This repository inlcudes all the tools developped and used in the article The geography of COVID-19 pandemic: a data-driven approach to exploring geographical driving forces (currently in peer-review). The repository contains notebooks and scripts for animation of data, spatial autocorrelation, emerging hotspot analysis and machine learning prediction.
The shapefile AI_Covid_NUTS-Regions.shp is made to match the regions that the Europeans regions use to report Covid-19 cases to the ECDC, these should be NUTS-2 regions but it varies from country to country.
The shapefile contains all the variables used in the modelling:
Total infected (dependent variable), Population Density, Restaurants per cap., Cafes per cap., Fast Food Places per cap., Pubs per cap., Bars per cap., Gas Stations per cap., Schools per cap., Doctors’ Offices per cap., Kinder gardens per cap., Annual mean temperature, Nitrogen dioxide yoy change, Particulate matter < 2.5 µm (PM2.5) yoy change, Non-methane VOCs yoy change, Particulate matter < 10 µm (PM10) yoy change.
Visualizing Covid-19 infection rates over time can be done in a single animation. The animation is made by loading a CSV file as a pandas dataframe and plotting each column with a world shapefile, the maps can then be combined in to a single gif.
Infection data for the world animation is the daily number of new reported COVID-19 cases and deaths worldwide from the European Centre for Disease Prevention and Control
Similar animations can likewise be made for regions in the European union or municipalities on a country level.
 |
 |
|---|
Bi-weekly infection rates for the European Regions are from the Data on the weekly subnational 14-day notification rate of new COVID-19 cases from the ECDC.
Data on daily municipal infection rates in Denmark are from Statens Serum Institut (SSI)
To ensure a fixed legend a dummy feature is included in the data and in the shapefile outside the mapview. Find shapefiles and CSV-files in the Data folder, the animation scripts are located in the Notebook/Animations folder.
The spatial autocorrelation analysis checks for spatial distribution of the infection rates in the included regions. The tool locates hot- and coldspots of infection rates, together with doughnuts and diamonds - being regions with low infection rates sorrounded by high infection rates and vice versa.
 |
 |
|---|
The spatial autocorrelation analysis can be done with accumulated infection rates with the purpose of locating the clusters of the pandemic as a whole. To visualize how the patterns of clusters has changed over time, all the individual time steps can be visualized in a single animations.
The spatial autocorrelation notebook contains tools to carry out the analysis. Under Notebooks/Animations is the script for making a hot- and coldspot animation.
The emerging hotspot analysis, inspired by ESRI , is a method that utilize a three dimensional data set to locate clusters through both time and space. By doing a spatial autocorrelation analysis for each time slot of the data and adding these as layers on top of each other, it is possible to analyze how hot and coldspots change in space over time. The method presented in this repository is a simplified version of the ESRI tool and runs in Python alone and outputs a shapefile that can be visualized in QGIS, find the script and QGIS-styling file in the Notebooks folder.
The emerging hotspot analysis can be implemted over different timespans, as long as a reasonable number of time steps are included. The tools is very usefull in identifying the different categories of hot and coldspots and hereby also locate emerging hot and coldspots, this could for example be done by looking at daily infection rates one month back or for a specific timeframe.
 |
 |
|---|
A notebook for Geographic Weighted Regression is located in the Notebook folder. This notebook inputs a shapefile with infection numbers and a set of variables and outputs GWR statistics together with maps of variable significance, coefficients and GWR evaluation metrics (Local R2, Standardized Residuals and Condition Number)
The machine learning notebook contains ML-frameworks for predicting Covid-19 infection rates based on the attribute-values of the region, 3 algorithms are included in the notebook: Random Forrest Regressor, Lasso Regression and Support Vector Regression. The predictors outputs a plot showing how well the test regions have been predicted, the accuracy metrics: Accuracy, R2 and MAE, and variable importance for the random forrest regression.
The notebook also include tools for simple data inspection together with a complete section for carrying out and visualizing k-means clustering of the data.
All feedbacks are appreciated, feel free to reach out on LinkedIn or Email
This work is licensed under the GNU General Public License v3.0