• Stars
    star
    234
  • Rank 171,630 (Top 4 %)
  • Language
    R
  • License
    MIT License
  • Created over 1 year ago
  • Updated about 2 months ago

Reviews

There are no reviews yet. Be the first to send feedback to the community and the maintainers!

Repository Details

Source code and data for The Economist's Ukraine war-fire model

The Economist's Ukraine war-fire model

This repository contains the source code and data associated with The Economist’s Ukraine war-fire model. This model uses statistical techniques from machine learning and publicly available satellite data on temperature anomalies to detect war events. Between February 24th, 2022, and February 18th, 2023, we detected 14,068 such events in the country. We have since added many more.

See the latest data, updating several times daily, here: Tracking the Ukraine war: where is the latest fighting?

If you have any ideas for how we might improve this work or the presentation of it, email [email protected] or message @Sondreus. This data is not perfect - see section on limitations below.

In the following maps, red indicates that fires are classified as war-related. Size reflects approximate population density.

Events in the past week

Chart of fire events assessed as war-related

Events since Feb 24th, 2022

Chart of fire events assessed as war-related
Animation of fire events assessed as war-related

Events day-by-day, since Feb 24th, 2022

Animation of fire events assessed as war-related

Spotlight: Donetsk

Chart of fire events assessed as war-related
Chart of fire events assessed as war-related

Spotlight: Bakhmut

Chart of fire events assessed as war-related
Chart of fire events assessed as war-related
Chart of fire events assessed as war-related
Chart of fire events assessed as war-related

Events in the past month

Chart of fire events assessed as war-related

Events in the past week

Chart of fire events assessed as war-related

Charts of events per day

Chart of fire events assessed as war-related
Chart of fire events assessed as war-related
Chart of fire events assessed as war-related
Chart of fire events assessed as war-related

Scripts, sources and output data

Generating the model ensemble, as well as model outputs, maps, and analysis charts can be done by running the master script:

This script calls component scripts as well as auxiliary scripts. These produce, among other things, the following map, where each of the 14,068 dots is a fire event with a precise location (in most cases, to within 500m accuracy), a date, and a time of day.

Chart of fire events assessed as war-related

To update the model, run:

This automatically downloads the latest data, updates predictions, get exact population densities of events, runs the classifier to detect war activity, and updates all maps, charts, and animations.

Classification of events

All details on the procedure used to classify events are available in the above scripts. In brief, events are classified as war-related if they met the following thresholds:

  • Excess fire activity in a given 0.1 latitude by 0.1 longitude area of Ukraine on a given day is so large as to have less than 5% probability of occuring in a normal year.
  • Such excess is substantial: it has happened regularly in that area of the country, defined as twice within a single cell less than 50km away, at least 7 days apart and less than 180 days apart (i.e. two spaced-out events within a six-month period).
  • or: A fire event takes place in a cell less than 10 days after it has meet the above two requirements. (In other words, other activity producing fire events, typically from agriculture, does not resume until 10 days after an area has seen a war event.)

Limitations

Many war events do not produce heat at a level detectable by the satellite systems we use, and even if they do, events may go unrecorded because they happen under cloud cover, which regularly obscures much of the country from such satellite monitoring, or have cooled by the time the satellites pass overhead. This means not all war events are detected.

Moreover, our statistical method classifying events as war-related (or not war-related) is probabilistic. This means that it will sometimes categorise events which were unrelated to the war as war-related, and more frequently, given our strict thresholds, classify events related to the fighting as insufficiently abnormal to be labelled war-related.

Historical versions and sources

This model was first published on February 23th, 2023.

Acknowledgements

The Economist gratefully acknowledge helpful discussions with Daniel Arribas-Bel and Francisco Rowe of the University of Liverpool on the construction of training features and modelling approach.

If you have any suggestions or questions, please email [email protected] or open an issue.

Licence

This software is published by The Economist under the MIT licence. The data generated by The Economist are available under the Creative Commons Attribution 4.0 International License.

The data and files that we have generated from open sources are freely available for public use, as long as The Economist is cited as a source.

Suggested citation

The Economist and Solstad, Sondre (corresponding author), 2023. The Economist war-fire model. First published in the article "A hail of destruction", The Economist, February 25th issue, 2023.

More Repositories

1

big-mac-data

Data and methodology for the Big Mac index
Jupyter Notebook
1,581
star
2

us-potus-model

Code for a dynamic multilevel Bayesian model to predict US presidential elections. Written in R and Stan.
HTML
1,231
star
3

covid-19-excess-deaths-tracker

Source code and data for The Economist's covid-19 excess deaths tracker
R
641
star
4

covid-19-the-economist-global-excess-deaths-model

The Economist's model to estimate excess deaths to the covid-19 pandemic
R
459
star
5

graphic-detail-data

Data and code behind the Economist's Graphic Detail section.
Jupyter Notebook
398
star
6

the-economist-gdp-per-hour-estimates

Code and data for Graphic detail on GDP and productivity
R
81
star
7

2022-france-election-model

R
74
star
8

covid-19-risk-estimator

Diagnostic code mappings and replication code for the covid-19-risk-estimator
Jupyter Notebook
35
star
9

oecd_regional_inequality

Regional inequality data and analysis
34
star
10

normalcy-index-data

The Economist's "normalcy index" data repository
20
star
11

the-economist-conquests

Source code and data for "Vladimir Putin is dragging the world back to a bloodier time"
R
14
star
12

banana-index-data

Data and methodology for the Banana index
9
star
13

covid-19-italy-herd-immunity-GD

R
8
star
14

covid-19-us-seroprevalence-model

The Economist's seroprevalence model estimating American weekly covid-19 infections based on a combination of death by age and sex, IFR by age and sex, and confirmed cases.
R
8
star
15

fires-in-sudan

Code and data to analyse fire events in Sudan
R
7
star
16

2021-germany-election-model-PUBLIC

An election-forecasting model for the German federal parliament
R
7
star
17

Grim-Tallies

Replication data and code for "Briefing: Grim Tallies", including The Economist's estimate of global infections up to September 15, 2020.
R
6
star
18

dark

Chrome extension to read economist.com in dark mode
CSS
3
star
19

covid-19-age-adjusted-ifr

Estimated IFR based on country's population age distributions
R
3
star
20

nutrition

The Economist's modelling of nutritional outcomes, informed in part by demographic projections
R
3
star
21

the-economist-nationalism-and-corruption

R
3
star
22

us-midterms-2022-change-data

2
star
23

GraphicDetail_WagesAndPlaces

From the graphic detail 15.01.22
R
2
star
24

covid-19-wuhan-excess-deaths

Estimates of excess deaths in Wuhan for The Economist "Daily Chart"
R
2
star
25

britain-mrp-2024-estimates

Estimates of 2024 voting intention by constituency from our MRP model with We Think
2
star
26

GraphicDetail_GlobalGrain

Data and code for the Graphic Detail dated 25.06.22
R
1
star
27

moving_democrats

HTML
1
star
28

vbm-turnout-analysis

R
1
star
29

2022-iran-protests

Replication data for "Protest movements as deadly as Iran’s often end in revolution or civil war"
R
1
star
30

malaria-eradication-estimates

A repository containing The Economist's estimates of effects of a successful malaria eradication campaign.
R
1
star
31

HPV

R
1
star
32

covid-19-uk-seroprevalence-model

The Economist's seroprevalence model estimating weekly covid-19 infections in England and Wales based on a combination of death by age and sex, IFR by age and sex, and confirmed cases.
R
1
star