The Role of Climate in the Spread of COVID-19 in Different Latitudes across the World

Abstract

Most transmittable diseases appear in a specific season and the effect of climate on COVID-19 is of special interest. This study aimed to investigate the relationship between climatic variables and the R₀ of COVID-19 cases in a list of areas in different latitudes around the world. The daily confirmed cases of COVID-19 and climatic data of each area per day from January 2020 to March 2021 were utilized in the study. The GWR and MLR methods were used to identify the relationship between the R₀ of COVID-19 cases and climatic variables. The MLR results showed a significant (p-value < 0.05) weak inverse relationship between the R₀ of COVID-19 cases and wind speed, but a positive significant (p-value < 0.01) relationship with precipitation. This implies that lower COVID-19 cases were recorded with high wind speed and low precipitation. Based on GWR, the R₀ of COVID-19 infection assessed against air temperature, rh, and precipitation was found to be statistically significant using the Monte Carlo p-value test, and the effect of climatic variables on COVID-19 infection appears to vary geographically. However, besides climatic variables, many socioeconomic factors could influence the virus’s transmission and will need to be considered in future studies.

1. Introduction

The coronavirus disease COVID-19 emerged in China between the end of 2019 and early 2020. Since then it has rapidly spread to other countries around the world, and the World Health Organization declared COVID-19 a pandemic [1,2]. The number of confirmed cases and casualties of SARS-CoV2 has exceeded the number of SARS-CoV1 and MERS-CoV casualties [3].

Most transmittable diseases appear in a specific season; for instance, influenza is not readily transmitted in hot and humid situations and shows seasonality in the regions with a temperate and climate where the peak of infections happens during winter [1,4]. Likewise, SARS-CoV1 and MERS-CoV also showed seasonal patterns and correlations with temperature. The peak of the first occurred during the spring while the second was transmitted in warm climates during spring and summer seasons [5]. People stay indoors more in winter, which can assist the spread of diseases; and vitamin D levels in people tend to drop in winter, which may reduce their immunity [6]. These may be among the reasons for the more rapid spread of infectious diseases in winter.

R naught or zero (R₀) refers to the reproduction number or the effective reproduction number (R_e) of the virus. It describes the average number of people each infected person will infect where there is no pre-existing immunity in the community, based on three factors: the duration, the likelihood of infection and the frequency of contact [7]. R₀ is an exponent of the spread of a virus, firstly used to evaluate the capability of a disease to infect society, and secondly to locate the fraction of the society which should be vaccinated [8].

Even though cultural factors and population density are influencing the growth and spread of COVID-19, the effect of climate on the virus is of special interest [5] and cannot be neglected. Visual assessment of the global maps of COVID-19 infection display that the disease is less prevalent in hot and humid countries that lie on the Equator [6]. At the local scale, Ahmadi et al. [9] detected that COVID-19 infection in Iran is high in cities with low degrees of wind speed, humidity, and solar radiation. Rasul and Ibrahim [10] assessed the influence of climatic variables and sociodemographic characteristics on COVID-19 infections in Iraqi cities. In China, Wang et al. [11] used linear regression to reveal that high temperature and high humidity significantly decreased the spread of COVID-19 in 100 cities. However, further research is essential to assess the influence of climatic elements on the virus in different latitude areas on the global scale using multivariate methods instead of univariate approaches.

Therefore, this study aimed to investigate the relationship between climatic variables and the R₀ of detected COVID-19 cases by comparing the transmission performance of the virus in small regions around the world. The novel contribution of this research is the quantitative analysis of relationships between SARS-CoV2 spread and principal climatic elements in a list of areas in different latitudes around the world using the multivariate regression method. The research results will support policymakers in monitoring and improving climate-related health in cities, states, and small countries.

2. Materials and Methods

2.1. Data

A list of states, provinces (e.g., states of Canada, China, and Australia that provide COVID-19 data at the state level), and small countries (smaller than 20,000 km²; e.g., Malta, Qatar, Fiji, Lebanon, Kuwait, and Gambia) was selected for this study. The reason for selecting small regions was that COVD-19 infection data on the city scale are not available in all countries, so we ignored large countries where only COVID-19 data at the country scale were available.

