Spatial modeling could not differentiate early SARS-CoV-2 cases from the distribution of humans on the basis of climate in the United States

Abstract

The SARS-CoV-2 coronavirus is wreaking havoc globally, yet, as a novel pathogen, knowledge of its biology is still emerging. Climate and seasonality influence the distributions of many diseases, and studies suggest at least some link between SARS-CoV-2 and weather. One such study, building species distribution models (SDMs), predicted SARS-CoV-2 risk may remain concentrated in the Northern Hemisphere, shifting northward in summer months. Others have highlighted issues with SARS-CoV-2 SDMs, notably: the primary niche of the virus is the host it infects, climate may be a weak distributional predictor, global prevalence data have issues, and the virus is not in population equilibrium. While these issues should be considered, we believe climate's relationship with SARS-CoV-2 is still worth exploring, as it may have some impact on the distribution of cases. To further examine if there is a link to climate, we build model projections with raw SARS-CoV-2 case data and population-scaled case data in the USA. The case data were from across March 2020, before large travel restrictions and public health policies were impacting cases across the country. We show that SDMs built from population-scaled case data cannot be distinguished from control models (built from raw human population data), while SDMs built on raw case data fail to predict the known distribution of cases in the U.S. from March. The population-scaled analyses indicate that climate did not play a central role in early U.S. viral distribution and that human population density was likely the primary driver. We do find slightly more population-scaled viral cases in cooler areas. Ultimately, the temporal and geographic constraints on this study mean that we cannot rule out climate as a partial driver of the SARS-CoV-2 distribution. Climate's role on SARS-CoV-2 should continue to be cautiously examined, but at this time we should assume that SARS-CoV-2 will continue to spread anywhere in the U.S. where governmental policy does not prevent spread.

Keywords: COVID-19; Climate; Coronavirus; SARS-CoV-2; Species distribution modeling; US.

Figure 1. Probability densities of SARS-CoV-2 coronavirus cases (using population scaled data; curves in red) compared to the probability densities of human populations (curves in blue) in each US county for each of seven climate variables (A–C) Average Temperature (C); (D–F) Max Temperature (C); (G–I) Minimum Temperature (C); (J–L) Precipitation (mm); (M–O) Solar Radiation (kJ m⁻²/day); (P–R) Wind Speed (m/s); and (S–U) Water Vapor Pressure (kPa) at three time periods (March 3, 2020; March 16, 2020; March 30, 2020).

Probability density curves are standardized to an area of one.

Figure 2. The relationship in the US between human population size and SARS-CoV-2 coronavirus cases, using (A) total viral cases and (B) population scaled viral cases.

New York City, an outlying point, has been excluded for clearer visualization.

Figure 3. (A) Species distribution model of the SARS-CoV-2 coronavirus (using population scaled data) for 30 March 2020. (B) Human population distribution model for the US from 2010.

Figure 4. (A) Niche overlap and (B) similarity tests for Maxent species distribution models built with population scaled SARS-CoV-2 coronavirus data compared to one built with human population density as occurrence data; actual model overlap indicated by a red marker in both plots.

Significant p-values correspond to greater niche overlap or similarity than expected by random models.