The rapid global spread of SARS-CoV-2, across diverse climates, prompted speculation about weather's role in transmission patterns. Analogies to seasonal respiratory viruses like influenza suggested potential modulation by temperature and humidity. However, the specific mechanisms remain unclear, involving host immunity, viral stability, and weather-influenced behaviors. Dynamic transmission models indicated that meteorological variables were unlikely to be dominant risk factors early in the pandemic due to high population susceptibility. Existing studies on the relationship between weather and COVID-19 transmission showed methodological weaknesses and conflicting results, with varying associations between temperature and COVID-19 response variables. Many lacked controls for government restrictions, socio-economic factors, population density, and age structure. This study aimed to address these methodological limitations by employing a two-stage ecological modeling approach.

Several studies explored the relationship between meteorological factors and COVID-19 transmission, but yielded inconsistent results. Some found positive associations between temperature and COVID-19 cases or deaths, while others found negative or non-linear associations, or no significant association at all. These inconsistencies stemmed from methodological weaknesses, such as the lack of control for crucial confounding factors like government interventions, socio-economic status, and population density. The lack of a complete annual cycle of data for the novel SARS-CoV-2 also made it difficult to distinguish between seasonal signals, inter-annual trends, and the effects of meteorological factors. The study highlighted the need for a robust approach that accounts for these confounding factors and addresses the data limitations.

The study utilized data from the Multi-Country Multi-City (MCC) Collaborative Research Network, encompassing daily COVID-19 case time series from January 11 to April 28, 2020, in 409 locations across 26 countries. A two-stage ecological modeling approach was employed. The first stage involved estimating the effective reproduction number (Re) for each city using a renewal equation-based approach, accounting for latent infections, incubation period, reporting delays, and a negative binomial observation model. This stage focused on the early phase of the pandemic to minimize biases introduced by later interventions. Meteorological data (temperature, humidity, solar radiation, wind speed, precipitation) from the ERA5 reanalysis dataset were averaged for each city over the early-phase time window. The second stage used a multilevel meta-regression model to estimate the association between city-level Re and meteorological variables, while controlling for population, population density, GDP per capita, percentage of population over 65, PM2.5 pollution levels, and the lagged Oxford COVID-19 Government Response Tracker (OxCGRT) Government Response Index. This accounted for the two-level structure of the data (cities nested within countries). Sensitivity analyses were conducted to evaluate the robustness of the results under different assumptions and model specifications. The authors used a Bayesian latent variable method using EpiNow2 1.3.2 for estimating Rt. The study accounted for uncertainty in R_t estimates by using a multilevel meta-analytic approach.

A modest, non-linear association was found between mean temperature and Re, with Re initially rising, peaking at 10.2°C, then falling to a trough at 20°C, and rising again. A similar non-linear association was observed for absolute humidity (AH). Relative humidity showed a weaker association. No significant associations were found for solar radiation, wind speed, and precipitation. The OxCGRT Government Response Index showed a strong, negative association with Re, explaining 13.8% of its variability (compared to 2.4% for temperature and 2.0% for AH). Temperature and AH were highly correlated which made it hard to separate out their individual effects. The study found little evidence that meteorological conditions influenced the early stages of local epidemics.

The findings suggest a relatively minor influence of meteorological factors on early-phase COVID-19 transmission compared to the impact of government interventions. The non-linear relationship between temperature/humidity and Re might be explained by mechanisms affecting viral survival, immune responses, and human behavior. The stronger effect of interventions underscores their critical role in controlling COVID-19 spread, regardless of meteorological conditions. The study's results align with some, but not all, previous research, possibly due to differences in methodology and data limitations. The inconsistencies in previous studies are attributed to factors like methodological challenges, incomplete seasonal data, and confounding effects of interventions. The ecological approach and the control of confounders in this study offer a more robust analysis.

This study, using a large dataset and advanced methodology, demonstrates that early public health interventions had a much more significant impact on COVID-19 transmission than weather patterns. While a minor non-linear relationship between temperature and humidity and transmission was identified, it was comparatively insignificant to the effectiveness of public health interventions. Future research should focus on investigating the complex interplay between weather, immune responses, and human behavior, and on developing more sophisticated models of transmission dynamics.

The study had several limitations. Cities in the northern hemisphere were overrepresented, potentially limiting the generalizability of findings to the southern hemisphere. Incomplete knowledge about the interaction between weather, viral characteristics, and immune responses introduces uncertainties. Variations in case reporting and data availability across cities and countries, and the limited timeframe of the data (less than a complete annual cycle), also limit the analysis. The high correlation between temperature and absolute humidity also makes it difficult to definitively separate their individual effects.