and population density). Besides, we hope that our research will help make our cities more inclusive, safe, resilient, and sustainable. 650 communities with confirmed COVID-19 cases in Wuhan were selected as the research objects. We utilize deep learning semantic segmentation technology to calculate the Visible Green Index (VGI) and Sky View Factor (SVF) of street view and use Partial Least Squares Structural Equation Modeling (PLS-SEM) to study the driving forces of pandemic situation. Temperature and humidity information recorded by sensors was also used for urban sensing. We find that the more SVF has a certain inhibitory effect on the virus transmission, but contrary to our intuitive perception, higher VGI has a certain promotion effect. Also, the structural equation model constructed in this paper can explain the variance of 28.9% of the number of confirmed cases, and results (path coef.) demonstrate that residential density of community (0.517) is a major influencing factor for pandemic cases, whereas convenience of community living (0.234) strongly influence it. Communities with good suitability of community human settlement (e.g., construction time, price) are safer in the face of pandemic events. Does the influence of SVF and VGI on the results of the pandemic situation mean that sunlight can effectively block the spread of the virus? This spatial heterogeneity in different communities is helpful for us to explore the environmental transmission route of COVID-19.

Keywords: COVID-19; GIS; PLS-SEM; Sensor; Social sensing; Street view.

Fig. 1

Study area profile and different community streetscape.

Fig. 2

Method introduction and chapter connection.

Fig. 3

Atrous Spatial Pyramid Pooling modular.

Fig. 4

ResNet101 module structure.

Fig. 5

The segmentation effect of Deeplabv3.

Fig. 6

Comparison of different semantic segmentation algorithms.

Fig. 7

Scene classification based on deep learning methods: street view image (A) and after the segmentation (B).

Fig. 8

Verification of satellite ground fusion (According to the house price>30,000, 20,000

Fig. 9

Variable correlations and interactions (**Representative passed the test of p = 0.01, *passed the test of p = 0.05).

Fig. 10

Two-by-two fitting relationship between variables (after reject outlier).

Fig. 11

Structural equation model structure, path coefficients and factor loadings.

Fig. 12

Temperature and humidity and the development of pandemic time series (Wuhan city adjusted the diagnostic criteria on February 12, 2020, and we compressed the data point on this day).