2021 Shortlist

Oldies but goodies? We looked at 40 years of “anomalies” in French nuclear plants

Country/area: France

Organisation: Contexte

Organisation size: Small

Publication date: 21 Feb 2020

Credit: Yann Guégan,Victor Roux-Goeken

Project description:

For the first time, journalists were granted access to detailed information about 30.000+ so-called “significant security events” that took place in French nuclear power plants since 1977. We explored this huge and very technical dataset to answer a highly sensitive question: are older reactors really more dangerous?

Working closely with independent experts from the Radioprotection and Nuclear Safety Institute, we established that even if the number of incidents is not rising as a plant gets older on average, a growing share of the reported anomalies are caused by / related to the aging of the facilities.

Impact reached:

The results of this months-long investigation fueled a long debate about the future of existing power plants. France relies heavily on civic nuclear power to produce the electricity it needs (70%, a world record), and 20+ of the 58 nuclear reactors built in the country are nearing the age of 40. Decisions are to be made about a possible life extension, that would require costly and time-consuming renovation processes.

“Pro-nukes” and “anti-nukes” spend a lot of energy promoting their views, and the climate crisis has raised the stakes even more: despite all his flaws, nuclear energy is a low-carbon mean to generate electricity, Therefore, some experts consider that keeping the current power plants up and running offers a favorable risk-reward ratio. Others warn that the probability of an accident is getting higher by the day, and that a nuclear-free production mix is a possible goal to reach by 2050, even in France, if the country shifts its focus towards renewable energy.  

Yet until the release of our story, the actual publicly available data about what really happen in French power plants was scarce. Occasionally, journalists ran stories about a particular incident, based on press releases from official agencies or environmental NGOs. But they could not put it in a broader context and therefore attempt to draw a conclusion.

Even if Contexte is a B2B independent news outlet relying on subscription, we decided to remove the paywall for such a public-interest story. Our work was praised by the most skilled experts in the field, and heavily commented on social media by stakeholders from all sides. It’s by far the most read content of our website in 2020.

Techniques/technologies used:

We used Python and specifically the Pandas module to parse, clean and explore the 30 MB of Excel files that were provided by the Radioprotection and Nuclear Safety Institute (RNSI). We produced 60+ summary tables to be imported and studied in Google Sheets. We came up with 25+ charts created with Datawrapper and embedded the most relevant of them in the main story.

We also released an additional interview with two experts from the RNSI, to give more information about the database itself and how they used it today. Called “Sapide”, it was created in the 1970s by the RNSI. Its first iteration was a collection of paper sheets stored in wooden drawers.

Sapide was digitized and beefed up throughout the years, but the core structure remained the same: each “significant security event”, even minor, must be declared by the plant workers and registered in it, triggering rapide response as well as long term investigations from the regulator.

What was the hardest part of this project?

While exploring the dataset, we soon came to realize that its content was as difficult to comprehend as a nuclear reactor itself. Even the basic description of the event (a plain text field) does not make sense to the profane reader – at first glance, they all look like the beginning of the Chernobyl TV series, but that’s about it.

We therefore had to familiarize with the taxonomy used for each field by going through hundreds of documentation pages. They go from the pieces of equipment impacted by the event to its causes and consequences on the power plant.

Understanding and scoring the severity of a particular event was even more challenging. We has to evaluate the numerous indicators available, understand how they are computed and what they tell us about the safety status of the reactor.

Lastly, to answer our initial question, we had to figure out if the age of the reactor had something to do with the recorded incident – even a brand new reactor can run into a problem, in fact, younger installations experience more issues than mature ones. We decided to base our conclusions on two criteria: was aging listed as one of the causes of the incident in the related field? Is aging mentioned in the description of the incident in some way?


What can others learn from this project?

Here are some takeaways:

  • you can work on a dataset even if you don’t understand every tiny bit of it.
  • you probably need help from specialists to understand the bits that matter to you (and they will happyly do so).
  • there is no subject so complex that you can not make sense of it for you and your reader. It will just take more time and more energy.
  • Excel is not the best tool to work on Excel files if they are too big.
  • a journalistic work can be deemed worthy of interest by the most skilled experts – not because it’s better, just because we offer a different perspective on things.  

Project links: