2026 Kumamoto Earthquake — Multimodal Damage Analysis

Hello. My name is Erick Mas, from the Disaster Geoinformatics Laboratory.

This is our laboratory team.

On behalf of the team, I would like to present part of the analysis we are currently carrying out.

This analysis is carried out as part of "Building a Smart Disaster Prevention Network", under the Cabinet Office's Cross-ministerial Strategic Innovation Promotion Program (SIP).

Let us look at this event from three perspectives.

The first is human movement. Here I want to discuss how many people were present at a location, and where they went afterwards.

Next, we carried out an analysis with the help of news media and AI, asking what was reported to have happened, and where.

Finally, analysis of satellite radar imagery, aimed at detecting change to buildings, bridges and similar structures.

Since the presentation time is short, reading the "Point" line on each slide is enough to follow the flow.

Let me explain the data used for the first topic.

Two datasets were available for human mobility.

First, NTT Docomo's Mobile Spatial Statistics. Through the collaboration between Tohoku University and NTT Docomo, data from 2016 onwards is available; today we used the data from before and after the earthquake.

We also carried out a joint analysis with GeoTechnologies Inc., using population estimated from their mobile application.

For seismic intensity, we used the QUIET+ system, which can supply rapid estimates.

We focused on the areas where the shaking was strongest, in particular Yatsushiro City and Uki City.

From the left: the population present at 15:00 before the earthquake, the population present at 16:00 when it struck, the peak ground acceleration, and the impact index combining the two.

The first thing to notice is that the two panels on the left look almost identical. This is not a failure of the figure — it is the result.

The point of this analysis is…

Next we stack up the number of people by strength of shaking.

125,000 people, 93 percent, at intensity 6-lower or above. 59,000 people, 44 percent, at 6-upper or above.

And 33,000 people — roughly one in four — were in cells equivalent to intensity 7.

Of those, 27,000 were in Uki City.

To repeat: this is the number of people exposed to shaking. It is not a count of casualties or of injured people.

Intensity here is an estimate, so I would ask you to avoid saying that intensity 7 was "observed".

The last analysis in the first topic is a joint analysis with GeoTechnologies on where people went.

We examined mobility change at four specific locations; today I will show the result for Yumetown.

In this figure, the red squares are locations with many people before the earthquake.

Green marks locations with many people after the earthquake.

Olive marks locations where the change was small.

What we found is that the number of people fell considerably on the day after the earthquake.

Reading mobility data in this way lets us understand the situation in an area.

For the second topic, we collected news data from the web and used a large language model to classify it, and to find the location of the places mentioned in the articles.

This lets us understand spatially what happened where.

The results are displayed like this. If you scan this QR code, you can open the map application.

Finally, we also analysed change before and after the earthquake using satellite radar imagery.

The results can be viewed freely through the same QR code.

By compositing PALSAR-2 images from before and after the earthquake, we can identify locations with large change.

Some of it is visible to the eye, but there are also machine-learning methods for identifying areas of major change.

Looking at another image — this time Sentinel-1 — coherence change analysis lets us identify places where damage is likely.

The two panels on the right are zoomed in around the Aeon Mall. Comparing our group's result with Prof. Matsuoka's result from Institute of Science Tokyo, we confirmed that the two are consistent.

In summary, this three-part multimodal analysis gives us the number of people exposed to shaking and their movement, together with the reported damage and the change captured by radar.