Applications of GIS and geospatial analyses in COVID-19 research: A systematic review

Guideline and data sources

This systematic review was conducted using the reporting guidelines as stated in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement¹³. A protocol was prepared before conducting this review that was not registered with PROSPERO or any other organization. This protocol was uniformly followed by the reviewers at each stage of the review, which is available upon request. The data for this systematic review were retrieved from MEDLINE and Web of Science. Both databases have their own competitive advantages that provide a broader coverage of scholarly articles. MEDLINE is considered as the largest bibliographic databases for health sciences, whereas Web of Science provides access to journals from multiple scholarly disciplines. We used the following steps to identify relevant literature for this review. First, we used the following search query in each database: (“COVID-19” OR “2019-nCoV” OR “2019 coronavirus” OR “2019 novel coronavirus” OR “novel coronavirus” OR “SARS-CoV-2”) to identify COVID-19 related studies. Further, we used another search query to retrieve GIS-related studies as following: (“GIS” OR “ArcGIS” OR "Geographic information systems" OR "Geographic mapping" OR "Spatial analysis" OR "Geospatial analysis"). At the next step, we combined both these queries with “AND” operator to identify literature that is likely to contain studies referring to both these topics. Furthermore, as COVID-19-related literature is evolving rapidly, we also searched the Google Scholar database to identify studies that may align with the objective of this review. Also, we performed a reference searching and contacted subject matter experts for additional studies beyond the scope of the databases, if there were any. We limited the timeline for literature searching between 2019 and 2020 considering the origin of the outbreak in late 2019. The search was first conducted on May 7, 2020 and updated on June 28, 2020.

Eligibility criteria

Articles were considered eligible for this review if they were a) published in English language, b) available as peer-reviewed journal articles, c) the primary focus of the paper was on any aspect of the COVID-19 pandemic, d) demonstrated the applications of GIS or geospatial analyses. Any article that did not comply any of these criteria were excluded from this review. For example, non-English studies, articles that were not peer-reviewed (e.g., letters, editorials, comments), studies that did not focus on COVID-19, or did not use GIS or geospatial techniques (conceptual papers without providing any findings) were considered ineligible for inclusion.

Selection of studies and synthesis of evidence

All citations retrieved through database searching were imported in Endnote reference manager software for curating the collective bibliography. Further, this library was exported to Rayyan QCRI, a cloud-based software for citations screening and assessment. Two authors independently assessed each citation against the eligibility criteria stated above. At the end of the process, a third reviewer was consulted to review the conflicts and a consensus on inclusion or exclusion was made upon discussion. The full texts of the primarily selected articles were reviewed and reviewed by all authors.

Further, articles fulfilling all criteria for this review were retained and data were extracted using a predesigned extraction sheet in Microsoft Excel on the following variables:

a) publication details,

b) study objectives,

c) the sources of data,

d) countries of origin,

e) COVID-19 specific domain presented in the papers,

f) the use of GIS or geospatial analyses, and

g) the research outcomes or key findings of those studies.

The applications of GIS or geospatial analyses on COVID-19 studies were identified and narratively synthesized as major themes alongside tabulation of the key findings. A narrative synthesis is appropriate where quantitative or qualitative synthesis may not be feasible due to the methodological differences and other measures of heterogeneity across the included studies¹⁴. Moreover, quality appraisal or risk of bias assessment were not considered in this review due to a large volume of studies and profound heterogeneity in methodological approaches, data sources, measurements, types of applications adopted in different contexts, and research outcomes in respective studies.

