In recent months, the COVID-19 pandemic has surged in Europe, the United States, and some other areas. In many regions, this surge began coincident with a transition towards colder weather in late fall and early winter. This post will examine the possible role that these changing weather conditions may be playing in the spread of COVID-19.
The transition to colder weather has corresponded with an increase in the spread of COVID-19. Considered across many regions, we find that the average reproduction number (R) gradually increased from ~1.0 to ~1.2 in coincidence with daily high temperatures falling from >27 °C to ~10-15 °C (>80 °F to 50-60 °F). Other factors, including decreases in sunlight and/or UV exposure, may have slightly increased this further.
This is consistent with an epidemic that moved, on average, from an approximately stable rate of spread to a rate of spread that doubled every 20 days. As discussed below, it is plausible that weather changes, such as decreasing temperature or sunlight, and the associated human responses to these changes have directly and/or indirectly contributed to the recent resurgence of COVID-19.
Preliminary data suggests that the average rate of spread may stop increasing and actually decline somewhat when daily high temperatures fall from ~10 °C towards 0 °C (~50 °F to 30 °F); however, it is unclear if this finding is robust. It may be an artifact of the many changes in pandemic countermeasures that occurred in the late fall.
The COVID-19 weather-related response suggested here is relatively modest, and will often have less impact than other changes in human behavior, such as compliance with mask-wearing directives. However, compared to warm weather, these findings do suggest that additional countermeasures will generally be necessary to control the spread on COVID-19 during cold weather.
Introduction
Back in April, we looked at how COVID-19 had been responding to weather conditions. In that discussion, we concluded that COVID-19 was at most only modestly inhibited by warm weather, and was likely to continue spreading through the summer. The summer wave of infections in the United States and the continued spread in warm regions like India has made clear that this prediction was accurate. Unlike diseases like influenza, which largely disappear during summer months, the virus that causes COVID-19 was robust enough to continue spreading during warm months.
However, even if warm weather did not, by itself, stop the spread of COVID-19, it may still have affected it somewhat. With a return of colder weather, environmental conditions and associated human behaviors may have allowed COVID-19 to spread more easily. This could occur either because the virus itself is sensitive to environmental conditions (e.g. if temperature and/or humidity affect the virus’s longevity in the environment), or because humans behave differently in cold weather (e.g. spending more time clustered together indoors), or both. For convenience, environmental effects on the transmissibility of the virus itself can be termed “direct effects”, and effects on human behavior that impact the disease’s spreads can be termed “indirect effects”.
Recently, a systematic review of 17 research papers concluded that temperature and humidity were likely to have an effect on COVID-19 disease transmission, though the evidence remained of low quality, and the effect was likely of less importance than other variations in human behavior (e.g. group gatherings, mask-wearing). The evidence supported a view that COVID-19 spreads somewhat less rapidly in warm, wet environments.
Initially, COVID-19 surged worldwide in March and April, corresponding to spring in the Northern Hemisphere. During the warm months that followed, many nations and regions (though not all) managed to greatly reduce their local COVID-19 epidemic when compared to the spring, but most have subsequently seen a resurgence in late fall.
We will examine the timing of the recent resurgence and make a case that weather conditions are probably a contributing factor. However, as with any disease outbreak, weather is certainly not the only factor. Human behavior, and adherence to COVID-19 counter-measures, have also played a large role.
The focus of this discussion will be the period of weather from July 1st to December 1st, with an assumption that COVID-19 is transmitted approximately 1 week before it is diagnosed, so that new cases are assumed to be most influenced by the weather that occurred one week prior to being reported.
Background: The new surge
Section Summary: We will start by looking briefly at how COVID-19 has spread in the territories of the USA and European Union and what that might suggest about the dynamics.
Before we begin, it is worth clearly characterizing the evidence of a new surge. While it is likely that many people are already familiar with this, globally new COVID-19 infections have recently been hitting record highs, as have COVID-19 deaths. This effect is not present everywhere, but has been particularly prominent in the United States and Europe.
A comparison of the United States and the European Union show that both first saw an outbreak in the spring. Subsequently, the EU managed to reduce the rate of new reported infections by ~85% during the summer months. By contrast, the USA saw a second wave in the middle of the summer. Finally, these and other regions are now experiencing a very large resurgence in cases since October and November.