Data of daily confirmed COVID-19 from January 2020 to March 2021 in selected areas were downloaded from Humdata.org (2020) [12]. Daily data of weather components (air temperature, relative humidity, atmospheric pressure, visibility, and wind speed) across weather stations in the study areas were obtained using the Worldmet package of R programming [13]. Surface meteorological data of regional weather stations were downloaded via the Worldmet package from the NOAA Integrated Surface Database (ISD).

2.2. Statistical Analysis

To obtain an accurate rate of transmission, only those data collected after 30 cases of COVID-19 were reported in each region were used in the R₀ production [14]. The basic R₀ transmissibility projected by Cori et al. [15] framework was used to estimate each COVID-19 patient spread to how many individuals. The method generates robust analytical speculates of R. When the result of R₀ is higher than 1, this means the virus spreads fast and has a larger spread with a higher number of R₀ values [16]. This tool estimates R_t from the time series of cases and therefore reproduces the R₀ of the pandemic [15].

The association between the R₀ of the daily confirmed cases of COVID-19 of selected regions and daily weather variables was assessed. We converted the hourly data of downloaded weather stations to daily data before data analyses. Geographically weighted regression (GWR) and multiple linear regression (MLR) approaches were used to quantify the linear and spatial regression analyses.

Statistical processing was performed in R programming (i.e., spgwr [17] and sp [18] packages). GWR multivariate regression was performed and preferred over univariate regression because in multivariate regression it is possible to quantify the partial contributions of each variable and obtain a robust assessment correlation between the R₀ of COVID-19 infection and explanatory climatic variables. GWR is the localized regression suggested by Brunsdon et al. [19], which assesses non-stationary variables [20]. The model is stated as Equation (1) and MLR is expressed as Equation (2). The GWR Gaussian weights function was applied for modeling GWR in Equation (3), while sp, ggplot2 [21] and some other packages were used for visualization results of the study.

$$y_{i }={\beta }_0\left(u_i, v_i\right)+{\displaystyle \sum }_k{\beta }_k(u_i, v_i)x_{ik}+{\epsilon }_i$$

(1)

where $\left(u_i, v_i\right)$ implies the coordinates of the $i$th point in space, ${\beta }_0$ and ${\beta }_k$ are parameters to be estimated, and ${\epsilon }_i$ is the random error term at point $i$.

$$y=b_0+b_1{x}_1+b_2{x}_2+\dots +b_n{x}_n+e$$

(2)

where $y$ is the dependent variable, $b_0$ is the y-intercept, $x$ the explanatory variables, $b_n$ slop coefficient for $x_n$, and e model error.

$$w\left(g\right)=e^{-{\left(d/h\right)}^2}$$

(3)

where d is the distance and h is the bandwidth.

3. Results and Discussion

The research aimed to display the influence of climatic variables on COVID-19 at state and country levels in different latitudes around the world. Results on this association will assist researchers to better compare the conditions favoring the new virus. Based on the daily climatic variables and R₀ of confirmed COVID-19 case analyses, the results are as follows.

3.1. Number of Cases and R₀ of COVID-19 Infection

As an example of areas examined in the study, Figure 1 displays the number of recorded cases and the R₀ in eight areas. In Victoria Australia (37.81 S), the peak occurred at the end of July and the start of August 2020 (700 people per day) while the highest R₀ occurred on 22 March 2020 when R₀ reached 11.6. Figure 1 displays three peaks in Bahrain (26.27 N): 12 February 2021 (896 cases per day), 16 September 2020 (841 cases per day), and 15 June 2020 (786 people per day). In terms of R₀, it reached the highest rank on 11 March 2020, which was 15.1. The figure displays the peak of cases in January 2021 and fluctuation in the number in Alberta Canada (53.08 N). The highest R₀ occurred at the beginning of the disease, in March 2020 (R₀ = 16.8). In May 2020 it fell to less than one, then increased to more than one during June and July 2020.