Summary and characteristics of reviewed articles

The PRISMA diagram for this review is shown in Figure 1. We found a total of 79 articles which met all the inclusion criteria. A PRISMA checklist enlisting the contents of this systematic review is available on Open Science Framework repository¹⁵. A summary of these 79 reviewed articles are listed in Extended data, Table 1¹⁶. The articles included in the review were published in a diverse group of journals, mostly in public health, urban planning, geography, and interdisciplinary journals. Although COVID-19 is more of a public health and welfare issue, these articles covered from public health issues to planning techniques, environmental concerns, and, most importantly, used geospatial analysis tools and techniques as part of their methods. In total, 32.9% of the articles (n=26) used China and 21.5% the United States (n=17) as their study site, which indicates to the disparity of the geographic coverage. Figure 2 shows the geographic distribution of the study area used in the 79 articles we included in our review. Wuhan, China, first reported COVID-19 cases, and the GIS has the highest number of confirmed cases, which also was reflected in the over-representation of these two countries in these articles. Africa and Europe had the lowest number of articles (n=4). Among these reviewed articles, 15% (n=12) worked on a global or multi-national scale. The global studies are more focused on mobility and how COVID-19 transmitted via airports and other human mobility^17–19.

Figure 1. GIS Applications in COVID-19 research literature search and evaluation for inclusion.

Figure 2.

(a) article count per country and geographic distribution of the study sites; (b) number of articles in each continents based on study site along with Global scale studies; (c) number of articles in each thematic groups used in this review; and (d) bubble chart showing the frequency of each software or tool used for spatial analysis.

A wide range of geospatial tools were used in the reviewed articles. Most of these articles used regression analysis (n=18) and correlation analysis (n=11) for statistical analysis. Most prominently used spatial analysis was hotspot analysis using either kernel density function or other density techniques (n=16) followed by the spatial autocorrelation analysis using either global or local Moran’s Index (n=13) and proximity analysis (n=5). All the articles used different visualization techniques to display either the intermediate or the end products. Along with that, suitability analysis (n=3) and sensitivity analysis (n=4) was also used to find either access to hospitals or how the cases are distributed. One of the less frequently implemented but interesting technique was to track transmission patterns using the mobility data^20–22. Similarly, remote sensing data was used for air quality and pollution measurement, but not that prominently (n=6). Nighttime imagery and solar radiation data were used by n=5 studies to compare the energy consumption differences between before and after pandemic situation.

ESRI ArcGIS is the most used platform (n=36) in these articles, but other open-source platforms like RStudio, where spatial and statistical analysis can be done simultaneously, were often used as well (n=6). Besides RStudio, QGIS (n=4) is another open source software platform that was highly employed by the researchers. Apart from that, a couple of the studies developed their own platforms and one of the studies used GPS. data. Most of these articles used COVID-19 confirmed cases as part of their analysis (n=77). In most cases, researchers used either WHO provided data or JHU provided data for the research purpose. As several spatial analysis was performed based on remotely sensed data, NASA and USGS satellite images were the primary source for those (n=5). Apart from that, the air quality index (AQI) was used in the articles focused on environmental analysis (n=8). As mentioned in the methodology, we did not assess the risk of bias among studies included in this review.

Thematic groups

The rapid diffusion of COVID-19 and data convenience had enforced the global scientific community to work more vigorously on geospatial analysis of this pandemic. These studies had focused on distinct aspects of the pandemic and have different inputs. The articles included in this review was divided into six thematic groups- environment (n=25), socio-economic and cultural (n=13), Population Health Surveillance (n=8), spatial transmission (n=9), computer-aided spatial and statistical analysis and modeling (n=17) and big data, social media and mobile data (n=10).

Environment. We found a total of 25 articles that emphasized the use of GIS and spatial analysis on environmental issues related to COVID-19^18,20–43. Although these studies had a similar research interest, they differ significantly based on the spatial scale- from global^23–27 to regional^28,29, national^30–38 and small local scale studies^39–45. Several works examined the relationship between different meteorological factors and the transmission potential of the virus. They included a diverse set of characteristics in the studies: temperature^{21,25–27,30,31,33,35–37,39,42,44,46}, humidity^{21,26,27,31,33,35–37,39,44,46}, precipitation^21,39, daylight hours^25,46, solar radiation^21,31,35, and wind-speed^27,33,35,39. One primary hypothesis of these studies was that these climatic factors and the dispersion of COVID-19 are correlated. In an attempt to examine the hypothesis, statistical and geospatial analysis was used as a key analytical tool^25,31,39,44.

