Satellite Imagery Unveils Hidden War Impact

Is it just a false impression, or have armed conflicts really increased?

Valerie Sticher: The impression is not misleading. We can see that there are more armed conflicts worldwide than at any time since the end of the Cold War in 1989. In addition, violence against civilians is increasing.

Casualty figures are often seen as the main yardstick for the scale of a conflict. Why is that not enough, in your view?

Casualty figures are a very important indicator because loss of life is among the most severe consequences of war. But violent conflicts have many other consequences - for example, people are displaced, or their livelihoods are destroyed.

Data on casualty figures is based on textual data, primarily from media reports. However, this type of data often has gaps when it comes to incorporating other forms of violence into the analysis of a conflict.

You suggest making greater use of satellite data. What can this data reveal?

Satellite data from conflict regions has been in use for some time. What is new, however, is that we now have the ability to analyse satellite data automatically and thus collect data systematically. In our study, we combined text-based and satellite-derived data from two regions - Ukraine and Myanmar - and examined how conflict dynamics evolve spatially and over time. This sheds light on where different types of data have their strengths and weaknesses, and how we might combine them effectively.

Could you give an example of this?

In Myanmar, text-based sources indicate that the worst massacres against the Rohingya occurred during the first week. Satellite-derived damage data, however, shows that violence continued for months afterwards. This changes how we understand the dynamics of the conflict.

And what insights have you gained regarding the war in Ukraine?

War damage occurs far more often during Russian territorial gains than during Ukrainian recaptures. This is not surprising. Ukraine has an interest in destroying as little land, infrastructure and buildings as possible on its own territory. However, by combining satellite and text-based data, we were able to clearly demonstrate this pattern.

Where does satellite data reach its limits?

Satellite data can only capture events that leave a physical footprint - in other words, only things that are visible from space. We see burned-out houses, but not who destroyed them, or whether people had already fled or were driven out by the attack. We therefore advocate not viewing satellite data in isolation, but combining it with other existing data wherever possible.

In our study, we looked at two large Ukrainian cities that had suffered a similar degree of destruction. However, the casualty figures were very different because, in one of the two cities, a large proportion of the residents had fled before the Russian offensive began. Anyone looking solely at satellite-derived data would see two cities that appear similarly devastated. Anyone looking solely at casualty figures, however, would see two cities affected in completely different ways. Combining both sets of data provides a more complete picture.

The specific datasets used to train the models for the automatic analysis of satellite data also play a major role.

In what way?

It is problematic if models are trained on too little or too one-sided training data. To date, much of the reference data has come from a handful of high-profile conflicts such as the one in Ukraine, which is why the models work well in these cases. However, many conflicts take place in regions where, for example, the built environment is different, or where houses are burnt down rather than destroyed by heavy weapons. To ensure that automated analysis works reliably in these situations too, we need reference data from these types of conflicts.

In your opinion, where is there a lack of data needed to assess conflicts more effectively?

Above all for conflict-related sexual violence. It is inadequately represented in text-based data, and satellite data is of no help here. As a result, we almost certainly underestimate this form of violence.

How could your approach help aid organisations such as the ICRC or the UN? After all, ETH Zurich also collaborates with both of them.

The main aim of our study is to demonstrate how we can combine different types of conflict data to gain a better understanding of conflicts and to target humanitarian aid more effectively. But these organisations also need specific tools, and above all tools that can be deployed in a wide variety of conflict regions.

What are your hopes for the future?

We worked closely with computer scientists and humanitarian actors for this study. I hope that we can further expand this interdisciplinary collaboration. My greatest hope is that our research will ultimately help to better protect civilians in conflicts.