In the Hubei Province of China (31.37 N), which includes Wuhan, the first center of the pandemic, cases started at the end of January 2020 and reached a high on 13 February 2020, which was 14,840 cases per day. One of the highest R₀ rates occurred on 28 January 2020 (R₀ = 28.9). In Gambia (13.34 N), two peaks occurred in the number of cases, while the number of cases start rising during July 2020 in Montenegro (42.37 N). The figure displays fluctuations in the number of cases and the highest peak (1716 cases per day) on 4 March 2021, while the R₀ during the same day was 1.77. It stayed at more than one during the final day of the period of our research study but the highest R₀ was 7.47 during 14 March 2020. In Singapore, two peaks in the number of cases occurred, then the rate fell. The highest number of cases was 1426, recorded on 20 April 2020, but the highest R₀ (5.3) occurred on 10 February 2020 with the beginning of spread of the disease in the country.

3.2. Relationship of R₀ of COVID-19 Infection and Climatic Variables Based on MLR Method

“Pr(>|t|)” in

Table 1. Statistics of regression between climatic variables and lag7 R₀ of COVID-19 based on the multiple linear regression method.

	Estimate	Std.Error	tvalue	Pr(>|t|)
Intercept	−7.133151	3.209931	−2.22	0.0263 *
Wind Speed	−0.016102	0.009561	−1.68	0.0922.
Average Temperature	−0.002282	0.001980	−1.15	0.2492
Precipitation	0.008256	0.003130	2.64	0.0084 **
Rh	0.001559	0.001504	1.04	0.3000
Air Pressure	0.000101	0.000464	0.22	0.8276
After Removing Insignificant Variables
Intercept	−7.658160	3.023472	−2.533	0.01133 *
Wind Speed	−0.015446	0.009546	−1.618	0.10567
Precipitation	0.008842	0.002971	2.976	0.00293 **

Signif. codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1.

refers to significance and 0.05 has been used as a threshold for the statistical significance of associations. A small p-value indicates that we can reject the null hypothesis, which allows us to conclude that there is a relationship between the R₀ of COVID-19 spread and examined climatic variables. The model of MLR is significant (p < 0.01) in general, while at variables level, only precipitation and wind speed are significantly correlated with lag7 of the R₀ COVID-19 infection. This result is in agreement with Sajadi et al. [5], who found a close relationship between COVID-19 R₀ cases and climatic drivers exists. In particular, wind speed was negatively correlated with COVID-19 infection and increasing wind speed could potentially decrease COVID-19 infection by 0.016 per unit. In contrast, precipitation was found to be positively correlated with COVID-19 infection, meaning that an increase in rainfall could increase COVID-19 infection by 0.008 per mm unit (Table 1). In order to improve the model, after we removed insignificant variables and ran the model again, based on the MLR method, precipitation was the main variable found to influence the spread of COVID-19 (coefficient = 0.008842; p < 0.01).

3.3. Relationship between R₀ of COVID-19 Infection and Climatic Variables Based on GWR Method

The R₀ of COVID-19 infection assessed against precipitation, rh, and average air temperature was found statistically significant using the Monte Carlo p-value test. Generally, the GWR method showed a negative relationship between R₀ and wind speed (insignificant) and air temperature (significant) while the relationship was positive with precipitation and relative humidity (

Table 2. Summary statistics of GWR coefficient parameter estimates between daily climatic variables and lag7 R₀ of COVID-19.

	Min.	Median	Max.	Global	Monte Carlo p-Value
Intercept.	−62.7119	−5.2736	2.2711	−7.133	0.00
Wind Speed	−0.078	−0.0026	0.006	−0.016	0.33
Average Temperature	−0.016	−0.0027	0.022	−0.002	0.03
Rh	−0.0047	0.0065	0.0139	0.002	0.00
Precipitation	−0.00137	0.0058	0.0627	0.008	0.00
Air Pressure	−0.00053	−0.0002	0.00137	0.0001	0.15
After Removing Insignificant Variables
Intercept.	−65.8430	−5.4395	2.2491	−8.2481	0.00
Average Temperature	−0.0140	−0.0026	0.0182	−0.0021	0.04
Rh	−0.0045	0.0063	0.0146	0.0016	0.00
Precipitation	−0.0007	0.0060	0.0655	0.0093	0.00

). However, this negative relationship was changed to positive in some areas (

Table 3. GWR coefficient estimates between daily climatic variables and lag7 R₀ of COVID-19 in different latitudes.