Studies analyzing the effects of temperature, humidity, and solar radiation found that an association exists between the transmission of the virus and humidity. These studies claimed that humidity is directly associated with the transmission³¹, while temperature has no relation to the viruses diffusion^35,37. However, another study refuted these findings and claimed both temperature and average daylight hours are influencing the spread of the virus²⁵. Other studies incorporated demographic variables as well as mobility and infections data with these meteorological factors to investigate how they are correlated with the COVID-19 outbreak³⁵. Instead of concentration mapping, they showed the spatial distribution and sensitivity of each factor in the study area map and asserted that population density and human mobility among the provinces are directly influencing the accelerated diffusion of COVID-19 cases.

Statistical and spatial analysis and modeling were used in other studies to identify the association between COVID-19 and climatic factors to understand the spatial distribution pattern of COVID-19 cases^26,44 and spatio-temporal prediction of the pandemic for a different period³⁶. These studies concluded that without adequate and effective control measures, there is no evidence of humidity, and summer weather substantially limiting the pandemic growth²⁶. Additionally, Pearson's correlation, generalized additive model, and regression models were also utilized in different studies to understand the influence of climate on the virus transmission. These works widely used GIS for mapping the observed and predicted COVID cases^30,33, mean temperature and humidity variation and correlation with COVID cases²⁷. None of these studies found any evidence of slower COVID-19 transmission with the change of temperature and humidity.

Different countries around the world have enforced different lockdown strategies since mid-January, even before WHO declared COVID-19 as a pandemic. As a result of these lockdowns, and lowered human and industrial activities, a substantial decline in air pollution was noticed^{23,24,28,29,34,40,41,43}. Several studies had reported a significant reduction in NO₂, CO_, and SO₂ concentration in the environment^{23,24,28,34,40}. Studies reported significant improvement in AQI³⁹ and significant reductions in particulate matter (PM) concentrations⁴¹. Similar studies at a global scale reported a significant decrease in NO₂ concentration and minor decrease in CO concentration, and aerosol optical depth²³.

Other studies identified a global reduction of environmental pollution by up to 30%, human mobility by up to 90%²⁴, and vessel activities by up to 69%²⁹. A study using satellite images to compare the pollutants concentration before, during, and after the Chinese Spring Festival observed that the usual trend of NO₂ and SO₂ concentrations decrease before the festival and increase afterwards was not noticed in 2020³⁴. ArcGIS was extensively used in several studies to quantify the association between AQI and the distribution of COVID-19 cases^32,33. These studies claimed that air quality is a core driver of COVID-19 dispersion around the world, and the dispersion enhances in a temperature ranged from 10 to 20°C. ArcGIS was also used to perform sensitivity analysis, and for the calculation of Global Moran's I and LISA to analyze the impact of PM concentration in the air on the fatality rate in China⁴⁵. The study concluded that the fatality rate is positively correlated with the pollutant concentrations in the air.

Socio-economic and cultural. In response to the coronavirus pandemic, alongside the booming clinical and public health research, social scientists are also retooling existing studies, methodologies, and data to understand the people's behavior responding to the pandemic and its impact on socio-economic and cultural settings around the globe. A total of 12 articles were identified that emphasized on socio-economic and cultural aspects in COVID-19 and used GIS^43–54. Researchers used gross domestic product (GDP), demographic and household compositions data, population density, and accessibility data⁴⁶. Several researchers utilized existing social vulnerability index data and applied geospatial tools and techniques to examine the spatial pattern of COVID-19 emergence across different socio-cultural settings^47–49. Studies developed dot density and choropleth maps of COVID-19 cases and reported that population density diseases along with poverty and unemployment rates are the major indicators associated with a higher COVID-19 mortality rate⁴⁹.

