NCDP Perspectives

Source: Iván Portugués and Luis del Romero

LEAP Wallerstein Panel Series: AI + Extreme Weather Preparedness

Based on panel presentations and discussions by Gustau Camps-Valls, Félix Francés, Luis Gómez-Chova, and Ana Ruescas.

On October 29, 2024, Spain’s Valencia region was devastated by an extreme weather event: one year’s worth of rain fell in a period of eight hours, causing catastrophic flash floods that claimed 228 lives. The disaster highlighted the need for disaster preparedness and offered valuable lessons as governments adapt to the “new normal” of increasingly extreme events.

Dr. Pierre Gentine, the Director of LEAP, was in Valencia at the time of this natural disaster, working at the University of Valencia during his sabbatical. For him, the Valencia floods highlighted the importance of forecasting, early warning systems, and the need for accurate climate projections to aid in climate adaptation, a topic he has been working on for decades.

At the inaugural Wallerstein LEAP Panel Series: AI + Extreme Weather panel, Dr. Gustau Camps-Valls, Dr. Luis Gómez-Chova, and Dr. Ana Ruescas from the University of Valencia, as well as Dr. Félix Francés from the Polytechnic University of Valencia, discussed Valencia’s flood event with LEAP Moderator, Geneva List, exploring how AI can help bolster disaster preparedness and mitigation in the future.

Learnings from the Floods: From Extreme Rainfall to Humanitarian Impact

According to the panelists, Spain, with its advanced meteorological capabilities, identified the potential meteorological and hydrological disaster and issued flooding alerts on time. AEMET, Spain’s meteorological service, warned authorities and the public two days before the catastrophe occurred that there was a 70 percent chance of torrential rain and issued a red alert for severe weather at 7:30 AM on the day of the disaster.

However, warning texts to residents’ cell phones and mobile devices came ten hours later, with regional authorities sending text warnings to residents just after 8 PM, when many were already driving in their cars or outside their homes, with some already stranded by rising floodwaters.

​​Flood forecasting carries an added degree of uncertainty from meteorological prediction due to the geography of the catchment and inundation area. This uncertainty increases with longer lead times, presenting local authorities with a trade off: more time for preparedness and mitigation measures but less certainty in the hazard prediction and potential impact.

Nevertheless, while timely warnings could not have prevented the flooding hazard itself, early warning text alerts could have helped mitigate some of the disaster impacts that followed (i.e., the exposure and thus risk).

The Valencia floods, though an unprecedented climatic event, exposed significant shortcomings in coordination and institutional fragmentation across agencies in the dissemination of warning texts. In particular, the disconnect between national and regional authorities hindered swift decision-making. The differing levels of jurisdiction between state and regional governments created further confusion, complicating the disaster response process.

Public awareness and response added another dimension. Many residents were unaware of the true severity of the risk, leading to delayed or inadequate evacuations. Even when warnings were eventually sent out, some residents were unsure of the evacuation procedures or safety protocols. Further compounding the challenges, there was also a perceived lack of trust between the general populace and political officials.

Lastly, there were technological shortcomings, as flood sensors malfunctioned due to extreme conditions and high water levels, and critical data integration and sharing between agencies was lacking.

What role can AI play in helping to mitigate the impact of future extreme weather events?

As the Valencia flood event has demonstrated, prediction is just a small part of an effective early warning system. To move from observations and forecasts to warnings and decisions, several “bridges of death” must be crossed, all of which rely on different data, tools, disciplinary specializations, and corresponding institutions. If forecasts, for example, are not effectively translated to warnings, the chain will break. Luckily, AI can play an important role in each step along the chain to help achieve more effective decision-making and risk reduction outcomes.

The Early Warning Chain from Observation to Decision (and the “Bridges of Death”!)

Recent advancements underscore the growing importance of AI in understanding and mitigating extreme weather events. Dr. Camps-Valls’ recent article in Nature Communications emphasizes the critical role of AI in modeling extreme climate events and highlights the importance of creating accurate, transparent, and reliable AI models, given challenges like limited data and understandability. By integrating AI into climate models, we can better anticipate events like the Valencia floods and develop more effective preparedness and mitigation strategies.

AI-powered early warning systems can also enhance real-time forecasting by integrating satellite data with on-ground sensors. Nepal, for example, is rolling out an AI-powered early warning system that warns residents of hillside villages about landslides and helps them evacuate.