Area	Latitude	Longitude	X.Intercept.	Wind Speed	Average Air Temperature	Relative Humidity	Precipitation	Air Pressure
French Guiana	64.1667	–51.7500	–4.1572	–0.0034	–0.0056	0.0088	0.0048	–0.0004
Isle of Man	54.0833	–4.6239	1.6814	0.0062	–0.0039	–0.0047	–0.0001	0.0003
Alberta	53.5667	–113.5167	2.2667	–0.0224	–0.0002	0.0065	–0.0014	–0.0004
Saskatchewan	50.2000	–104.7000	1.2325	–0.0163	–0.0008	0.0065	–0.0004	–0.0004
Manitoba	49.8833	–97.1333	0.3030	–0.0120	–0.0013	0.0066	0.0005	–0.0004
Luxembourg	49.6266	6.2115	1.6083	0.0059	–0.0012	–0.0044	–0.0001	–0.0001
Channel Islands	49.4350	–2.6020	1.7668	0.0051	–0.0029	–0.0045	–0.0002	0.0004
Newfoundland and Labrador	47.6186	–52.7519	–4.7149	–0.0026	–0.0046	0.0090	0.0053	–0.0004
Liechtenstein	47.1333	9.5167	1.5150	0.0053	–0.0001	–0.0040	–0.0000	–0.0002
Prince Edward Island	46.2833	–63.1333	–4.2993	–0.0022	–0.0039	0.0089	0.0049	–0.0005
New Brunswick	45.9200	–66.6000	–3.9759	–0.0023	–0.0036	0.0087	0.0046	–0.0005
Jilin	43.9964	125.6808	–62.7119	–0.0785	–0.0160	0.0124	0.0628	0.0002
Ontario	43.6772	–79.6306	–2.4008	–0.0039	–0.0027	0.0076	0.0031	–0.0005
Montenegro	42.3667	19.2500	1.0683	0.0038	0.0030	–0.0028	0.0003	–0.0003
Xinjiang	41.7333	85.8167	–52.2264	–0.0167	–0.0037	0.0111	0.0522	–0.0001
Liaoning	40.6667	122.2000	–62.2737	–0.0679	–0.0155	0.0128	0.0623	0.0002
Hebei	40.0801	116.5846	–61.7733	–0.0566	–0.0149	0.0133	0.0617	0.0002
Shanxi	37.7469	112.6284	–61.5048	–0.0453	–0.0139	0.0137	0.0614	0.0002
Ningxia	37.4833	105.6833	–60.8694	–0.0317	–0.0121	0.0140	0.0607	0.0001
Shandong	36.1833	118.1500	–61.7997	–0.0536	–0.0146	0.0132	0.0617	0.0002
Gibraltar	36.1512	–5.3497	2.0293	–0.0002	–0.0013	–0.0035	–0.0005	0.0010
Malta	35.8575	14.4775	1.4610	0.0015	0.0031	–0.0027	–0.0001	0.0001
Cyprus	35.1500	33.4000	0.2711	–0.0004	0.0070	–0.0006	0.0009	–0.0002
Lebanon	33.8209	35.4884	0.1218	–0.0017	0.0074	–0.0003	0.0010	–0.0001
Jiangsu	32.8500	120.2833	–61.4212	–0.0524	–0.0146	0.0130	0.0614	0.0002
Bermuda	32.3667	–64.6833	–4.6579	–0.0013	–0.0035	0.0089	0.0052	–0.0005
Hubei	31.1667	112.6333	–61.2376	–0.0355	–0.0131	0.0136	0.0611	0.0002
Sichuan	29.9833	103.0000	–60.7495	–0.0179	–0.0101	0.0139	0.0605	0.0001
Kuwait	29.3833	48.0000	–2.5896	–0.0099	0.0075	0.0009	0.0036	–0.0000
Zhejiang	28.4500	119.9167	–60.3267	–0.0450	–0.0142	0.0127	0.0603	0.0002
