Spatial Data Analysis of Civilian Harm in Ukraine

This project aims to extract statistical insights and produce a meaningful cartographic visualization of civilian harm in Ukraine. The data for this project based on incidents in Ukraine that have resulted in potential civilian harm. All computations were made in the Jupyter notebook, and I really like Jupyter for its ability to represent your work without additional storytelling in the article.

The data for this project based on incidents in Ukraine that have resulted in potential civilian harm. This data began collection by the OSINT group Bellingcat on February 24, 2022, and intends to be a living document that will continue to be updated as long as the war persists. Therefore, this data is not an exhaustive list of civilian harm in Ukraine but rather a representation of all incidents that Bellingcat has collected and determined the exact locations.

Skip table of contents

Table of contents

Required modules

import pandas as pd
import geopandas as gpd
import json
import plotly.express as px
import plotly.graph_objects as go
import contextily as cx
import folium
from folium.plugins import HeatMap

Importing and exploring the data

Importing json file

fp = '../data/ukr-civharm-2022-05-24.json'

with open(fp, 'r') as file:
     data = json.loads(file.read())

Flatten nested JSON structures to pandas DataFrame

incidents_df = pd.json_normalize(data, record_path='filters', meta=['id', 'date', 'latitude', 'longitude', 'location', 'description'])

incidents_df.head(2)

Data munging

incidents_df = incidents_df[incidents_df['key'] == 'Type of area affected'].rename(columns={'value': 'area_type'}).reset_index()

columns = [
    'id',
    'date',
    'latitude',
    'longitude',
    'location',
    'area_type',
    'description'
]

incidents_df = incidents_df[columns]

incidents_df[["latitude", "longitude"]] = incidents_df[["latitude", "longitude"]].apply(pd.to_numeric)

incidents_df['date'] = pd.to_datetime(incidents_df['date'])

Exploring the data

incidents_df.head(2)

incidents_df.info()

<class ‘pandas.core.frame.DataFrame’>
RangeIndex: 456 entries, 0 to 455
Data columns (total 7 columns):
# Column Non-Null Count Dtype
— ------ -------------- -----
0 id 456 non-null object
1 date 456 non-null datetime64[ns]
2 latitude 456 non-null float64
3 longitude 456 non-null float64
4 location 456 non-null object
5 area_type 456 non-null object
6 description 456 non-null object
dtypes: datetime64ns, float64(2), object(4)
memory usage: 25.1+ KB

Statistical analysis

Analysing civilian harm by type of affected area

incidents_by_type = incidents_df.groupby('area_type').size()

incidents_by_type = incidents_by_type.to_frame(name='n_events').reset_index()

incidents_by_type.sort_values(by='n_events', ascending=False)

fig = px.pie(incidents_by_type, values='n_events', names='area_type', title='Civilian harm by type of affected area',
             hover_data=['n_events'], labels={'n_events':'Number of incidents'})

fig.update_traces(textposition='inside', textinfo='percent+label')
fig.update_layout(uniformtext_minsize=12, uniformtext_mode='hide', width=800, height=600)

fig.show()

Analysing civilian harm by date

incidents_by_day = incidents_df.groupby('date').size().to_frame(name='n_events').reset_index()

fig = px.line(incidents_by_day, x='date', y='n_events', title='Number of incidents per day', labels={'n_events':'Number of incidents'})

fig.show()

fig = px.bar(incidents_by_day, x='date', y='n_events', color='n_events')

fig.show()

Geospatial dimension

Converting to a geospatial dataset

geo_incidents_df = gpd.GeoDataFrame(incidents_df, geometry=gpd.points_from_xy(incidents_df['longitude'], incidents_df['latitude']), crs="EPSG:4326")

geo_incidents_df.to_crs(epsg=3857, inplace=True)

ax = geo_incidents_df.plot(figsize=(20,10))
cx.add_basemap(ax)

From the map above, we can define that few points are outside Ukraine.

Combining with regions of Ukraine

I downloaded the publicly available regions reference from humdata.org (Ukraine - Subnational Administrative Boundaries), and added the Shapefile with the borders to the project repo: data/ukr-adm-sspe-20220131-zip-1/ukr_admbnda_adm1_sspe_20220114.shp.

regions = gpd.read_file(r'../data/ukr-adm-sspe-20220131-zip-1/ukr_admbnda_adm1_sspe_20220114.shp')

regions.head(3)

regions.to_crs(epsg=3857, inplace=True)

ax = regions.plot(figsize=(20,10))
cx.add_basemap(ax)

geo_incidents_df.crs == regions.crs

True

combined_geo_incidents_df = gpd.sjoin(geo_incidents_df, regions[['ADM1_EN', 'geometry']], predicate='within').drop(columns=['index_right'])

combined_geo_incidents_df.head(3)

Heatmap visualization

m = folium.Map(location=[49.107892273527504, 31.444630060047018], tiles = 'stamentoner', zoom_start=6, control_scale=True)

heat_data = list(zip(combined_geo_incidents_df["latitude"], combined_geo_incidents_df["longitude"]))

HeatMap(heat_data).add_to(m)

m

Analysing civilian harm by regions

incidents_by_region = combined_geo_incidents_df.groupby('ADM1_EN').size().to_frame(name='n_events').reset_index()

incidents_by_region.head(2)

fig = px.pie(incidents_by_region, values='n_events', names='ADM1_EN', title='Civilian harm by region',
             hover_data=['n_events'], labels={'n_events':'Number of incidents', 'ADM1_EN': 'Region'})

fig.update_traces(textposition='inside', textinfo='percent+label')
fig.update_layout(uniformtext_minsize=12, uniformtext_mode='hide', width=800, height=600)

fig.show()

Choropleth classification

regions_incidents = pd.merge(regions[['ADM1_EN', 'geometry']], incidents_by_region, how='left', on='ADM1_EN')

regions_incidents['n_events'] = regions_incidents['n_events'].fillna(0)

regions_incidents.head(3)

regions_incidents['geoid'] = regions_incidents.index.astype(str)

m = folium.Map(location=[49.107892273527504, 31.444630060047018], tiles='cartodbpositron', zoom_start=6, control_scale=True)

folium.Choropleth(geo_data=regions_incidents,
                  data=regions_incidents,
                  name='Choropleth Map',
                  columns=['geoid','n_events'],
                  key_on='feature.id',
                  bins=8,
                  fill_color='YlOrRd',
                  line_color='white',
                  line_weight=0,
                  legend_name='Number of incidents').add_to(m)

folium.features.GeoJson(regions_incidents, name='Labels',
               style_function=lambda x: {'color':'transparent','fillColor':'transparent','weight':0},
                tooltip=folium.features.GeoJsonTooltip(fields=['ADM1_EN', 'n_events'],
                                              aliases = ['Region', 'Number of incidents'],
                                              labels=True,
                                              sticky=True)).add_to(m)

folium.LayerControl().add_to(m)

m