key: cord-0719328-t4k5csgy
authors: wan, x.; Cheng, C.; zhang, z.
title: Early transmission of COVID-19 has an optimal temperature but late transmission decreases in warm climate
date: 2020-05-18
journal: nan
DOI: 10.1101/2020.05.14.20102459
sha: e28d59d91447cf0f1e890d4628ef67a43799d24f
doc_id: 719328
cord_uid: t4k5csgy

The COVID-19 novel virus, as an emerging highly pathogenic agent, has caused a pandemic. Revealing the influencing factors affecting transmission of COVID-19 is essential to take effective control measures. Several previous studies suggested that the spread of COVID-19 was likely associated with temperature and/or humidity. But, a recent extensive review indicated that conclusions on associations between climate and COVID-19 were elusive with high uncertainty due to caveats in most previous studies, such as limitations in time and space, data quality and confounding factors. In this study, by using a more extensive global dataset covering 578 time series from China, USA, Europe and the rest of the world, we show that climate show distinct impacts on early and late transmission of COVID-19 in the world after excluding the confounding factors. The early transmission ability of COVID-19 peaked around 6.3{degrees}C without or with little human intervention, but the later transmission ability was reduced in high temperature conditions under human intervention, probably driven by increased control efficiency of COVID-19. The transmission ability was positively associated with the founding population size of early reported cases and population size of a location. Our study suggested that with the coming summer seasons, the transmission risk of COVID-19 would increase in the high-latitude or high-altitude regions but decrease in low-latitude or low-altitude regions; human intervention is essential in containing the spread of COVID-19 around the world.

the lower transmission ability under 6.3°C was likely caused by human behaviors. In cold conditions, people are not as active as in warm conditions, which did not favor the person to person transmission of COVID-19. Besides, in cold conditions, droplets can freeze, which prevent their spreading in the air.

The impacts of climate on control efficiency have never been assessed before. We found air temperature showed a significant and negative association with the control efficiency ( ) in China and Europe (Fig. 2G , Table 1 ). High precipitation showed a significant and positive association with the control efficiency ( ) in China and the world ( Table 1 ). These results indicated that cold and wet climate decreased the control efficiency (Note: smaller indicates the better control efficiency) on COVID-19.

This is likely because cold and wet conditions did not favor human movement outside, thus the lockdown and social distancing measures may have worked better in warm and dry conditions rather than in cold and dry conditions. Another possible explanation could be that in cold conditions, people and facilities for disease control or prevention may not be easily mobilized. The efficiency of detection, disinfection or sanitation may be low in cold condition. Therefore, although the COVID-19 virus had an optimal temperature around 6.3°C, the poor control efficiency in cold conditions resulted in the general negative association of temperature with the average daily increase rate.

Our results suggest that climate may affect the transmission of COVID-19 directly is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.14.20102459 doi: medRxiv preprint be arrested in cold condition; thus extreme conditions would make it hard for the virus to survive or spread in cough droplets of infected patients or on the surfaces of contaminated goods. Besides, the control efficiency would be high in warm conditions but low in cold conditions, which might contribute to the observed association between climate and transmission ability.

The effects of human factors on the spread of COVID-19 and control efficiency have been widely modelled in previous studies (e.g. (Gilbert et al., 2020; Hellewell et al., 2020; Tian et al., 2020), but evidence using empirical data is still limited. In this study, we found the founding population size of early reported COVID-19 patients showed a consistent, significant and positive association with the maximum daily increase rate ( ) in China, USA, Europe and the world (Table 1) . It had a positive association with the control efficiency ( ) in China and the world (but a negative association in Europe).

These results suggested that human migration increased the transmission severity of COVID-19. A higher number of early reported COVID-19 cases stimulated the increase of control efforts in Europe, but not in China and the world. The population size of a location showed a consistent, significant and positive association with the average daily increase rate ( ) and the control efficiency ( ) of COVID-19 in the world (Table 1) The Allee effect suggested that founding population size was essential for the successful establishment of alien species (Liebhold & Bascompte, 2003) . It has been widely . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.14.20102459 doi: medRxiv preprint supported in studies of biological invasions (Courchamp, Clutton-Brock, & Grenfell, 1999; Stephens & Sutherland, 1999) , as well as by our observations.