Different studies reported association among spatial distribution of the socioeconomic variables and the temporal progression of the pandemic using a regression model based spatio-temporal analysis. They noticed that per capita GIS and public transit access is closely related to COVID-19 incidence⁴⁶. Using a similar analytical tool, another study estimated the association between virus infection and social vulnerability considering county-level socioeconomic data, demographic composition, disability and minority status, language, and housing and transportation database. The study concluded that the increase of COVID-19 cases are highly associated with minority status and language⁴⁷. To understand the association between racial inequality and COVID-19 mortality, Kim and Bostwick (2020) utilized principal component analysis (PCA) and hotspot analysis in ArcGIS. The study reported that African American communities are the ones with the highest COVID-19 related deaths in the USA⁴⁸. Other studies also used ArcGIS and regression analysis to investigate the spatial patterns of the COVID-19 in relation to socio-economic variables. This study used GIS to map the spatial aspects and disparities between metropolitan and nonmetropolitan communities and the regression analyses to test the hypotheses of positive correlations between COVID-19 incidence and mortality rates and socio-economic factors in the GIS⁵⁰.

Proximity and hotspot analysis in GIS has become a widely used geospatial analysis technique in the social research to understand the feasibility of social distancing⁵¹, accessibility analysis of specific age group^52,53. Gibson and Rush (2020) calculated the distance to each dwelling's nearest neighbors to identify units that are unable to practice social distancing effectively. Dryhurst et al. (2020) developed a global risk perception index (R.P.I.) using a linear regression model and used GIS to plot the mean risk perception of COVID-19 in 10 countries⁵⁴. Sarkar (2020) used ArcGIS to reclassify the administrative units of Bangladesh based on COVID-19 susceptibility using multicriteria analysis based pairwise comparison⁵⁵. Cavalcante and Abreu (2020) applied Moran’s I and LISA to identify the type and degree of spatial clustering and scatter plots of socio-economic indicators⁵⁶. Similarly, Exploratory Spatial Data Analysis (ESDA) technique was used to identify spatial relationships between the density of built heritage resources and Airbnb listings. Based on the calculation and mapping, they concluded that the distribution of Airbnb listings has a certain degree of spatial autocorrelation⁵⁷.

Population health surveillance. From the first case in Wuhan to the global pandemic, an enormous number of public health studies have been conducted to help the policymakers to understand how best to manage the current and future public health responses. A total of nine articles were identified that emphasized public health issues and used GIS^50,55–62. Multiple studies developed a multicriteria decision making index to assess the risk and resilience of the existing healthcare system. Requia et al. (2020) employed GIS techniques to construct a geodatabase comprising land use, income, population, health condition, number of hospital beds, and staff at the municipality scale. They predicted a deficit of 17 beds in Brazilian municipalities⁵⁵. Similarly, Jovanović et al. (2020) developed a global index comprising of 57 indicators using ArcGIS-based network analysis for hospital accessibility and resilience mapping^58,59.

GIS was used to find out whether there is any association between orthopedic surgeons’ age and COVID-19 confirmed cases⁶⁰. The study reported a spatial relationship between the number of cases and number of surgeons in a state. In a similar study, Ruthberg et al. (2020) studied the potential risk of otolaryngologists above age 60 due to COVID-19, where they used a heat map to show the state-wise potential risk zone using QGIS. The study indicated that New York, New Jersey, Massachusetts, and Michigan were the riskiest zone according to the ratio of the number of confirmed cases to the number of total ENT's over 60⁶¹. Kuupiel et al. (2020) calculated the distance from the primary healthcare (PHC) clinic as well as to the nearest health facility in ArcGIS to measure geographical accessibility (in terms of distance and time) to COVID-19 specialized hospital facilities in Ghana. The analysis reported that the current mean travel time (more than an hour) and distance (more than 35 kilometers) to PHC is significantly higher than the globally accepted standards⁶². Similar accessibility analysis was done in Australia using proximity analysis and hotspot analysis to measure travel time to the closest hospital facility for aged population resulted in a similar outcome as well⁵³.

