AI가 토네이도 이후 재난 복구를 혁신하는 방법

AI가 자연 재해 복구를 개선하는 방법

When natural catastrophes strike, the immediate need is to save the local inhabitants and restore infrastructure to a functional state, especially power, fresh water, and healthcare.

자연 재해가 발생하면, 즉각적인 필요는 지역 주민을 구하고 전력, 식수, 의료 등 인프라를 기능적인 상태로 복구하는 것입니다.

이는 기후 변화와 인구 증가로 인해 지난 10년 동안 이러한 사건에 대한 비용이 지속적으로 증가하면서 점점 더 큰 문제로 부각되고 있습니다.

출처: Spire

하지만 지역 사회의 실제 복구는 훨씬 더 오랜 시간이 걸리며, 피해 평가와 재건에는 수개월, 경우에 따라서는 수년이 필요합니다.

Often, this is delayed by the sheer overwhelming load of problems to identify and properly assess. Individual inspection of each building can take a lot of time, especially if insurance companies, emergency services, and other stakeholders are short on personnel.

“수동 현장 검사는 노동 집약적이며 시간이 많이 소요되어, 중요한 대응 노력을 지연시키는 경우가 많습니다.”

Abdullah Braik – 텍사스 A&M 대학 토목공학 박사과정 학생

This could now change thanks to AI technology. Two researchers at Texas A&M University have developed an AI system using remote sensing, deep learning, and restoration models to predict accurately in less than one hour tornado damage assessments and recovery.

This method could help better organize recovery efforts, and was published in Sustainable Cities and Society¹, under the title “Post-tornado automated building damage evaluation and recovery prediction by integrating remote sensing, deep learning, and restoration models”.

“우리 방법은 고해상도 감지 영상을 활용하고 딥러닝 알고리즘을 적용해 몇 시간 안에 피해 평가를 생성함으로써, 최초 대응자와 정책 입안자에게 즉시 실행 가능한 정보를 제공합니다.”

Abdullah Braik – 텍사스 A&M 대학 토목공학 박사과정 학생

토네이도 파괴: 미국에서 증가하는 위협

When unleashed, nature can be devastating for people and the cities they live in. One such example is tornadoes, a relatively regular occurrence in many parts of the world, including the so-called Tornado Alley in the USA.

출처: Weather and Radar

예를 들어, 2011년 봄에 미주리주 조플린은 EF5 토네이도로 인해 풍속이 200mph(321km/h)를 초과하는 것으로 추정되는 폭풍에 휩쓸렸습니다. 이 폭풍은 161명을 사망하게 하고 1,000명 이상이 부상했으며 약 8,000가구와 사업체가 손상·파괴되었습니다. 토네이도는 인구가 밀집된 남중부 지역을 마일 폭으로 가로질러 파편이 된 잔해를 남기고 20억 달러 이상의 피해를 남겼습니다.

이번 달에도 치명적인 토네이도가 중서부와 남부 지역의 많은 주택을 파괴하고 큰 피해를 입혔습니다.

It will likely take a lot of time to fully grasp the full scale of the damage these tornadoes caused. And this is where the Texas A&M University researchers think they can help.

토네이도 이후 피해 평가의 과제

After such an event, it is the standard procedure to have an array of first responders, city planners, and insurance experts move in to assess the destruction and how to react to it.

This is especially true with tornadoes, as the wind and heavy things lifted by them can destroy even the strongest buildings.

“Post-disaster damage surveys are typically time-consuming and labor-intensive, primarily focused on continuous model and code refinement rather than immediate and rapid updates.”

Some tentative steps have been made to use existing data to analyze the post-catastrophe damages, notably Geographic Information Systems (GIS), which bring together multiple layers of data about a given location.

But the way this data is used is often insufficient to provide accurate evaluations. It also requires a lot of manual interventions and human judgment to be turned into usable metrics. This is where adding extra information and AI can help.

AI 기반 손상 평가 및 복구 모델

기존 자원을 AI와 결합하기

The researchers combined 3 different tools together: remote sensing, deep learning, and restoration modeling.

To better evaluate the immediate damages and changes after the catastrophe, they used remote sensing like high-resolution satellite or aerial images.

“이러한 이미지는 영향을 받은 지역을 거시적으로 파악할 수 있게 해 주어, 빠르고 대규모의 손상 탐지를 가능하게 합니다.”

Abdullah Braik – 텍사스 A&M 대학 토목공학 박사과정 학생

Then, deep learning methods were used to automatically analyze these images to identify the severity of the damage accurately.

The AI was trained on thousands of images of previous disasters and learned to recognize visible signs of damage such as collapsed roofs, missing walls, and scattered debris. It then classifies every building into categories like no damage, moderate damage, major damage, or destroyed.

The last element is restoration modeling. One part is using data about infrastructure details and community factors, like income levels or access to resources.

Another part is using existing and tested recovery models to judge how long it might take for homes and neighborhoods to recover under different funding or policy conditions.

This completely changes how damage assessment is performed:

• Remote sensing gives an immediate overview of the entire situation.
• The AI model can analyze this data in less than an hour, compared to the months required for on-the-ground humans to perform the damage evaluation.
• Restoration modeling turns the AI evaluation into actionable metrics on which areas need the most help and what resources are required.

