UPK.1 | Keynote Presentation Understanding Weather & Climate Processes
Keynote Presentation Understanding Weather & Climate Processes
Co-organized by PSE.keynote
Convener: Frank Beyrich
Orals K-Thu
| Thu, 11 Sep, 17:30–18:00 (CEST)
 
Kosovel Hall
Orals K-Thu
Thu, 17:30
The keynote presentation on Flood warnings everywhere - data-driven rainfall-runoff modeling at global scale will be given by Frederik Kratzert from Google Research.

Frederik is a research scientist in the Flood Forecasting team at Google. His academic background is in civil (BSc) and environmental (MSc) engineering, as well as machine learning (PhD). Besides his work and research on the intersection of hydrology and machine learning that mostly focusses on improving reliable and actionable flood forecasts, Frederik also cares deeply about open source software and open data. He is the creator of NeuralHydrology, an open source Python library for training deep learning models in the context of hydrology, and also the creator of Caravan, a global community dataset for large-sample hydrology.

Orals: | Kosovel Hall

Chairperson: Frank Beyrich
17:30–18:00
|
EMS2025-481
|
solicited
|
Onsite presentation
Flood warnings everywhere - data-driven rainfall-runoff modeling at global scale 
Frederik Kratzert

Floods are one of the most common natural disasters, with a disproportionate impact in developing countries. The World Bank has estimated that upgrading flood early warning systems in developing countries to the standards of developed countries would save an average of 23,000 lives per year. With that in mind, Google started working on flood forecasting back in 2017, first by concentrating only on individual regions that are among the most affected regions by floods (e.g. India and Bangladesh). Over the following years we expanded to a handful of other partner countries but it was not until 2021, that we started concentrating on a global coverage. 

This was, to a large extent, made possible by the recent advances in deep learning based rainfall-runoff modeling. Taking the idea of regionally trained Long Short-Term Memory networks (LSTMs) by Kratzert et al. (2019) to an extreme, we started building global LSTM based forecast models that predict streamflow up to 7 days in advance (Nearing et al. 2024, Deborah Cohen 2024).

In this presentation, we will highlight some of the research that was crucial for building this model as well as some of the lessons learned, from operating a global-scale flood forecasting model and collaborating closely with different governmental agencies and non-governmental organizations. Lastly, we will look at ongoing research efforts to improve our flood forecasting system.

References:

Kratzert, F., Klotz, D., Shalev, G., Klambauer, G., Hochreiter, S., and Nearing, G.: Towards learning universal, regional, and local hydrological behaviors via machine learning applied to large-sample datasets, Hydrol. Earth Syst. Sci., 23, 5089–5110, https://doi.org/10.5194/hess-23-5089-2019, 2019.

Nearing, G., Cohen, D., Dube, V. et al. Global prediction of extreme floods in ungauged watersheds. Nature 627, 559–563 (2024). https://doi.org/10.1038/s41586-024-07145-1

Cohen, D. An improved flood forecasting AI model, trained and evaluated globally, https://research.google/blog/a-flood-forecasting-ai-model-trained-and-evaluated-globally/

How to cite: Kratzert, F.: Flood warnings everywhere - data-driven rainfall-runoff modeling at global scale, EMS Annual Meeting 2025, Ljubljana, Slovenia, 7–12 Sep 2025, EMS2025-481, https://doi.org/10.5194/ems2025-481, 2025.

Show EMS2025-481 recording (28min) recording