Automated satellite analysis widens the measure of war beyond the fatality count
A Nature analysis led by Valerie Sticher argues that death tolls drawn from media text capture only part of a war, and its own Ukrainian case shows that damage seen from orbit misleads just as badly on its own.
The Scientist · Science desk
What happened
- A Nature study led by Valerie Sticher proposes three named ways of joining media-derived conflict event data to satellite-derived data: improvement, enrichment and fusion, tested on Ukraine and Myanmar.
- In Myanmar, text sources place the worst massacres against the Rohingya in the first week, while satellite damage data shows the violence continuing for months after that.
- In Ukraine, the combined data shows war damage occurring far more often during Russian territorial gains than during Ukrainian recaptures of the same ground.
- The study's starting point is that there are more armed conflicts worldwide than at any point since the Cold War ended in 1989, with violence against civilians rising.
Why it matters
- What's now possible capability Automated analysis makes satellite data collectable systematically rather than case by case, which is what turns destruction into a measurable variable across whole conflicts and long periods.
- What it limits constraint Anyone tempted to use damage maps as a casualty proxy inherits a hard limit: imagery cannot say who destroyed a building, or whether the people who lived in it had already left.
- Who has to decide decision Organisations that rank conflict severity by death toll alone now have a published alternative structure, and have to decide whether a broader measure is worth its added ambiguity.
- Who's exposed exposure Conflict-related sexual violence gains nothing from either data stream, so the most under-measured category of harm stays under-measured after integration.
Next story · Science · 5h ago
Baker lab's from-scratch zinc proteases cut their chosen target in 36% of 135 designsConfirmed2 publishers
The two kinds of data fail for different, unconnected reasons, and that difference is central to the argument. Casualty figures come from text, mostly media reports [5], and media attention is uneven: coverage thins in places and periods that draw few reporters, and thins again for whole categories of event that go unreported or are reported only in unspecific terms, among them the destruction of dwellings and crops and the displacement of populations [8].
Satellite-derived data breaks the other way, and the paper's Ukrainian city pair shows how. Two large cities had suffered a similar degree of destruction but recorded very different casualty figures, because a large proportion of one city's residents had fled before the Russian offensive began [12]. The imagery alone makes those two places look alike, but the death tolls alone make them look like different wars. Sticher's own conclusion is that satellite data should not be read in isolation but combined with other data wherever possible [17], which is a narrower claim than the one usually attached to automated remote sensing: it does not close the reporting gap by itself, it merely fails somewhere else.
The training data complicates the promise further. Much of the reference data used to teach these models what damage looks like comes from a handful of high-profile conflicts, Ukraine among them, which is why the models perform well there; many other conflicts unfold in places where the built environment is different, or where houses are burned rather than flattened by heavy weapons [13]. Media reporting concentrates its attention in much the same small set of wars [8]. So the automation is currently strongest where the text record is already thickest, and weakest in exactly the under-reported settings that motivated the exercise. The paper treats this as fixable, and names the fix directly: reference data drawn from those under-represented conflicts [13].
On the denominator, the paper names Gaza, Ukraine, Sudan and Myanmar as the conflicts where heavy weaponry, drone warfare and arson have caused widespread death, destruction and displacement [15]. Its empirical demonstrations cover two of those four [18]. Two case studies are enough to show that integration is feasible and that it surfaces patterns neither dataset produced alone, which is what the authors claim for it: complementary strengths and novel analytical insight [16]. What remains untested is how accurate an integrated severity measure would be in a conflict that has not yet had reference data labelled for it.
Media-based event datasets have been the field's main instrument for about fifteen years [7]. What the study offers is a second instrument whose blind spots do not overlap with the first one's, plus three named ways of joining them [2]. Read that way it is a real methodological gain, and the Ukrainian city pair is the caution attached to it: destruction measured from space is a poor stand-in for harm to people, in either direction.
What to watch
- Publication of labelled reference data from conflicts where dwellings are burned rather than shelled, which is the stated precondition for the models working outside Ukraine.
- Any validation study reporting error rates for automated damage detection in a conflict outside the training set, rather than analytical insight from case studies.
- Whether the established media-based conflict event datasets add destruction or displacement indicators alongside their fatality counts.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence58
- Adoption
- Not scored
- Hype gap+15
- Incentives
- Not scored
- Confidence55
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
A new study published in the journal Nature examines how different types of data can be combined for a fuller understanding of armed conflicts; the lead author is Valerie Sticher, who uses Ukraine and Myanmar as examples.
- [2]
The paper proposes three concrete approaches to integrating text-based conflict event data with satellite-derived data: improvement, enrichment and fusion, illustrated with case studies from Ukraine and Myanmar.
- [3]
The current paradigm in conflict research relies heavily on text-based data, using fatalities as the primary and often sole proxy for violence intensity.
- [4]
There are more armed conflicts worldwide than at any time since the end of the Cold War in 1989, and violence against civilians is increasing.
- [5]
Data on casualty figures is based on textual data, primarily from media reports, and this type of data often has gaps when incorporating other forms of violence into conflict analysis.
- [6]
Satellite data from conflict regions has been in use for some time, but what is new is the ability to analyse satellite data automatically and thereby collect data systematically.
- [7]
Over the past 15 years, conflict research has advanced with the emergence of detailed conflict event datasets based primarily on media sources.
- [8]
Because media reporting is attention-driven, there is a lack of reliable data from places and periods that receive little media coverage, and from types of event that often go unreported or are reported only in unspecific terms, such as the destruction of dwellings and crops or the displacement of populations.
- [9]
In Myanmar, text-based sources indicate that the worst massacres against the Rohingya occurred during the first week, while satellite-derived damage data shows that violence continued for months afterward.
- [10]
In Ukraine, war damage occurs far more often during Russian territorial gains than during Ukrainian recaptures, consistent with Ukraine's interest in destroying as little land, infrastructure and buildings as possible on its own territory.
- [11]
Satellite data can only capture events that leave a physical footprint; burned-out houses are visible, but not who destroyed them or whether people had already fled or were driven out by the attack.
- [12]
The study looked at two large Ukrainian cities that had suffered a similar degree of destruction but had very different casualty figures, because in one of the two a large proportion of residents had fled before the Russian offensive began.
- [13]
Much of the reference data used to train models for automated satellite analysis has come from a handful of high-profile conflicts such as Ukraine, which is why the models work well in those cases; many conflicts take place where the built environment is different or where houses are burned down rather than destroyed by heavy weapons, so reference data from those conflicts is needed.
- [14]
Sticher says there is above all a lack of data on conflict-related sexual violence: it is inadequately represented in text-based data and satellite data is of no help, so it is almost certainly underestimated.
- [15]
The paper names Gaza, Ukraine, Sudan and Myanmar as conflicts in which the use of heavy weaponry, drone warfare and arson has caused widespread death, destruction and displacement.
- [16]
The authors state that the different data generation processes of text-based and satellite-derived data create distinct, complementary strengths, and that integration yields novel analytical insights.
- [17]
The authors advocate not viewing satellite data in isolation but combining it with other existing data wherever possible.
- [18]
The paper's empirical demonstrations cover two of the four conflicts it names as examples of heavy-weapon, drone and arson violence.
Sources
2 independent publishers whose own reporting we read for this story.
- nature.comAdvancing conflict research and response through satellite-derived data
1 article · September 8, 2026
- phys.orgQ&A: Satellite imagery reveals what casualty figures alone do not tell us
1 article · September 9, 2026
Topics and entities
Topics
- Satellite damage assessment
- Humanitarian response
- Conflict Event Data
Entities
- ETH Zürich
- Nature
- International Committee of the Red Cross