Suitability Map of COVID-19 Virus Spread

This image reports a Maximum Entropy model that estimates <em>suitable </em>locations for COVID-19 spread, i.e. places that could favour the spread of the virus just in terms of environmental parameters. The model was trained just on locations in <em>Italy </em>that have reported a rate of new infections higher than the geometric mean of all Italian infection rates. The following environmental parameters were used, which are correlated to those used by other studies: Average Annual Surface Air Temperature in 2018 (NASA) Average Annual Precipitation in 2018 (NASA) CO2 emission (natural+artificial) averaged between January 1979 and December 2013 (Copernicus Atmosphere Monitoring Service) Elevation (NOAA ETOPO2) Population per 0.5° cell (NASA Gridded Population of the World) A higher resolution map, the model file (in ASC format) and all parameters used are also attached. The model indicates highest correlation with <em>infection rate</em> for CO2 around 0.03 gCm^−2day^−1, for Temperature around 11.8 °C, and for Precipitation around 0.3 kg m^-2 s^-1, whereas Elevation and Population density are poorly correlated with <em>infection rate</em>. <strong>One interesting result is that the model indicates, among others, the Hubei region in China as a high-probability location</strong>, <strong>and Iran (around Teheran) as a suited location for virus' spread, but the model was not trained on these regions, i.e. it did not know about the actual spread in these regions.</strong> <strong>Evaluation: </strong> A <em>risk score</em> was calculated for each country/region reported by the JHU monitoring system (https://gisanddata.maps.arcgis.com/apps/opsdashboard/index.html#/bda7594740fd40299423467b48e9ecf6). This score is calculated as the summed normalised probability in the populated locations divided by their total surface. This score represents how much the zone would potentially foster the virus' spread. We assessed the reliability of this score, by selecting the country/regions that reported the <em>highest rates of infection</em>. These zones were selected as those with a rate higher than the upper confidence of a log-normal distribution of the rates. The agreement between the two maps (covid_high_rate_vs_high_risk.png, where violet dots indicate <em>high infection rates </em>and countries' colours indicate estimated <em>high risk score</em>) is the following: <strong>Accuracy </strong>(overall percentage of correctly predicted high-rate zones): <strong>77.25%</strong><br> <strong>Kappa </strong>(agreement between the two maps): <strong>0.46</strong> (Good, according to Fleiss' intepretation of the score) <strong>This assessment demonstrates that our map can be used to estimate the risk of a certain country to have a high rate of infection, and indicates that the influence of environmental parameters on virus's spread should be further investigated.</strong>本图展示了一款最大熵模型（Maximum Entropy Model），该模型仅依据环境参数，即可预估新冠病毒传播的*适宜区位*——即能够助力病毒传播的区域。本模型仅以*意大利*境内新增*感染率*高于全意大利感染率几何均值的区域作为训练样本。 本研究采用了与其他同类研究相关联的以下环境参数：2018年平均地表气温（NASA）、2018年平均年降水量（NASA）、1979年1月至2013年12月平均的二氧化碳（自然+人为）排放量（哥白尼大气监测服务，Copernicus Atmosphere Monitoring Service）、海拔（NOAA ETOPO2）、每0.5°网格单元的人口数（NASA世界人口网格数据集，NASA Gridded Population of the World）。 本研究同时附上了更高分辨率的风险分布图、模型文件（ASC格式）以及所用全部环境参数数据集。 模型结果显示，当二氧化碳排放量约为0.03 g·cm⁻²·day⁻¹、地表气温约为11.8℃、年降水量约为0.3 kg·m⁻²·s⁻¹时，与*感染率*的相关性最高；而海拔与人口密度则与*感染率*相关性较弱。**值得关注的是，模型将中国湖北省判定为高传播概率区域，同时将伊朗（德黑兰周边）判定为病毒易传播的适宜区域，但本模型并未以这些区域作为训练样本，即模型并未接触过这些区域的实际疫情传播数据。** **评估环节**： 我们为约翰·霍普金斯大学（JHU）监测系统（https://gisanddata.maps.arcgis.com/apps/opsdashboard/index.html#/bda7594740fd40299423467b48e9ecf6）所报告的每个国家/区域计算了*风险评分*。该评分的计算方式为：有人聚居区域的归一化概率总和除以该区域的总表面积，可反映对应区域潜在的病毒传播助力程度。 为评估该*风险评分*的可靠性，我们选取了报告*最高感染率*的国家/区域作为分析样本：即感染率高于感染率对数正态分布（log-normal distribution）上限置信区间的区域。两张分布图（covid_high_rate_vs_high_risk.png，其中紫色圆点代表*高感染率*区域，各国配色代表估算的*高风险评分*）的一致性结果如下： **准确率（正确预测高感染率区域的整体占比）：77.25%** **Kappa系数（两张分布图的一致性指标）：0.46，根据弗莱斯（Fleiss）的评分解读标准，该结果属于良好水平。** **本次评估证明，本研究所用的风险分布图可用于预估各国/地区的高感染率风险，同时也提示环境参数对病毒传播的影响仍有待进一步深入研究。**