Ahmadi et al. (2020) developed an epidemic prediction model for spatial-temporal analysis to predict and estimate the number of patients and deaths at the end of pandemic (infected, cured, and dead cases). The study predicted that approximately 7900 and 4620 deaths would occur in Iran from May 13 to June 1, 2020, respectively, and then the curve will flatten⁶³. To analyze the epidemiology of COVID-19, studies also utilized georeferencing. These studies first geocoded all COVID-19 cases and then joined them to the county layers by administrative codes in ArcGIS and afterwards, applied LISA. This study reported that the spatial clustering is not random and shows significant spatial heterogeneity in China⁶⁴. A wastewater-based epidemiology (WBE) tool was proposed as a surveillance tool to monitor the COVID-19 pandemic⁶⁵. The study used multiple variables to run the GIS-based hydraulic model and network analysis using the SWMM modeling environment and ArcGIS. The result effectively served as a justification to use WBE as a rapid and efficient tool to track COVID-19, which the authors claimed could be used with clinical testing to save billions of dollars⁶⁵.

Spatial pattern analysis of COVID-19. A total of nine articles were identified that emphasized on identifying the spatial transmission pattern analysis of COVID-19^{14,16,19,62–67}. So far, only one study had used geospatial analysis to identify Spatio-temporal clusters and prediction modeling for COVID-19 transmission. They utilized the Poisson probability distribution model, Kernel density analysis, and space-time scanning analysis to identify high-risk Spatio-temporal clusters for transmission of COVID-19 in Brazil and detected active Spatio-temporal clusters comprising six municipalities in the south-central region of Brazil⁶⁶. Only two studies so far studied how travel restrictions may have limited the epidemic trajectory. One of them developed a global model based on internationally reported cases and mobility data, to project the impact of travel limitations on the national and international spread of the epidemic and revealed that Wuhan travel ban only hindered the overall epidemic trajectory by 3 to 5 days in other cities of China but had a significant influence on the international scale dispersion²². Studies also estimated the probability of COVID-19 cases transportation before January 23 among 369 cities in China and reported that 130 cities in China have more than 50% chance of having a COVID-19 case imported from Wuhan⁶⁷.

In a global study to identify geographic risks of COVID-19 transmission using online Rasch Modeling Algorithm, the authors reported that Iran, South Korea, Italy, Germany, Spain, China (Hubei), and France, are the top countries with higher outbreak potential respectively⁶⁸. Other studies utilized accessibility analysis techniques to assess the spatial diffusion of COVID-19, using GIS-based origin, destinations, and frequencies mapping of public transportation from Wuhan. The study claimed that increase of COVID-19 cases have a direct relationship with the frequency of public transport from Wuhan⁶⁹. Other studies utilized time-series analysis and plots to portray the spatial and temporal variation of COVID-19 cases in China and to elucidate the role of case importation in transmission in cities across China using spatial analysis⁷⁰.

One other global study used GIS to visualize the worldwide distribution of air transport passengers from Wuhan and infected traveler’s ratio around the world¹⁹. Geo-referencing of confirmed cases also played an interesting role in the spatial pattern analysis studies. These studies measured and identified the regions that have a high risk of transmission at an early stage⁷¹. Similarly, a generalized linear regression model was used to analyze the spread and control of COVID-19 cases using case reports, and human movement data. This study utilized spatial distribution mapping of the travel movements from Wuhan to each province and modeled the spatial dispersal pattern of COVID-19 trajectories with or without Wuhan travel ban. They found that Wuhan shutdown delayed arrival of COVID-19 in other cities by around three days¹⁸.