A few weeks after the surge in cases, both the United States and European Union have also being experiencing a surge in COVID-19 deaths.
Notably, though cases are ~6 times higher now than in the spring, the overall death rate has been similar to the spring. Much of this difference may be do to the underreporting of cases in the spring, largely due to inadequate testing early in the pandemic. However, some of the difference is also likely due to improvements of treatment protocols and other factors that have helped to reduced the mortality rate.
One curious feature of this comparison is that the surge occurs ~2 weeks earlier in the EU than in the USA. While there might be several reasons for this, one possible explanation is that the EU is generally cooler than the United States, and transitioned into cooler temperatures ~3 weeks earlier than the US this year.
However, looking at the EU and USA as monolithic units is inherently problematic since there is wide variation in both environmental conditions and pandemic responses across these territories.
Briefly, we will offer a more local example to note the relative timing. Switzerland, began a large surge in COVID-19 cases shortly after a cold wave in Europe reduced the average daily high temperature in Switzerland from ~20 °C to ~10 °C (70 °F to 50 °F). Based on the findings below, it appears plausible that the cold snap was a significant trigger for the surge of cases. In the Swiss case, the surge continued until additional counter-measures (e.g. new partial lockdowns) were re-introduced in some of the worse affected areas in late October / early November.
Reproduction Numbers over Time
Section Summary: The rate at which COVID-19 spreads, as measured by its “reproduction number” has changed over time. Averaged across many countries and US states, the reproduction number was slightly above 1 during March to June, moved to ~1.0 during July and August, and has then increased again slightly above 1 in September to November.
An important way to characterize the spread of COVID-19 is through its reproduction number (R). This measures the number of new people that each infected person will be expected to subsequently infect. A value of R higher than 1 indicates that the local epidemic is accelerating, lower than 1 means the epidemic is declining, and values approximately equal to 1 indicate a constant rate of new infections.
In the absence of any counter measures, it was initially estimated that COVID-19 had a basic reproduction number of ~2.5, but subsequent studies suggests it could be 4.5 or higher. In that case, on average, each infected person would spread the disease to 4.5 additional people. This would correspond to the number of reported new cases doubling every 2-3 days, as was reported in some of the territories that were most impacted in the beginning of the pandemic, before COVID-19 awareness became widespread.
However, after the world became aware of COVID-19, most countries and individuals began taking steps to reduce exposure. Such counter measures have the effect of reducing the effective reproduction number. Since widespread awareness developed in April, the global number of new cases has generally grown at a rate ranging from R = 1.0 to 1.2, essentially varying from an approximately constant rate of new infections to a rate of new infections that doubles every ~20 days. This is obviously much slower than would be expected from uncontrolled spread, and reflects the widespread efforts to reduce rates of transmission. However, the global average also conceals wide variation between countries with very rapid spread and countries with nearly complete suppression of the virus.
In the following plot, measures of the reproduction numbers over time for every country and US states have been aggregated together. Due to the volume of measurements, a density plot is used to better characterize the result. For individual locations and times, there is wide variation from values well below 1 to values substantially higher than 1. To emphasize the average behavior, the median of this collection of measurements over time is shown via a red curve.
Aggregating data on the rate of spread across a wide range of locations finds a relatively narrow range of variation in the average over time. Since May, the median R value has generally been between 1.0 and 1.1. This suggests that control over the pandemic, in a global sense, might be described as more or less stagnant with a slow level of increase. Notably, a transition from ~1.0 to ~1.1 occurred in October and November as pandemic spread picked up again.
Reproduction Rate vs. Temperature
Section Summary: Comparing effective rates of COVID-19 spread vs. environmental temperature shows that R values increased from 1.0 to 1.2 coincident with the daily high temperature falling from ~27 °C to ~15 °C (80 °F to 60 °F).
While it is well-known that COVID-19 rates of spread have increased in October and November, it is useful to consider whether there is any correlation between changes in the rate of spread and environmental conditions.
We observe that for locations with daily high temperatures above ~27 °C (80 °F), the median rate of spread has been around R = 1.0. In other words, though the pandemic continued to spread during warm weather, on average it was not accelerating. This is consistent with the observation that many locations saw relatively less severe pandemics during their local summer, manageable with only moderate control measures.
