AI Forecasting Pushes Disaster Work Into the Data Center

Researchers in Switzerland have copied around 100 petabytes of public NASA climate and Earth-observation data onto servers next to one of the world’s most powerful supercomputers, according to reporting by Phys.org. Earth-observation data means information gathered from satellites and other sensors that watch the planet’s land, oceans, ice, and atmosphere. The goal is to train artificial intelligence, or AI, which is software built to find patterns in data and use those patterns to make predictions. In this case, the predictions could improve weather forecasts and help identify warning signs before some natural hazards become disasters.

The project is based at the Federal Institute of Technology Zurich, known as ETH, and uses a supercomputer called Alps at the Swiss National Supercomputing Centre in Lugano. A supercomputer is not just a fast desktop; it is a large system built from many connected processors that can run huge calculations at the same time. The copied NASA archive includes about 6 billion files. Phys.org reports that it took about a year to move the data, which researchers compared to about 20 million feature-length films in storage size, or about a million times the storage on a typical computer.

That storage detail matters because forecasting is becoming as much a data-handling job as a weather job. Traditional weather and climate models use mathematical equations to simulate how air, water, heat, and pressure move through the atmosphere and oceans. Those models are powerful, but they can take hours to run. The AI approach described by ETH researchers uses statistical models, meaning models that learn relationships from existing data instead of recalculating every physical process from scratch. Reto Knutti, a climate physics professor involved in the work, said such a model can run a multi-day global forecast in about a minute.

For the people doing the work, speed changes the workflow. If a forecast takes hours, there are fewer chances to test different starting conditions or check whether a specific danger pattern is forming. If a model can run quickly, teams can do more iterations, which are repeated runs of a model with adjusted inputs or conditions. The machine is still only useful if the data pipeline works. A data pipeline is the full path that moves raw data from storage through cleaning, processing, model training, and final output. In this project, putting the data physically near the supercomputer reduces the delay that would come from waiting for massive files to move across slower connections.

The jobs behind this kind of system are not limited to one title. Climate scientists and computational physicists design the questions and judge whether model results make scientific sense. Data engineers build and maintain the systems that move, store, and organize files without losing track of them. Supercomputing technicians and system administrators keep the compute cluster, which is a group of linked machines working together, powered, cooled, connected, and secure. Remote-sensing specialists interpret satellite images and sensor data. Emergency managers and public officials would be the people turning a warning into an evacuation order or other action, if the science supports it.

The article points to two kinds of hazards where better data scanning could matter: landslides and glacier collapses. A landslide is a sudden movement of rock, soil, or debris down a slope. A glacier collapse happens when a large section of ice breaks loose, sometimes sending water, ice, and debris downhill. ETH researchers noted that not all disasters can be predicted. But they said some geology-related hazards can show warning signs in satellite data if there is enough monitoring and enough computing power to notice the pattern. The article cites the Swiss village of Blatten, which was evacuated before a glacier collapse last year, avoiding mass casualties.

For hiring, the takeaway is that technical workers in this field need more than one narrow skill. The article does not list job requirements, but the work described depends on people who can handle very large datasets, understand statistics well enough to question a model, and communicate risk without overstating certainty. A fast AI forecast is not automatically a correct forecast. Someone has to check whether the model is seeing a real pattern, a sensor error, or a coincidence in old data. That makes quality control, documentation, and teamwork part of the technical job, not extras.

The promise is not that AI will replace forecasters or emergency planners. The more realistic change is that AI may become another tool in the shop: fast, powerful, and only useful when trained, tested, and supervised well. The work moves between field knowledge, computer hardware, software models, and public safety decisions. In a career pathway, that matters because the future job may not be labeled “AI forecaster.” It may be data technician, satellite analyst, climate modeler, systems operator, or emergency planning specialist working on the same chain of information.