Bookmark and Share

Alex Bäcker's Wiki / High temperatures correlate with slower Covid-19 infection growth rates
  • If you are citizen of an European Union member nation, you may not use this service unless you are at least 16 years old.

  • Finally, you can manage your Google Docs, uploads, and email attachments (plus Dropbox and Slack files) in one convenient place. Claim a free account, and in less than 2 minutes, Dokkio (from the makers of PBworks) can automatically organize your content for you.


High temperatures correlate with slower Covid-19 infection growth rates

Page history last edited by Alex Backer, Ph.D. 9 months, 3 weeks ago Saved with comment

A more up to date version of this research can be found at .


High temperatures correlate with slower Covid-19 infection growth rates

Alex Bäcker, Ph.D.

There is a lot of controversy about the effect of temperature on Coronavirus transmission and Covid-19 infection rates. This is a very important question, for a number of reasons. First, it will determine whether rising temperatures with spring and summer will change the course of the epidemic. Second, it will determine the course of the epidemic in different locations, and thus the recommendations on where people should weather this out, or if it matters. Third, it can have implications for the temperature at which to keep hospitals, homes and buildings.


The largest initial outbreaks of Covid-19 have all occurred in a narrow latitude band characterized by consistently similar weather patterns (5-11 degrees C and 47-79% humidity)  (“Temperature and latitude analysis to predict potential spread and seasonality for COVID-19”, by Mohammad M. Sajadi, MD et al.):

Qasim Bukhari & Yusuf Jameel, postdocs at MIT, argued that 90% of the 2019-nCoV transmissions have so far occurred within a certain range of temperature (3 to 17C) and absolute humidity (4 to 9g/m3 ) and the total number of cases in countries with mean Jan-Feb-March temperature >18C and and absolute humidity >9 g/m3 is less than 6%. But their analysis compared numbers from different stages of the outbreak, making it possible that the differences are simply historical in nature, with outbreaks at latitudes farther from the original epicenter in Wuhan are simply behind. 


Wang et al found that the higher the temperature, the lower the number of cases, but that again could be explained historically:


They then plotted the daily effective viral reproductive number, R, as a function of temperature 

 for each of all 100 Chinese cities with more than 40 cases from January 21 to 23, 2020:


Yet that data is confined to China, and dates back to January. Since then, outbreaks have occurred in Brazil, Australia and a number of other countries in the southern hemisphere with temperatures quite different from those in the narrow band where the original outbreaks occurred, prompting some to suggest that there are no signs that warmer weather regions being different.




To shed light on this question, I looked at the number of cases reported in a number of countries by March 20th as a function of the average temperature at the most affected city in that country the week of March 20th:

The results showed every single location with over 2,000 cases to date occurred at or below an average temperature of ten degrees Celsius.


To investigate whether that could be simply a historical accident related to the location of the initial outbreak, I looked at the growth in number of cases in the week following the first day in which 100 or more cases were reported at each of those locations. The results show a significantly faster growth curve for countries with a lower average temperature:




This could no longer be explained by a historical accident, since the curves are aligned on the day each particular geography passed 100 cases.


To quantify the effect of temperature on growth rate, I plotted the number of cases on the seventh day after hitting 100 cases at each location versus the average temperature of that place around the time of the epidemic in each location:

Cases on the seventh day after hitting 100 cases has a statistically significant Pearson's correlation coefficient of -0.53 with average temperature (p<0.01, paired T-test, two samples, unequal variance).


I then repeated the analysis a week later, and found the trend got even clearer:


Indeed, *every single* locale with a temperature above 10°C (solid lines with triangles) had a slower growth in the number of infections detected than *every single* locale with a temperature 10°C or lower (dashed lines with stars). That is statistically significant at p=0.0001 (1/2^13).


Furthermore, the three locales with the highest growth in infection detections (Hubei, New York and Spain) are the three locales with the lowest temperatures among those analyzed. That is statistically significant at p<0.01 ((3/14)^3).


The correlation also got stronger four days later (r=-0.65, p<0.01):




The results show that, for the dataset analyzed, places with temperatures at or below ten degrees Celsius had a significantly higher growth rate in the number of cases reported.


To test whether the effect was due to sunlight or temperature, we obtained irradiance and temperature data from Solcast, and carried out the same analysis on both. The Solcast temperature data confirmed the effect described above. The Solcast irradiance data proved there was an even larger coefficient of correlation between COVID 19 case growth and irradiance (GHI in W/m2 (irradiance in the sunlight)) than between COVID 19 case growth and temperature (CC=-0.58, p<0.01):





Whilst Covid-19 infections are clearly able to grow at a wide range of temperatures, and steps to slow down their growth, including social distancing, should be taken seriously everywhere, it appears clear that higher temperatures (15 to 30 degrees Celsius) seem to slow down the growth rate. Thus, keeping the at risk population at such higher temperatures might reduce the probability of individual infection, and decrease the overall number of infections. An obvious next step would be to test infection rates as a function of indoor ambient temperatures, in the hope that higher temperatures and/or humidity can lead to a reduction in the viral infection growth rate.


