key: cord-1027716-ioyl1gla authors: Pramanik, M.; Chowdhury, K.; Rana, M. J.; Bisht, P.; Pal, R.; Szabo, S.; Pal, I.; Behera, B.; Liang, Q.; Padmadas, S. S.; Udmale, P. title: Climatic influence on the magnitude of COVID-19 outbreak: a stochastic model-based global analysis date: 2020-06-04 journal: nan DOI: 10.1101/2020.06.02.20120501 sha: d1f84f1c0661ec7acfea182878ce95e5dbb68172 doc_id: 1027716 cord_uid: ioyl1gla This study examines the association between community transmission of COVID-19 cases and climatic predictors, considering travel information and annual parasite index across the three climatic zones, i.e., tropical, subtropical, and temperate. A Boosted Regression Tree model has been employed to understand the association between the COVID-19 cases. The results show that average temperature and average relative humidity are the major contributors in explaining the differentials of COVID-19 transmission in temperate and subtropical regions whereas the mean diurnal temperature range and temperature seasonality are the most significant determinants in tropical regions. The average temperature is the most influential factor affecting the number of COVID-19 cases in France, Turkey, the US, the UK, and Germany, and the cases decrease sharply above 10oC. Among the tropical countries, India found to be most affected by mean diurnal temperature, and Brazil fazed by temperature seasonality. Most of the temperate countries like France, USA, Turkey, UK, and Germany with an average temperature between 5-12oC had high number of COVID-19 cases. The findings are expected to add to the ongoing debates on the influence of climatic factors influencing the number of COVID-19 cases and could help researchers and policymakers to make appropriate decisions for preventing the spread. The global surge of pandemic 1 Severe Acute Respiratory Syndrome (SARS) coronavirus disease Historical evidence shows that meteorological conditions, e.g., temperature and relative humidity S2), and hence these two variables were dropped from the analysis. 44 COVID-19 cases were 2 2 4 selected as outcome variable along with a set of six independent variables or predictors (see 2 2 5 Table 1 ): average temperature, diurnal temperature change, temperature seasonality, relative 2 2 6 humidity, number of travelers, and API. The motivation for boosting regression was to improving various weak learners by combining 2 2 9 two powerful procedures: regression tree and boosting. 40, 45 The following gradient boosting 2 3 0 model considers the forward stage-wise manner by adding the trained model from F, an 2 3 1 approximation function of the response variable. 2 3 3 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 4, 2020. (2) 2 3 9 which is similar to 2 4 1 β m in equation (3) It indicates that β m equalize the negative gradient of the squared loss function. Moreover, To avoid overfitting, a simple regularization strategy is to scale the contribution of each 2 5 0 regression tree by a factor The parameter ν is the learning rate as it magnifies the length of the gradient descent procedure. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 4, 2020. In this model, a 25% sample were used for training, and 75% sample distributed for testing. This 2 6 0 method has been simulated 1,000 times to generate statistical inference by using ten times the 2 6 1 loss function by cross-validation. In each BRT model, the subsampling procedure requires a 2 6 2 parameter called the 'bag fraction' which was set at 0.75 49 , and at least 1,000 nodes/trees were 2 6 3 used. 40 In addition, a sensitivity analysis was conducted by setting a bag fraction of 0.5. All 2 6 4 results presented in the following sections were calculated by averaging the predicted values of 2 6 5 50 bootstrap replicates. All analyses were conducted using DISMO package version Rv3.4.0. Moreover, the marginal association was assessed for all independent variables across climatic 2 6 7 regions and the countries with major COVID-19 cases spillover. The relative contribution of . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 4, 2020. . https://doi.org/10.1101/2020.06.02.20120501 doi: medRxiv preprint 1 5 Table 3 presents the association between COVID-19 and climatic parameters, number of 2 9 9 international passengers and API based on aggregate global model. We excluded the information 3 0 0 about the number of international travelers for country-level analysis due to single data for the 3 0 1 country level. Although it is a primary source of infection, it has no role in community Table 3 represents the region and country wise association between climatic parameters and the COVID-19 cases. The results show that in France, Turkey, the US, the UK, Germany, the 3 5 2 number of COVID-19 cases were non-linearly but highly associated with average temperature. Maximum cases were found during the temperature range of 5 to 10°C, and after the temperature increased beyond 10°C infected cases declined. Similarly, the average relative humidity was a 3 5 5 contributing factor in Spain, the UK, and Italy, and favorable relative humidity for the disease 3 5 6 transmission was found to be 60 to 70% in temperate countries. Most interestingly in the case of 3 5 7 Turkey, it was found that the cases were increasing after crossing the 73% threshold of relative humidity. The temperature seasonality mostly influenced the Russian cases. About 92% of the 3 5 9 cases in Russia were influenced by temperature seasonality, followed by Italy (64.3%), and the 3 6 0 US (21.5%). It concludes that more than 70% variation of temperature (temperature seasonality) 3 6 1 may cause a significant increase in COVID-19 community transmission. But with the 80% of 3 6 2 temperature seasonality, there was a declining trend for the US cases, whereas Russian cases Russia is northwards than the US where extreme seasonality was found. Another important . