A Spatio-Temporal Analysis of the Environmental Correlates of COVID-19 Incidence in Spain

Abstract

The novel SARS-CoV2 has disrupted health systems and the economy, and public health interventions to slow its spread have been costly. How and when to ease restrictions to movement hinges in part on whether SARS-CoV2 will display seasonality due to variations in temperature, humidity, and hours of sunshine. Here, we address this question by means of a spatio-temporal analysis in Spain of the incidence of COVID-19, the disease caused by the virus. Use of spatial Seemingly Unrelated Regressions (SUR) allows us to model the incidence of reported cases of the disease per 100,000 population as an interregional contagion process, in addition to a function of temperature, humidity, and sunshine. In the analysis we also control for GDP per capita, percentage of older adults in the population, population density, and presence of mass transit systems. The results support the hypothesis that incidence of the disease is lower at higher temperatures and higher levels of humidity. Sunshine, in contrast, displays a positive association with incidence of the disease. Our control variables also yield interesting insights. Higher incidence is associated with higher GDP per capita and presence of mass transit systems in the province; in contrast, population density and percentage of older adults display negative associations with incidence of COVID-19.

Figure 1

Mean weekly incidence of COVID‐19 by province, in reported cases by 100,000 people.

Figure 2

Distribution of mean temperatures and humidities in the Autonomous Communities in Spain between March 12, 2020 and April 11, 2020. The Autonomous Communities have been sorted by latitude, with communities to the left being the northermost, and to the right the southernmost.

Figure 3

Spatial distribution of control variables by province.

Figure 4

Distribution of daily correlations of the independent variables with daily incidence of COVID‐19 (all variables have been log‐transformed).

Figure 5

Goodness of fit of the SUR systems: by date and pooled.

Figure 6

Temporal evolution of the spatial autocorrelation coefficient (rho) and the intercept of the model; dots are the point estimates and vertical lines are 95% confidence intervals. In yellow is the period after the declaration of the state of emergency, and in orange is the period when only essential activities were allowed.

Figure 7

Temporal evolution of coefficient for the control variables; dots are the point estimates and vertical lines are 95% confidence intervals. In yellow is the period after the declaration of the state of emergency, and in orange is the period when only essential activities were allowed.

Figure 8

Temporal evolution of coefficient for the environmental variables; dots are the point estimates and vertical lines are 95% confidence intervals. In yellow is the period after the declaration of the state of emergency, and in orange is the period when only essential activities were allowed.

Figure 9

Temporal evolution of direct, indirect, and total effects by date.