The importance of integration

Beyond early warnings and communication, AI can help break institutional silos by integrating multi-modal data (such as geospatial and socio-economic data) into a unified platform. It can facilitate better coordination between meteorology, hydrology, and emergency response agencies, streamlining data sharing and decision-making processes. Better coordination between national and regional agencies is crucial, and AI can help achieve this by improving data accessibility and ensuring timely information flow.

“Multi-modal AI can accelerate the shift from hazards to impacts, increase the locality and personalization of warnings, boost their accuracy and lead time, advance democratized access and enable a previously unseen level of interactivity.” (Reichstein et al., 2025)

AI also has significant potential in risk communication. It can improve public risk perception by generating clear, understandable warnings tailored to different audiences. Large Language Models (LLMs) could be leveraged to provide real-time, adaptive risk communication. However, AI must work alongside traditional models to ensure explainability and trust in predictions, making Explainable AI (XAI) a key component.

“Explainable AI (XAI) aims to unveil the decision-making process of AI models, […]  it facilitates debugging, improving models, and gathering scientific insight by revealing the model functioning, learned relationships, and biases.” (Camps-Valls et al., 2025) 

AI-generated warnings must be carefully designed and integrated with other monitoring and warning systems to avoid false alarms while ensuring timely evacuation. The public must also be trained that false alarms through AI (much like traditional forecasting approaches) are possible and that it’s better to have a false warning than no warning at all.

With extreme weather events becoming more frequent, improving disaster response systems is more essential than ever. Learning from extreme weather events like the Valencia flooding is critical to understanding the current state of disaster response. Our panelists concluded that AI, while not a panacea, can be a powerful tool in reducing disaster risks when combined with effective governance, community engagement, and strong institutional coordination.

This was produced by the National Science Foundation’s Center for Learning the Earth with Artificial Intelligence and Physics (LEAP) in collaboration with the National Center for Disaster Preparedness (NCDP).

__

References

Bhandari, B. (2024, June 11). Nepal rolls out AI-powered ‘crystal ball’ to predict deadly landslides. South China Morning Post. https://www.scmp.com/week-asia/health-environment/article/3266082/nepal-rolls-out-ai-powered-crystal-ball-predict-deadly-landslides

Camps-Valls, G., Fernández-Torres, M.-Á., Cohrs, K.-H., Höhl, A., Castelletti, A., Pacal, A., Robin, C., Martinuzzi, F., Papoutsis, I., Prapas, I., Pérez-Aracil, J., Weigel, K., Gonzalez-Calabuig, M., Reichstein, M., Rabel, M., Giuliani, M., Mahecha, M. D., Popescu, O.-I., Pellicer-Valero, O. J., … Williams, T. (2025). Artificial intelligence for modeling and understanding extreme weather and climate events. Nature Communications, 16(1), 1919. https://doi.org/10.1038/s41467-025-56573-8

Portalés-Julià, E., Mateo-García, G., Purcell, C., Gómez-Chova, L. (2023) Global flood extent segmentation in optical satellite images. Nature Scientific Reports, 13, 20316. DOI: 10.1038/s41598-023-47595-7 (https://www.nature.com/articles/s41598-023-47595-7)

Frost, R. (2024, November 7). A weather warning expert on how emergency texts failed Valencians. Euronews. https://www.euronews.com/green/2024/11/07/early-warnings-save-lives-could-valencia-have-been-better-prepared-for-deadly-flooding

Kurbalija, J. (2024, November 5). Valencia flooding: Why did one of the leading smart cities fail on basic safety? Diplomacy.Edu. https://www.diplomacy.edu/blog/valencia-flooding-smart-city-safety-failure/

Man presumed dead in 1994, found among the victims of the deadly Valencia floods. (2025, March 7). Majorca Daily Bulletin. https://www.majorcadailybulletin.com/news/local/2025/03/07/131663/man-presumed-dead-1994-found-dead-valencia-floods.html

Medrano, T., & Wilson, J. (2024, October 31). What to know about the unprecedented floods that killed more than 200 in Spain. AP News. https://apnews.com/article/flash-floods-spain-valencia-climate-change-what-to-know-f942142b82de24f5b4a18867bc32ae00

Reichstein, M., Benson, V., Blunk, J., Camps-Valls, G., Creutzig, F., Fearnley, C. J., … & Weldemariam, K. (2025). Early warning of complex climate risk with integrated artificial intelligence. Nature Communications, 16(1), 2564.

World Meteorological Organization. (2024).  Devastating rainfall hits Spain in yet another flood-related disaster. https://wmo.int/media/news/devastating-rainfall-hits-spain-yet-another-flood-related-disaster