Our study suggested that both human and climate factors determined the spread of COVID-19 by altering the transmission ability of COVID-19 and its control efficiency. Temperature showed distinct effects on the early (without human intervention) and later stage (under human intervention) transmission of COVID-19. We plotted the 6.3°C isocline of COVID-19 in January, April, July and October in Fig. 3E . It is notable that countries or regions with heavy infection rates of COVID-19 are mostly located within the isoclines of 6.3°C (the optimal temperature for COVID-19) from October to January. From winter, spring to summer, the 6.3°C isocline moves from subtropical to temperate, arctic zone; while from summer, autumn to winter, it back moves from arctic to temperate, subtropical zone (Fig. 3E) . The seasonal movement of the 6.3°C isocline of COVID-19 along latitude or altitude increases the transmission risk of COVID-19 in different climate zones. Phylogenetic analysis suggested that the novel coronavirus was a close relation with the . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 18, 2020. 

We obtained data of cumulative cases of COVID-19 in cities and prefectures in China from 1 January to 11 March from daily reports or announcements by each provincial or prefectural health commission (making up 99.23%), the World Health Organization (making up 0.61%), and news from official media such as the CCTV news channel (making up 0.10%), and announcements by local governments (0.06%). Data consisted . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 18, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.14.20102459 doi: medRxiv preprint the USA, Europe, and the rest of the world, which represent the incidence of COVID-19 of three large epicenters and the world (Fig. 1 , Table 1 , SI Appendix, Table S1 -3).

Similarly, cumulative case data was not used for countries or regions when the daily increase rate was smaller than 1%. Cumulative cases were normalized by (average value-minimum value)/range of the value for easily demonstrating the growth patterns of different locations in Fig. 1 , whereas the original data of cumulative cases was used for modeling analysis. It is notable that our data had various spatial resolution from prefecture to state or countries. However, spatial resolution was relatively comparable within China, USA and Europe.

The human population size ( ) of a city or prefecture was obtained from China Human population size was log transformed (with base = e) to make the data normally . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review)

The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.14.20102459 doi: medRxiv preprint 

We used a logistic model to estimate the transmission parameters of COVID-19 by . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.14.20102459 doi: medRxiv preprint +1 = (1− ) was the number of cumulative cases at day , was the maximum cumulative cases of COVID-19 patients. The daily increase rate ( ) of the number of cumulative cases of patients was defined as follow:

Thus, the daily increase rate should be negatively associated with the number of cumulative cases of patients under human intervention:

Here, , are constants, and all > 0. a represents the maximum daily increase rate ( ) without human intervention, represents the control efficiency under human intervention.

Because the mean incubation period of COVID-19 patients was estimated to be 5. April), Europe (n = 52, 25 January-4 th April) and the rest of world (n = 142, 20 January-. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.14.20102459 doi: medRxiv preprint

The process of a virus invading a new place is likely similar to the biological invasion of alien species. The founding population is very essential for the successful invasion as defined by the Allee effect (Allee et al., 1959) . Besides, the population size and climate factors may also play a significant role in affecting the spread of COVID-19 both directly and indirectly. Thus, we assumed that the transmission parameters ( , , GAMs were used to model the effects of the founding population size ( ), human population size ( ), and climate factors (air temperature and precipitation ) on the average daily increase rate ( ) , maximum daily increase rate ( ), and control efficiency ( ) in the th location by following (Wood, 2011). A Gaussian GAMs was firstly fitted by using a linear regression formula:

Here, Y represents the three transmission parameters ( , , ) separately. To examine the potential nonlinear effect of climate factors, we fitted the data using the . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.14.20102459 doi: medRxiv preprint following model:

Here, ( ) , ( ) and ( , ) were 2D smooth function (with k value, dimension of the basis = 4) for removing the effects of spatial autocorrelation. was uncorrelated random errors of zero mean and finite variance. To obtained model robustness as to the effects of human and climate factors on the transmission of COVID-19, we conducted the modeling analysis by using the cumulative cases > 0, 10, 20 and 30 infected cases of COVID-19 (SI Appendix, Table S2 , S3). Only significant effects of a factor on a transmission parameter detected in two modeling analyses were selected for making conclusions and discussions (Table 1 , SI Appendix, Table S1 ).

Using equation (2), (3), we also analyzed the association of the number of cumulative cases with human and climate factors (SI Appendix, Table S4 ).