By contrast, the median R value has increased to ~1.2 when daily high temperatures are around 10 to 15 °C (50 to 60 °F). At this R value, the rate of new cases would be expected to double every ~20 days. This is consistent with cooling weather in the fall contributing to a new pandemic surge.
Intriguingly, the R value seems to fall back below 1.1 at the coldest sampled temperatures; however, it is unclear if this effect is reliable due to sparser data and the fact that such temperatures often only occurred after countries had already introduced new control measures in response to local surges.
This analysis is fundamentally correlational in nature. It cannot prove that cooling weather caused COVID-19 infections to surge, though the data is consistent with that interpretation.
If the weather is affecting COVID-19 transmission, we think that indirect effects are likely to be a major factor. While temperature might directly affect the longevity or transmissibility of the virus, it is well established that most infections occur in indoor spaces where outdoor temperatures would have no direct impact. However, if cooling weather drives individuals to spend more time indoors, then that may increase the number of opportunities for virus transmission. Similarly, populations that rely on open windows for cooling during summer may have less indoor ventilation during cool weather. We consider it likely that a majority of any weather effect on COVID transmission is primarily the result of indirect effects on human behavior. However, we will note that laboratory studies have indicated that the coronavirus remains viable on surfaces longer at lower temperatures.
Lastly, we would note that similar shifts in the R value are observed if either daily average temperature or daily low temperature are used as the explanatory variable. We used daily high temperature above as we believe it makes the association easier to understand, but the analysis is unable to distinguish what aspect of daily temperature is most significant.
Reproduction Number vs. Humidity
Section summary: An examination of humidity is also considered as an alternative weather variable. We conclude that relative humidity has no apparent impact on the rate of spread. Absolute humidity does appear to have an impact; however, absolute humidity and temperature are highly correlated. After controlling for the effect of temperature, there does not appear to be any additional influence of absolute humidity on the rate of spread.
It is well-known that some viruses are sensitive to changes in humidity. With that in mind, it is useful to also consider whether changes in humidity might have affected the spread of COVID-19.
To begin, we find very little evidence that environmental relative humidity has affected the rate of COVID-19 spread. Across the full range of observed humidity values, R is close to 1.0 and only very slightly elevated for relative humidity greater than 50%. Given that the median R value over the last few months must also be somewhat elevated, we regard this pattern as consistent with no effect.
Relative humidity measures the amount of water vapor in the air as a fraction of the maximum possible water vapor content, which varies strongly with temperature. By contrast, absolute humidity measures the total water vapor directly.
Here we find that COVID-19 spread appears to increase at low values of absolute humidity. However, we must acknowledge that absolute humidity is highly correlated to temperature, so an effect at low absolute humidity could also be a direct result of low temperature.
To test the possibility that the apparent increase in spread at low absolute humidity is simply related to temperature, we adjusted the measured R values to account for the previously described temperature effect. After doing so, there does not appear to be any residual influence attributable to absolute humidity.
This conclusion suggests that either temperature or absolute humidity or a combination could be driving changes in COVID-19’s spread; however, the two variables are too tightly correlated to be empirically distinguishable. However, we favor the interpretation that temperature — and the changes in human behavior that it induces — is more likely to be the causal agent than changes in outdoor absolute humidity.
Reproduction Number vs. Sunlight & Ultraviolet
Section summary: Another possible environment change during cold weather is a reduction in sunlight and ultraviolet radiation. Similar to temperature, we find that R values increased from 1.0 to 1.2 coincident with declines in sunlight or ultraviolet (UV) exposure.
Most of this change is indistinguishable form the change in environmental temperature, since sunlight and temperature change in tandem; however, there may be a small residual effect due to sunlight change even after accounting for temperature changes.
In addition to cooling temperatures, another change that occurs during the fall and winter is a reduction in sunlight.
As one might expect, the changes in COVID-19 rate of spread observed versus temperature are largely mirrored by changes with respect to average daily sunlight. After all, changes in solar radiation are the main reason for the changes in temperature.
As with absolute humidity, we consider how the effect appears to change if we adjusted for the apparent influence of temperature.
As before, most of the change in apparent COVID-19 reproduction rates with respect to sunlight vanishes after adjusting for the implied role of temperature changes. However, the residual effect in the case of sunlight is somewhat larger than with humidity. This suggest there may be a small impact of variations in sunlight that is independent of the variations in temperature.