Computer-aided spatial and statistical analysis and modeling. A substantial number of studies were found to apply computer-aided spatial and statistical analysis and modeling techniques in analyzing distinct aspects of COVID-19. A total of 16 articles were identified that emphasized on computer-aided spatial and statistical analysis and modeling in COVID-19^69–84. Most of these works were focused on analyzing the spatial distribution pattern of COVID-19 cases using the confirmed cases data^72–78, or news reports of COVID-19 cases as proxy data⁷⁹. Most of the studies were focused on examining the spatial distribution and relationship between COVID-19 cases, deaths, and locations. Adekunle et al. (2020) examined the same relationship. They found a positive statistically significant relationship among spatial clusters, confirmed cases and potential deaths⁷². Another similar study in Hubei, China, demonstrated that the high-low cluster had no high-value incidence cluster where local Moran's I indicated that Hubei was the only province with High-Low aggregation⁷⁴. Moran's I was applied in analyzing the spatial and temporal distribution of cases in Hubei province by other studies as well⁷⁷.

Multiple studies examined the spatial and temporal distribution and modeled the trend of COVID-19 cases growth in ArcGIS and found that the highest risk place was those that had a high population inflow from Wuhan and Hubei province^75,76. Spatial modeling using Poisson space-time scan statistics to produce cluster mapping attempts also led to the proposal of first rapid surveillance to monitor the spread of COVID-19⁷³. Miller et al. (2020) demonstrated the worldwide spatial distribution of COVID-19 cases using the heat map technique in GIS The map reported that China, Italy, Iran, and Spain were the highest affected countries till March 17, 2020, which was also visible by the JHU and WHO reported data⁷⁸. Spatial panel data model used by Guliyev (2020) showed that the rate of deaths had a significant positive effect where the recovery rate had a negative with the confirmed COVID-19 cases⁸⁰. Irvine et al. (2020) used a SEIR model to estimate the transmission rate within Immigration and Customs Enforcement detention facilities in the GIS and the impacts on the I.C.U. capacity⁸¹.

Bai (2020) used two different models, including SEIRD model and Agent-Based Model (A.B.M.) to simulate the COVID-19 spread. They found that A.B.M. could be more effective and it also could be a useful tool to figure out new effective strategies⁸². In a similar work, Mollalo et al. (2020) compared five different models to develop a spatial model of COVID-19 incidence rate considering 35 variables using geospatial software’s. The results showed that MGWR could be a better model as it was able explain 68.1% of the total variation of COVID-19 incidences in the GIS⁸³. In a follow-up paper, the same authors used ANN to model the incidence rate of COVID-19 and used Moran's I index to create incidence hotspots. Out of included 57 variables, 10 variables found statistically significant in explaining the result⁸⁴. A similar global scale study revealed that age and population density have a statistically significant relationship with the spatial distribution pattern of COVID-19 cases⁸⁵.

Using Maxent based Ecological Niche Model, Ren et al. (2020) developed a potential risk zone map of China where population, public transportation demands, medical resources demands were used as explanatory variables. They suggested using this as an early forecasting model to predict the risk zone in China's other megacities⁸⁶. Kanga et al. (2020), on the other hand, provided a useful recommendation to local authorities in India by using proximity-based hot spot analysis to map the risk zones with relevant preventative measures to mitigate the COVID-19 crisis⁸⁷. Several other works were focused on suitability mapping with a focus to find a suitable location for the quarantine zone in Surat, India, using ArcGIS. The study revealed that the suitability analysis could help to control the spread as a prevention measure.

Underlying data

All data underlying the results are available as part of the article and no additional source data are required.

Extended data

Figshare: Application of geospatial techniques in COVID-19-related studies. https://doi.org/10.6084/m9.figshare.13229147.v2¹⁶.

This project contains a summary of the articles idenitifed in this study. This includes a detailed breakdown of what methods used in the reviewed article, along with the data type, spatial analysis tool/techniques and the findings from the analysis.

Reporting guidelines

Open Science Framework: PRISMA checklist for ‘Applications of GIS and geospatial analyses in COVID-19 research: A systematic review’. https://doi.org/10.17605/OSF.IO/ZGMP8⁹⁹.

Extended data are available under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0); the compelted PRISMA checklist is available under the terms of the Creative Commons Zero "No rights reserved" data waiver (CC0 1.0 Public domain dedication).