Pearson's correlation analysis was introduced to detect significant correlations 128 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.14.20102459 doi: medRxiv preprint CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.14.20102459 doi: medRxiv preprint CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 18, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 18, 2020. . factors (temperature, , and precipitation, ) , and spatial autocorrelation 208 based on analyses using Equation 2 (linear model). + denotes the significant effects of spatial autocorrelation or temperature (p < 0.05), NS 209 denotes non-significant effects. * p < 0.05, ** p < 0.01, *** p < 0.001. To obtain robust results, the linear model for each country or region was 210 repeated four times with the cumulative cases >0, 10, 20, and 30, respectively (SI Appendix, Table S2 ). Models presented here were based on the 211 observation that human or climate variables should have significant association with the average daily increase rate ( ), the maximum daily 212 increase rate ( ) or the control efficiency ( ) in at least two repeated models from SI Appendix, 

Correlation between the transmission severity and human/climate factors Table S1 to S4 Figure S1 to S7 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.14.20102459 doi: medRxiv preprint

Using Pearson's correlation analysis on data from 1 January to 11 March in China, we found the average daily increase rate was negative correlated with temperature (r = -0.11, p < 0.05), the maximum daily increase rate was positive correlated with founding population size during the initial 7 days after the first patient (r = 0.28, p < 0.001), the control efficiency was positive correlated with founding population size during the initial 7 days after the first patient (r = 0.21, p < 0.001), human population (r = 0.19, p < 0.001), and precipitation (r = 0.24, p < 0.001) (Fig. S2 ).

Using Pearson's correlation analysis on data from 20 January to 4 April in the USA, we found the maximum daily increase rate was positive correlated with founding population size during the initial 7 days after the first patient (r = 0.63, p < 0.001), the control efficiency was positive correlated with human population (r = 0.41, p < 0.01), and precipitation (r = 0.40, p < 0.01) (Fig. S3 ).

Using Pearson's correlation analysis on data from 20 January to 4 April in Europe, we found the maximum daily increase rate was positive correlated with founding population size during the initial 7 days after the first patient (r = 0.49, p < 0.001), the control efficiency was positive correlated with founding population size during the initial 7 days after the first patient (r = -0.33, p < 0.05) (Fig. S4 ).

Using Pearson's correlation analysis on data from 1 January to 4 April for the world data, we found the average daily increase rate was positive correlated with precipitation (r = 0.11, p < 0.01), maximum daily increase rate was positive correlated with founding population size during the initial 7 days after the first patient (r = 0.30, p < 0.001), but . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.14.20102459 doi: medRxiv preprint negative correlated with human population (r = -0.14, p < 0.01), temperature (r = -0.22, p < 0.001), and precipitation (r = -0.12, p < 0. 01), the control efficiency was positive correlated with founding population size during the initial 7 days after the first patient (r = 0.15, p < 0.001), temperature (r = 0.10, p < 0.05), and precipitation (r = 0.19, p < 0.001) (Fig. S5) .

. CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.14.20102459 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.14.20102459 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.14.20102459 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.14.20102459 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.14.20102459 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.14.20102459 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.14.20102459 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity.

The copyright holder for this preprint this version posted May 18, 2020. . https://doi.org/10.1101/2020.05.14.20102459 doi: medRxiv preprint Table S1 Significant associations of average daily increase rate ( ) of cumulative cases, maximum daily increase rate ( ), control efficiency ( ) of COVID-19 with founding population size during the initial 7 days after the first patient ( ), human population, climate factors (temperature and precipitation), and spatial autocorrelation. Bold values indicated the coefficients represent the significant effects (p < 0.05). + denotes the significant effects of spatial autocorrelation or temperature (p < 0.05), ns denotes non-significant effects. NA denotes not available (* p < 0.05, ** p < 0.01, *** p < 0.001). The significant association was extracted from Table S3 . Models presented here were based on the observation that human or climate variables should have significant association with the average daily increase rate ( ), the maximum daily increase rate ( ) or the control efficiency ( ) in at least two repeated models from Table S4 Associations of the number of cumulative cases (N) of COVID-19 with founding population size during the initial 7 days after the first patient ( ), human population, climate factors (temperature and precipitation), and spatial autocorrelation based on analyses using Equation 2 (linear model) and 3 (non-linear model) for locations with more than 0/10/20/30 cases. Bold values indicated the coefficients represent the significant effects (p < 0.05). + denotes the significant effects of spatial autocorrelation or temperature (p < 0.05), ns denotes non-significant effects. NA denotes not available (* p < 0.05, ** p < 0.01, *** p < 0.001). 

Principles of Animal Ecology