Similar analysis (not shown) where temperature effects are assessed after controlling for sunlight variations produces a similar residual effect.
It is plausible that sunlight variations have a small added effect not already captured by the consideration of temperature. One plausible interpretation is that reductions in sunlight also affect human behavior, for example in terms of the number of hours that one will spend outdoor vs. indoors. Even on an unusually warm winter day, people may well spend less time outdoors than during similarly warm days with more hours of daylight.
An alternative consideration is that sunlight might have a direct effect, either on durability of the virus or on human physiology. Laboratory findings suggest that exposure to sunlight can inactivate the virus.
We will also mention that changes in ultraviolet radiation (UV) are largely indistinguishable from changes in total sunlight. A comparison of UV flux to R values finds a similar trend, which also partially persists after correcting for temperature.
UV flux is of particular interest given that UV light is known to inactivate many viruses, so periods of low UV may allow the virus to persist longer in outdoor environments and on outdoor surfaces.
UV light also plays a physiological role by promoting the formation of Vitamin D when human skin absorbs UV. Several studies have found that patients with low Vitamin D levels may be more vulnerable to COVID-19 or tend to have more several outcomes. However, Vitamin D seems to be ineffective as a treatment when given to patients who are already hospitalized with severe COVID-19.
Since Vitamin D levels tend to decline during cold weather (due to less exposure to UV), then this could also be a possible route by which environmental conditions affect COVID-19’s rate of spread.
Estimated Environmental Effect over Time
By combining the factors considered above, specifically daily high temperature, sunlight, and UV, it is possible to construct an estimate of the environmental pressures affecting the spread of COVID-19 at different times, which is summarized for different months in the four maps below.
 |  |
 |  |
The combined effect of environmental factors adds about 0.25 to the effective R value, with most of the effect plausibly due to temperature. As shown here, this implies a role for seasonal variation in the transmission of the coronavirus, with an emphasis on locally cold months. However, as noted in the temperature section, the data suggests that transmission may peak around 10-15 °C and decline somewhat at colder temperatures, possibly leading to a greater pressure on mid-latitude countries during the winter rather than high latitude countries. This is particular evident in the November 2020 map above, where some high latitude countries appear less heavily shaded.
As emphasized above, it is entirely plausible that much of the observed weather effect is due to the human responses to colder weather, e.g. more time spent indoors with less ventilation. Though we can’t rule out that weather conditions direct affect the virus.
Also, we would reiterate that though weather appears to play a role in affecting COVID-19 transmission, it is only a modest effect. Other human factors, including efforts undertaken to limit transmission are likely more important. In particular, though the maps above do match up with some of the observed changes in national rates of transmission, there are also many exceptions due, in part, to the local differences in how nations have responded to the pandemic.
Discussion
COVID-19 has traveled in “waves”. The current surge, which began in many places in October and November, has been among the most severe in many Northern Hemisphere countries.
The reasons for the current surge remain unclear. Purely human factors, such as pandemic fatigue, may play a role, but it is useful to consider whether the nearly synchronized surge across many countries is due, in part, to changes in environmental factors.
As discussed above, it appears plausible that COVID-19 may be somewhat influenced by environmental conditions. Though warm weather did not stop COVID-19 from spreading, it appears that the virus may spread somewhat more easily during cold weather. Having a degree of seasonality would make COVID-19 similar to influenza and many other viruses. Though, in the case of COVID-19, the seasonal effects appear to be relatively weak.
Of the factors examined, it appears that declines in temperature and/or declines in sunlight/UV exposure are likely to be the most significant. Averaged across many populations, it appears that the transition to cold weather has been associated with a boost of ~0.2 in the effective reproduction number (R value). Other factors, including decreases in sunlight, may add to this slightly. These boosts in R value, occurring in a population with a preexisting stagnant rate of spread (e.g. R about 1.0), would be enough to jump start a further pandemic wave.