Within 12 days of passing the very same 100-case threshold, the difference between the fastest-growing and slowest-growing locales was 83X. Progress is being made fast, both in provisioning hospital beds, in learning about treatment effectiveness, in splitting ventilators, and more. Buying time is key.


Many different mechanisms could be behind the correlation observed. There is evidence that virus particles last longer on surfaces at cold temperatures. There is evidence that aerosol transmission of other viruses is blocked by high temperatures (30°C) (8). A study cited by the WHO showed heat at 56°C kills the SARS coronavirus at around 10000 units per 15 min (quick reduction). So it's not crazy to think that temperature halfway from 10°C to there will reduce viral load. Immune systems could be more compromised at lower temperatures. People tend to be more gregarious at higher temperatures. Sunlight exposure stimulates vitamin D production.


It would be interesting to follow this study up with an analysis of the impact of population density and to infer underlying infection rates given test rates per inhabitant and test negative result rates.


Underlying infection rates vs. detected infection rates


A problem with infection growth curves is that they are not actually infection growth curves, but infection *detection* growth curves. Since testing is uneven in time and across geographies, what we really want is to estimate the underlying hidden variable, infection rate. This is important for a number of reasons. First, in order to assess at what point saturation and a flattening of the infection and death growth curves is expected. Second, because comparing death rates to decide on best treatments cannot be done without knowing if different countries, states or counties are dealing with different subsets of the overall population.


I posit the estimation of this hidden variable can be done in two ways.


If one has control over the testing, by testing random samples of the population. It is my recommendation to testing authorities and the WHO that those start getting published at the same time as the non-random samples numbers.


If one doesn't have control over the testing, I posit one can estimate it by normalizing infection detection numbers making an assumption about a variable that is believed to be constant. For example, death rates may not be great for this use case because they could vary greatly based on the quality of healthcare, but an earlier variable less subject to manipulation, such as % of detected infections (let's call them diagnoses) that require oxygen for a given age (failing to normalize by age would expose you to variability based on differences in average age across populations) could work.


Est. infection rate in geo G = observed infection rate in G x (% of diagnoses of age band A that require oxygen in G / % of diagnoses of age band A that require oxygen in testing of a random sample of the general population)


The larger the % of diagnoses that require oxygen, the more narrowly the testing is likely happening on severe cases only, and the more testing is likely to be underdiagnosing.


I have not seen any metric like this applied in any of the plethora of websites tracking infection growth, and it seems to me like it would yield a better prediction of underlying infection rates, and thus of when the curves are likely to flatten out.


Applying such a metric would constitute a logical extension of this study.




Average temperatures for the corresponding week were obtained from . Covid-19 infection numbers were obtained from Carlos Brody’s data repository, which itself obtained data from the Johns Hopkins Covid-19 database


Conflicts of interest


No conflicts to report.




The author is deeply indebted to Pablo Bäcker Peral for his help gathering data. 


About the Author


The author holds a Ph.D. in biology from the California Institute of Technology, is a member of Caltech's Board of Information Science and Technology and is co-founder of QLess, Inc., the inventor and developer of mobile queueing technology that can be used to achieve social distancing while remaining in business. The opinions expressed herein are solely the author’s.




[2]“Temperature and latitude analysis to predict potential spread and seasonality for COVID-19”, by Mohammad M. Sajadi, MD et al.

[3]Qasim Bukhari & Yusuf Jameel, 'Will coronavirus pandemic diminish by summer?'

[4] Wang, Jingyuan and Tang, Ke and Feng, Kai and Lv, Weifeng, High Temperature and High Humidity Reduce the Transmission of COVID-19 (March 9, 2020). Available at SSRN: or

[5] Carlos Brody:



[8]  JOURNAL OF VIROLOGY, June 2008, p. 5650–5652 Vol. 82, No. 11. 0022-538X/08/$08.000 doi:10.1128/JVI.00325-08. High Temperature (30°C) Blocks Aerosol but Not Contact Transmission of Influenza Virus. Anice C. Lowen,1 John Steel,1 Samira Mubareka,1 and Peter Palese1,2*.

[9] Solcast, Solar irradiance data, (2020).


Nota Bene

The author is not an epidemiologist, but simply a Ph.D. in biology with 25 years of experience in data analysis who thinks that a global pandemic crisis that has already killed over 24,000 and is projected to kill may more cannot wait for the publication of peer-reviewed studies by epidemiologists alone. Rather, extraordinary times demand extraordinary solutions, and the contribution of all capable of proposing potential solutions that could alleviate the number of deaths ahead. 


First published online March 23rd, 2020. Updated March 28th, 2020.




Comments (0)

You don't have permission to comment on this page.