“궁극적으로, 이 연구는 신속한 재해 평가와 전략적 장기 복구 계획 사이의 격차를 메우며, 위험 기반이면서도 실용적인 프레임워크를 제공해 토네이도 이후 회복력을 강화합니다.”

Abdullah Braik – 텍사스 A&M 대학 토목공학 박사과정 학생

실제 토네이도 데이터를 통한 AI 모델 검증

To validate this approach, the researchers looked back at the 2011 Joplin tornado catastrophe.

This event was extensively documented, creating a rich dataset that could be used as a backtest for the AI system. The calculated assessments could then be compared to real-life, on-the-ground damage assessments done at the time.

And the AI-generated results proved remarkably close to the historical data. It also gave a record of how the catastrophe unfolded.

“가장 흥미로운 발견 중 하나는 고정밀도로 손상을 감지할 뿐만 아니라 토네이도의 이동 경로까지 추정할 수 있었다는 점입니다.

손상 데이터를 분석함으로써 토네이도의 경로를 재구성했으며, 이는 역사적 기록과 매우 일치하여 사건 자체에 대한 귀중한 정보를 제공했습니다.

Abdullah Braik – 텍사스 A&M 대학 토목공학 박사과정 학생

 다른 자연 재해로 AI 모델 확장

While it was developed and back-tested for tornado-caused damage, this method could be deployed for other situations, like hurricanes and earthquakes, as long as satellites can detect damage patterns.

This limitation could be less of a problem than initially expected, even if damages from earthquakes are, for example, less easily visible from the sky than blown-off roofs. This is because the model learns from real-life examples, and we often see AIs being able to detect patterns invisible to the human eye.

“모델의 일반화 가능성의 핵심은 특정 위험에 대한 과거 이미지를 활용하도록 훈련시키는 데 있으며, 이를 통해 각 사건에 고유한 손상 패턴을 학습할 수 있게 됩니다.”

Abdullah Braik – 텍사스 A&M 대학 토목공학 박사과정 학생

At the very least, it seems that the model will match well for another common disaster in the USA: hurricanes.

“우리는 이미 모델을 허리케인 데이터에 적용해 보았으며, 결과는 다른 위험에 적용할 수 있는 잠재력을 보여주었습니다.”

Abdullah Braik – 텍사스 A&M 대학 토목공학 박사과정 학생

This matches another application of AI in this field, with better hurricane prediction now becoming a reality, including with AIs like Graphcast,  Spire, and Climavision.

Another extension of this research could be to move beyond damage assessment. It could be used for creating real-time updates on recovery progress and tracking recovery over time.

This sort of automated, AI-driven feedback could then inform policy and optimize the rebuilding efforts.

위성 데이터 및 AI에 투자하기

Spire

Spire is a space data company that operates the world’s largest multi-purpose constellation of satellites in private hands.

The company strongly focuses on weather data, and its satellites can capture images in multiple spectra, giving a more data-rich picture of a given location.

For example, its satellite images of moisture measurement are precise up to 100m and can be used by farmers, but also insurance companies, commodity traders, environmental monitoring agencies, construction companies, and civil engineers to better understand soil condition and upcoming agricultural yields.

출처: Spire

The company offers to build for its client their own proprietary satellite constellation, with the LEMUR satellite platform.

출처: Spire

The company is active in security as well, notably with its aviation offer for Automatic Dependent Surveillance-Broadcast (ADS-B), which uses GPS to determine airspeed, location, and other information about aircraft.

Meanwhile, Spire has been selected for a $237M contract with the US Space Force to “to design, build, integrate, and operate small satellite buses for next-generation space experiments”.

It was also active in the maritime industry, with its satellites used for vessel tracking, but this branch of the company was acquired by Kpler in 2025년 4월.

Regarding AI, Spire is collaborating with NVIDIA (NVDA ) to integrate into  NVIDIA‘s Earth–2 Cloud APIs all of Spire’s Radio Occultation (RO) data and proprietary data assimilation (DA)

“Spire의 독점 데이터와 뛰어난 전 세계 날씨 커버리지를 NVIDIA의 최첨단 기술 및 전문성과 결합함으로써 날씨 예측 정확도를 크게 향상시킬 수 있습니다. 이 협업은 고객이 정보를 얻는 것을 넘어, 변화하는 기후 환경에 선제적으로 대응할 수 있도록 역량을 강화합니다.”

Michael Eilts – General manager of weather and climate at Spire

Overall, Spire is a data company embracing the conjunction of satellite imagery and AI for better weather forecasts, agriculture prediction, tracking of airplanes, and even defense purposes.

It is now reaching a scale where it could become profitable, an important potential turning point for investors, and maybe it will not need too much additional money raising after a successful $40M gross proceeds in selling new shares in Q1 2025.

출처: Spire

최신 Spire (SPIR) 주식 뉴스 및 개발

연구 참고:

^{1. Abdullah M. Braikand Maria Koliou. Post-tornado automated building damage evaluation and recovery prediction by integrating remote sensing, deep learning, and restoration models. Sustainable Cities and Society. Volume 123, 1 2025년 4월, 106286. https://doi.org/10.1016/j.scs.2025.106286}