Hunan	27.1833	111.4500	–60.6136	–0.0277	–0.0123	0.0135	0.0604	0.0001
Jiangxi	27.1167	114.9667	–60.3716	–0.0339	–0.0131	0.0131	0.0602	0.0002
Guizhou	26.5385	106.8007	–60.6065	–0.0192	–0.0109	0.0137	0.0604	0.0001
Bahrain	26.2708	50.6336	–4.1320	–0.0138	0.0073	0.0013	0.0051	0.0000
Qatar	25.2611	51.5651	–4.8284	–0.0151	0.0073	0.0014	0.0058	0.0000
Yunnan	25.0167	101.5167	–60.1998	–0.0108	–0.0088	0.0136	0.0599	0.0001
Guangxi	24.3500	109.4000	–60.0336	–0.0205	–0.0113	0.0133	0.0598	0.0001
Guangdong	23.3924	113.2988	–59.3091	–0.0258	–0.0123	0.0129	0.0592	0.0002
Hong Kong	22.4000	114.2000	–58.7490	–0.0261	–0.0123	0.0127	0.0586	0.0002
Cayman Islands	19.6870	–79.8828	–3.5209	–0.0014	–0.0027	0.0075	0.0042	–0.0005
Hainan	19.5167	109.5833	–58.1775	–0.0152	–0.0107	0.0127	0.0580	0.0001
Sint Maarten	18.0458	–63.1147	–5.2734	–0.0006	–0.0032	0.0090	0.0058	–0.0005
Jamaica	17.9357	–76.7875	–3.9823	–0.0010	–0.0028	0.0078	0.0046	–0.0005
Saint Lucia	17.3112	–62.7187	–5.3245	–0.0006	–0.0032	0.0091	0.0059	–0.0005
Dominica	15.3330	–61.3830	–5.4704	–0.0006	–0.0032	0.0091	0.0060	–0.0005
Cabo Verde	14.9333	–23.4833	0.3947	–0.0098	–0.0019	0.0019	0.0007	0.0008
Martinique	14.6000	–61.0000	–5.5158	–0.0006	–0.0032	0.0092	0.0060	–0.0005
Gambia	13.3380	–16.6522	1.8652	–0.0113	–0.0006	–0.0001	–0.0006	0.0014
Aruba	12.5000	–70.0170	–4.9454	–0.0004	–0.0029	0.0085	0.0055	–0.0005
Curacao	12.2000	–68.9670	–5.0529	–0.0004	–0.0030	0.0086	0.0056	–0.0005
Bonaire, Sint Eustatius and Saba	12.1500	–68.2830	–5.1142	–0.0004	–0.0030	0.0087	0.0057	–0.0005
Grenada	12.0000	–61.7830	–5.5728	–0.0004	–0.0031	0.0092	0.0061	–0.0005
Trinidad and Tobago	10.5833	–61.3500	–5.6484	–0.0004	–0.0031	0.0092	0.0062	–0.0004
Brunei	4.9442	114.9284	–42.5733	–0.0138	–0.0089	0.0071	0.0430	0.0002
Maldives	4.1918	73.5291	–38.9624	–0.0246	0.0075	0.0045	0.0394	0.0001
Seychelles	–4.6743	55.5218	–16.8494	–0.0527	0.0169	0.0014	0.0176	0.0001
Comoros	–11.5337	43.2719	–9.2050	–0.0702	0.0213	0.0013	0.0101	0.0000
Mayotte	–12.8047	45.2811	–11.1706	–0.0720	0.0223	0.0009	0.0120	0.0001
Fiji	–18.0433	178.5592	–11.8107	–0.0227	0.0081	–0.0037	0.0130	0.0007
New Caledonia	–22.0146	166.2130	–11.7658	–0.0220	0.0091	–0.0033	0.0129	0.0009
Western Australia	–31.9403	115.9669	–10.1166	–0.0075	0.0074	–0.0023	0.0112	0.0010
Australian Capital Territory	–35.3069	149.1950	–11.4782	–0.0185	0.0113	–0.0027	0.0125	0.0013