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 4, 2020. . https://doi.org/10.1101/2020.06.02.20120501 doi: medRxiv preprint It was found that the COVID-19 community transmission in the tropical zone was not strongly transmission of influenza epidemic. 12 In recent times, several efforts have also been made to 3 8 0 evaluate the association between climatic predictors and COVID-19 transmission. [26] [27] [28] [29] 53 Existing 3 8 1 studies mainly focused on regional perspectives of COVID-19 transmission and its association 3 8 2 with climatic conditions. However, studies at the macro level are limited, in particular, the . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 4, 2020. considerably low more than 10°C temperature. 27 The present study found a linear relationship 4 0 5 between the transmission of COVID-19 and temperature in the temperate region, while there was 4 0 6 no significant association between these two in the tropical region. As China is from a temperate 4 0 7 region, with an increase in temperature, the number of COVID-19 cases also increased in the for smooth transmission. Also, in the temperate and subtropical regions, COVID-19 transmission 4 1 0 was lower when the temperature remains below 10°C. Possibly, in these regions, the unfavorable 4 1 1 temperature keeps people inside their homes, and "social distancing" was maintained. Therefore, 4 1 2 the temperature might have played a significant role in the dispersion of the virus in the 4 1 3 . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 4, 2020. temperature were associated differently with the outbreak in different climatic regions such as 4 1 7 the temperate and tropical zones, it may also vary over regional/country levels due to changes in 4 1 8 geographical and ecological settings. Thus, the regional level analysis of heterogeneous climatic 4 1 9 associations with the transmission is equally necessary along the global assessments. The present study found that the role of average relative humidity on COVID-19 transmission China. 56 The relationship between relative humidity and COVID-19 cases can be complicated in CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 4, 2020. . https://doi.org/10.1101/2020.06.02.20120501 doi: medRxiv preprint weaken the symptoms of COVID-19. 35 The API values slightly positively influenced the number of COVID-19 cases in these countries, and the rate of influence was very low because cities are Other strains of coronavirus such as HCoV-HKU1, HCoV-229E, HCoV-OC43, and HCoV- Report 92 2020a. protein-on-the-new-coronavirus (Accessed on 21th April 2020). real time. The Lancet infectious diseases. 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 4, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 4, 2020. [50] Fang LQ, Li XL, Liu K, Li YJ, Yao HW, Liang S, Yang Y, Feng ZJ, Gray GC, Cao WC. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 4, 2020. personal protective equipment contamination by severe acute respiratory syndrome 6 5 8 coronavirus 2 (SARS-CoV-2) from a symptomatic patient. JAMA. 2020.. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 4, 2020. . https://doi.org/10.1101/2020.06.02.20120501 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 4, 2020. . https://doi.org/10.1101/2020.06.02.20120501 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 4, 2020. . . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 4, 2020. . https://doi.org/10.1101/2020.06.02.20120501 doi: medRxiv preprint Fig 4: Relationship between selected climatic variables, the number of international travelers, and the number of COVID-19 cases in the temperate region, and a similar relationship has been observed for the sub-tropical region. Fig (a) CoVID-19 , relative humidity, and temperature; (b) COVID-19, diurnal range of temperature, and relative humidity; (c) COVID-19, temperature seasonality, and diurnal temperature change; (d) COVID-19, average temperature and API values. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 4, 2020. . https://doi.org/10.1101/2020.06.02.20120501 doi: medRxiv preprint Table 3 . Y-axes are showing number COVID-19 cases, and X-axes represent predictors. The predictor, the number of international travelers, was omitted as this is a primary source of infection, but it has no role in community transmission. The API values considered only for the tropical region, where most of the countries of this region are malaria-prone. The shaded line shows a 95% confidence interval from the mean. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 4, 2020. . https://doi.org/10.1101/2020.06.02.20120501 doi: medRxiv preprint CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 4, 2020. . Table 3 . Y-axes are showing the number of COVID-19 cases, and X-axes represent predictors. The predictor, the number of international travelers, was omitted as it is a primary source of infection, but it has no role in community transmission. The API was omitted for fewer malaria cases in the temperate region. The shaded line shows a 95% confidence interval from the mean. (e) . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 4, 2020 . . https://doi.org/10.1101 /2020 Table 3 . Y-axes are showing COVID-19 cases, and X-axes represent predictors. The predictor, the number of international travelers, was omitted, as it is a primary source of infection, but has no role in community transmission. The value of API was considered in the analysis as these countries are malaria-prone. Shaded line shows a 95% confidence interval from the mean. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 4, 2020. . . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this preprint this version posted June 4, 2020. . https://doi.org/10.1101/2020.06.02.20120501 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 4, 2020. . https://doi.org/10.1101/2020.06.02.20120501 doi: medRxiv preprint . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) The copyright holder for this preprint this version posted June 4, 2020. . https://doi.org/10.1101/2020.06.02.20120501 doi: medRxiv preprint Climatic zones Tropical 25 The countries with the highest number of COVID-19 cases