Explore examples of environmental harm in Ukraine

CEOBS has been remotely tracking and assessing environmentally-relevant incidents in Ukraine since February 2022. We share our data with relevant stakeholders and it also informs our research and advocacy activities. This interactive map features 25 incidents from our database that help illustrate some of the types of environmental damage that have been caused or exacerbated by the conflict.

You can learn more about our Ukraine database and the assessment methodology for the case studies below.

CEOBS’ database incidents
Facility Type

Each of the case studies features a timeline of key incidents, and satellite and social media data has been used to undertake a remote assessment of the damage and the potential risks to the environment. It is important to note that remote analyses have their limitations and that ground assessments will be needed at the sites to validate the extent of any potential risks identified. Although the risks are small, recent imagery and analysis could influence military decisions. For these reasons of operational security we only publish information on this map with a delay, the cut-off point is currently the end of September 2023.

Learn more about the data:

The CEOBS database

Following the 24th February 2022 it became clear that the scale, intensity and character of the conflict in Ukraine was generating incidents of environmental harm that posed risks to people and ecosystems, and which would impede Ukraine’s recovery if left unaddressed. The types of harm and volume of available open source intelligence informed our development of a database framework that aims to capture the breadth and depth of environmentally-relevant incidents, with a systematic and detailed characterisation of the potential environmental harm. The data we continue to collect and analyse has a number of end uses, which include prioritising site assessments, and informing remedial activities and policy development.

Identifying incidents

An incident is an instance discrete in time and space that has resulted in or may result in pollution, environmental health impacts, a pause/stop of ecosystem/environmental services, or which may contribute to climate or biodiversity breakdown or a combination of any or all of these criteria. We cast the net wide to identify potential incidents meeting this definition. The primary sources are searches of social and traditional media, including manual daily inspections of key feeds and ad-hoc keyword searches. Incidents have also been added from existing datasets, principally from the Ecodozor platform, from reports and academic papers and from examination of satellite data.

Documenting incidents

There are two principal components to the database. Level 1 data is a long-list of incidents we can identify, together with basic information and verification. The most environmentally risky incidents are progressed to a Level 2 analysis, in which a detailed and systematic search is undertaken to gather and archive all available remote information. Satellite imagery and data is inspected, including from open sources, radar damage maps and proprietary high resolution imagery. We also perform a deep multi-language screening of several social media platforms, which include local chat groups.

The detailed documentation helps us to verify, chrono-locate and geo-locate an incident, and provide historical context and information on subsequent impacts in the months and years after. This rich information is used internally for environmental assessment and reporting; but is also useful for a range of stakeholders.

Assessing environmental harm

For Level 1 data we calculate environmental risk based on two parameters: i) the theoretical environmental harm score, which is calculated automatically based on the profile of the facility and its proximity to environmental media; and ii) the event magnitude score, which is a manual assessment of the severity of the impacts by our research team. The detailed information in the Level 2 documentation then allows us to calculate an ‘actual’ harm score, based on a standardised questionnaire that we developed based on the type of information that is available remotely. The questions are based on four conceptual components of environmental harm: contaminant source or hazard; environmental media; loss of ecosystem or provisioning service; and vulnerability. Where there are multiple incidents at the same facility, these are added together so the total harm can be compared across facilities.

It should be noted that the CEOBS database is only part of the story. A complementary way to track environmental harm via discrete incidents is to take what we term a ‘continuous cohort based’ approach, where objects or impacts are self-similar and are treated as more homogeneous. Several groups, in particular academics, are working on these approaches, which include crater detection, crop losses, burned area detection and locating explosions.

Contributors: Dr Anna McKean, Dr Eoghan Darbyshire, Iryna Babanina, Jay Lindle, Jonathan Walsh, Dr Linas Svolkinas and Rob Watson.

Funded by