For example, a shift to R = 1.2 would predict a doubling in the number of new cases every ~20 days. Small increments beyond that would allow the even faster growth, such as if environmental factors coincided with other changes in behavior. We note that EU and USA saw the rate of daily new cases grow approximately 4-fold (two doublings) over several weeks in October and November.
| Effective R value | Approximate doubling time for new cases |
| < 1.0 | Never (epidemic declining) |
| 1.0 | Never (stable rate of new infections) |
| 1.1 | ~38 days |
| 1.2 | ~20 days |
| 1.3 | ~14 days |
| 1.4 | ~11 days |
We suspect that the most likely reason for an association between weather changes and increased spread is due to changes in human behavior, especially the tendency to spend more time indoors when weather cools and sunlight decreases. It appears likely that most virus transmission occurs indoors, and any factors that increase time spent indoors in mixed household groups are likely to encourage spread. However, the current presentation is limited in that we have not directly considered mobility data or other information which might directly measure such changes in behavior. In addition, we have not attempted to untangle the effects of changes in pandemic response over time.
While human behavior is expected to play a large role, we cannot rule out the possibility that temperature, sunlight, or absolute humidity may be having direct effects on the transmissibility of the virus. For example, by affecting how long the virus remains viable outside the body. In addition, to the extent that changes in sunlight and other factors affect human physiology (e.g. Vitamin D levels), this could also be affecting viral spread.
At this point in the pandemic, it has become clear that a warm climate does not by itself prevent local transmission of the coronavirus responsible for COVID-19. However, it may be helpful. Notably, Central Africa and Southeast Asia appear less severely affected than most regions. It is plausible that regions that remain warm year-round will have a somewhat easier time controlling transmission. Though even in such countries, vigilance and countermeasures are likely to remain necessary.
In mid-latitude countries, seasonal changes towards cold weather may trigger repeated, robust outbreaks when daily high temperatures fall to 10-15 °C (50-60 °F). This suggests that pandemic countermeasures may need to be consistently increased during periods of colder weather in order to have the same degree of effectiveness as during warm weather.
There is also a suggestion that outbreaks may naturally ease, somewhat, during the peak of winter in climates where daily highs near 0 °C are the norm. If true, some countries might expect to see surges in fall and spring with reduced severity in winter. However, this latter conclusion remains uncertain.
It is worth mentioning that the ~0.2 increase in R value observed here due to weather appears to be less significant than the potential for a 0.4-0.7 increase in R value due to the emergence of a new COVID-19 strain, “the UK variant B117”. This variant, which originated at least as early as September but only gained prominence in December, appears to be substantially more efficient at infecting humans. In the near future, the local prevalence of this new strain may be a dominant cause of regional variations in the rate of COVID-19 spread, with weather playing a reduced significance.
Lastly, we want to reemphasize that weather is, at best, only a modest factor affecting the transmission of COVID-19. Human behavior is the most important consideration in determining the rate of spread of COVID-19, and the rate of spread can be reduced by adherence to control measures such as limiting gatherings and wearing masks.
Methods
The analysis presented here is based on confirmed cases over time as reported by Johns Hopkins University: data portal & time series files. Values up till December 1st were considered here. For some locations, the JHU data includes very large single day spikes when backlogs of re-evaluated old cases were added to national totals. To avoid distortions associated with such events, days with reported counts much larger than average when compared to a surrounding 14-day period were excluded. Such exclusions were rare.
Temperature, humidity, solar radiation and UV flux data are compiled from the ERA5 reanalysis dataset. Weather variables are masked to political boundaries using the shapefiles provided by Natural Earth. Representative values for each political territory were constructed using population-weighted averages to better reflect conditions where people actually reside. The Gridded Population of the World Dataset v. 4 data for 2015 is used for this purpose.
In most of the analysis of cases, an offset of seven days was used when comparing case counts to weather conditions. This is motivated by the typical delay of ~5 days between exposure and the onset of symptoms, and assuming a further ~2 days for diagnosis. Both weather and case numbers were smoothed using a 7-day average to remove variability due to weekday/weekend effects (very common in the reported case data for many countries). R values where estimated by examining the change in the number of new cases on a weekly basis and assuming an effective serial interval of 5.2 days. Territories reporting fewer than 500 new cases per week were omitted.
Scatter plots of weather data vs. estimated R value were constructed by aggregating over all countries and US states. These were converted to heatmaps by treating each observation as a Gaussian in both variables and normalizing as a function of the weather variable (to account for variation in sampling density). Effective median curves, as a function of the weather variable, were then calculated from the heatmaps.
To estimate total environmental pressure, an index combining high temperature, sunlight, and UV was formed by iteratively determining median curves as described above and adjusting for each effect until an effective linear combination was determined.
All analysis and original graphics were produced using Matlab.