).

Regarding the spatial variation of regression, the effects of wind speed, temperature, humidity, precipitation, and air pressure on COVID-19 infection appear to vary geographically. Table 3 displays that increasing air temperature led to decreasing R₀ of COVID-19 in some areas; for instance, in the Chinese provinces, Aruba, Brunei, and New Brunswick in Canada (negative regression), while it led to increasing infection in some areas such as Qatar, Mayotte, and Western Australia (positive regression, Table 3). Wind speed was (statistically insignificantly) negatively correlated to COVID-19 infection in most examined areas in Asia, Australia, and Caribbean Islands while positively correlated with COVID-19 infection in the Isle of Man, Luxembourg, Channel Islands, and Montenegro. Relative humidity was found to be negatively related to infection in Europe (e.g., Gibraltar, Malta, Luxembourg, and Montenegro), while increasing relative humidity led to an increase in COVID-19 infection in the most examined areas, (e.g. Jilin, Ontario, Lebanon, and Hubei province). Air pressure was found insignificant in a negative relationship with infection of COVID-19 in areas such as Canada, Aruba, and Jamaica while in a positive relationship in countries such as Australia, Gambia, and Gibraltar. Precipitation was found to have a significant positive influence on the R₀ of COVID-19 in most examined areas, for instance in China, Brunei, and the Maldives, while having a minor negative influence in Gibraltar, Alberta, and Saskatchewan provinces in Canada.

Based on both MLR at the global scale and GWR at local levels, precipitation was significantly correlated with R₀ spread of COVID-19 and high rainfall was associated with increasing cases. In our research, comparing OLS and GWR, based on Brunsdon, Fotheringham, and Charlton’s (2002) ANOVA test, the GWR (SS residuals = 49,865, AICc = 33,514.041) provides a statistically significant (p-value 0.3) improvement over OLS (SS residuals = 5,0431, AICc = 33,591.833).

Figure 2 shows the GWR-based predicted R₀ of COVID-19 infection, the standard error of predicted R_0, and localR2. High pred_se was observed in East of Asia. In general, localR2 was low and less than 0.1. The highest localR2 values were 0.053 and 0.052 in Zhejiang and Hong Kong provinces in China, respectively. Figure 3 displays that predicted and observed R₀ was statistically significantly correlated and predicted R₀ based on GWR had less rmse (2.12) than the rmse of predicted R₀ based on OLS (2.13). The low R square values in Figure 3 (while the relationship was significant statistically), may be the reason only a small portion of y (COVID-19 spread) was explained by x (climatic variables). There are many measurements and socioeconomic factors that could be responsible for COVID-19 spread and are not included in our model.

4. Conclusions

While the cases and deaths of COVID-19 have risen, assessing the influence of climate on the spread of the virus in different latitude areas on the global scale is essential, using multivariate methods (e.g., GWR) instead of univariate approaches.

Our results report that based on both MLR and GWR methods, precipitation is the main significant climatic driver of the R₀ of COVID-19 cases both at the global and local scale. Based on GWR, the R₀ of COVID-19 infection assessed against precipitation, rh, and average air temperature was found statistically significant; however, the effect of climatic variables appears to vary geographically. In our research, comparing OLS and GWR, the GWR provides a statistically significant (p-value 0.3) improvement over OLS.

Previous studies focusing on socioeconomic parameters have shown that population density, hospital, diabetes, and aged patients have been found to be the main drivers controlling the R₀ of COVID-19 cases. In future research, we recommend examining the combined effects of socioeconomic and climatic parameters to assess sustainable cities, countries, and societies.