2026 Kumamoto Earthquake — Disaster Impacts Analysed Through News Reporting

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).

Background: news and information about conditions in the affected area arrive in fragments, and neither the progression of damage nor the state of the response is linked across separate reports.

Immediately after the earthquake, coverage concentrates on visible direct damage — collapsed buildings, severed infrastructure.

Weeks later, an invisible chain of consequences surfaces: economic loss, health impacts — effects produced by effects, that is, secondary and tertiary damage.

The obstacle is that manual analysis cannot keep pace. Information fragments as time passes, and causal relationships spanning several weeks become difficult to trace.

The proposed approach is to extract causal chains from the reporting itself, automatically linking the long-term connections that tend to be overlooked.

The chart shows daily article counts over the month after the earthquake: a peak of 1,169 articles on the second day, then a steady decline, with weekends shaded.

A local large language model converts unstructured news articles into structured database entries — one article to many records — and generates links between related disaster impacts.

Records are then matched across the whole corpus, beyond the boundaries of a single article, to extract both short- and long-term causal chains.

Each record is geocoded using the geographic named entities found in the text, with Esri Japan and GeoNLP.

On the remote-sensing side, the aim is to map damage that text alone cannot locate — housing damage leading to large-scale evacuation, or road damage restricting the inflow of supplies and hitting tourism. Geolocated news for an area is then used to check the damage-mapping results.

Localisation rate: 83% of the extracted and analysed causal-chain records could be placed on the map as spatial features.

Example 1 — a deployment request from Kumamoto Prefecture led to the formation of a Nagasaki Prefectural Police special security unit, which completed its burglary-prevention assignment (1 source).

Example 2 — water facilities in Mashiki were damaged and the supply was cut, prompting water trucks to be sent to Mashiki from Saga, Miyazaki and Oita (3 sources).

Example 3 — Kyushu Shinkansen suspensions and road closures led to accommodation cancellations, with Kagoshima reporting a very large impact from the transport cut (3 sources).

Arrows on the map show causal relationships crossing prefectural boundaries; the hatched area marks the water and power outage zone in Mashiki.

Key message — dynamic relationship building: the method operates across successive news releases in time order and can generate new links dynamically, building causal links between the latest reports and earlier ones so that a developing situation can be monitored.

Medical chain: Uki General Hospital suspended patient intake on 28 July for lack of supplies; on 29 July, 26 medical facilities in Kumamoto reported damage, and in Yatsushiro 11 facilities lost water, 2 lost power and 5 had trouble with medical gas supply; out-of-prefecture DMAT deployment was requested and DWAT placed on standby, with DMAT support arriving on 31 July.

Heat chain: the power cut left households in Uki without air conditioning from 30 July, and by 31 July heatstroke cases were rising at Uki General Hospital; broken shelter air conditioning at the Ogawa disaster-response base was still unresolved on 30 August.

Transport-to-economy chain: overhead line breaks, rail distortion and fracture, sound-barrier damage and girder displacement on the Kyushu Shinkansen on 29 July; no restart date as of 31 July; cancellations and cancellation fees for tourists and accommodation providers across Saga and Kagoshima by 3 August; and economic loss plus reputational damage by 12 August, reaching areas with little direct damage such as Aso and Hitoyoshi.

Rapid discovery of causal relationships: links between post-disaster events are extracted as soon as they are reported.

Quantifying the time course: the time it takes for an impact to surface and then propagate can be measured.

Finding — secondary impacts that surface late in the recovery process. Several secondary impacts on society only became visible after time had passed:

Economic loss in tourism, surfacing in the third week after the earthquake, spreading even to businesses that suffered no direct physical damage.

A rise in petty crime in the affected area, exceeding the response capacity of local police and public-order arrangements.

Strain on the medical system, with shortages of medical supplies, rising heatstroke cases and damaged facilities compounding to place a continuously increasing load on the system.