Insights
Terna is the operator of the Italian National Transmission Grid, one of the largest in Europe. Listed on the FTSE MIB index, it manages approximately 76,000 km of high- and extra-high-voltage power lines and hundreds of substations, ensuring the security, continuity and quality of the country’s electricity service.
In 2024, Terna tested Eoliann’s predictive models, based on artificial intelligence and satellite data, with the aim of creating a more comprehensive and consistent flood hazard map. This was an essential step in identifying the areas of the grid most exposed to risk and targeting resilience measures more effectively.
Although the maps produced by Italy’s River Basin Authorities remain the official reference, they cover only 39% of the country in terms of water-depth data, with information that is often inconsistent across districts. The maps produced by the European Commission’s Joint Research Centre offer broader coverage, but exclude smaller catchments and lack the resolution required to accurately capture local features such as flood defence structures.
Against this backdrop, Terna launched a Proof of Concept with Eoliann as part of the “Data Science for Resilience” Innovation Bootcamp, to develop an approach capable of overcoming the limitations of existing sources.
The project, completed in March 2024, was jointly developed by Terna and Eoliann using a twofold approach: two-dimensional numerical hydraulic simulations, designed to generate physically accurate maps, and the application of the Airis Flood machine-learning model, to assess whether the analysis could be extended to a national scale.
Terna selected three pilot areas in Piedmont, each centred around a substation affected by documented flood events in previous years. This made it possible to compare the project’s results with high-quality observational data provided by ARPA Piemonte.
For each area, Eoliann developed two-dimensional hydrological and hydraulic simulations using only Earth observation data, including a digital terrain model, precipitation, land cover and the hydrographic network.
The simulations produced water-depth and water-velocity maps at a spatial resolution of 30 metres for three return periods – 20, 100 and 200 years – in line with the relevant regulatory framework.
The results were systematically assessed against three independent sources:
This made it possible to measure not only the quality of the simulations, but also their added value compared with the tools already available.
Airis Flood, Eoliann’s machine-learning model for estimating water depth, was then applied to the same areas. The model is already operational at European scale.
It combines geomorphological, climatic and urban variables to provide a continuous and consistent view of flood hazard at a spatial resolution of 30 metres, addressing the main gaps in public data sources.
Comparison between the extent of the observed flood event and the simulated water-depth map.
The project produced water-depth maps for the three pilot areas and the three return periods considered, validating them against both observed flood events and the main official sources.
The results confirmed the reliability of the hydraulic models.
The hydraulic simulations generated water-velocity maps, a variable that is not currently provided by public sources but is essential for assessing risk to transmission towers and substations.
When compared with the extent of flood events documented by ARPA Piemonte, the hydraulic simulations showed a good level of consistency, producing overall reliable results in representing the areas affected.
The machine-learning model tested as part of the Proof of Concept made it possible to explore the extension of the analysis beyond the pilot areas and the development of a more consistent view of flood hazard at national scale.
Unlike public data sources, Eoliann’s model provides continuous territorial coverage, with estimated flood extents at a spatial resolution of 30 metres. This makes it possible to overcome both coverage gaps and inconsistencies in official sources.
The project with Terna shows that strengthening the resilience of a national grid begins with making its risk easier to interpret through consistent, continuous and usable data.
Only when hazard is measured consistently and through its most relevant physical variables does it become possible to understand where to intervene, which assets to prioritise and what action to take.
Want to put Airis to the test on your assets? Request a demo!
Want to go deeper?
