Seismologists often say that earthquakes do not kill people, buildings do, when they collapse because of a shock that they were not designed to withstand. An earthquake-prone country such as Italy needs a reliable seismic risk map that indicates what kind of earthquakes can be expected in each area, for authorities to use in designing a building code. Better information reduces the risk of tragedies such as the Irpinia 1980 earthquake, that killed almost 3,000 people; the one in L’Aquila in 2009 that killed 309 victims; or the series of earthquakes that hit the Marche, Lazio and Umbria regions in 2016, with a death toll of 299.
Italy now relies on a seismic map approved in 2004. As seismology evolves, the map must be updated with the data and knowledge produced in the last two decades, including from the many strong earthquakes that have hit Italy since then. For almost three years, the approval of an updated map has been delayed again and again, as the two institutions involved in the process fail to agree on its final form. The process, that Nature Italy has reconstructed through official documents and interviews with its protagonists, offers a glimpse into how earthquake science is undertaken, and highlights the tensions that emerge when scientific advice is translated into policy.
The story began in 2015, when the National Institute of Geophysics and Volcanology (INGV) started working on a new seismic hazard map for Italy, putting together a working group coordinated by INGV researcher, Carlo Meletti, and by Warner Marzocchi, then at INGV and now a professor at the University of Naples Federico II.
Together, they set out to develop a new version that would replace the 2004 map, called MPS041, on which the current Italian building code is based. INGV had been tasked by the government to develop that map in 2003, in the aftermath of the San Giuliano di Puglia magnitude 6 earthquake, in which 27 students and their teacher died.
Many new elements have emerged since 2004. “First, a richer catalogue of Italian earthquakes was issued in 2015, containing more accurate estimates of historical earthquakes magnitude”, explains Meletti. Secondly, there are many more accelerometric stations, that have helped refine the models that compute how much the ground will shake in a specific site as a function of an earthquake’s magnitude and the distance from its epicentre. “Some of those stations were located very close to the epicentre of major earthquakes in Italy in the last 15 years”, Meletti adds.
The new model, like the previous one, adopts a scientific framework known as Probabilistic Seismic Hazard Assessment. It starts by dividing a territory in seismic zones according to where faults or subducting zones are located, and to the history of past earthquakes. This allows to evaluate the frequency with which earthquakes of different magnitude may occur in each area, the so-called seismicity rates.
There are many different models to estimate seismicity rates. Marzocchi and Meletti brought in a large section of the Italian seismological community, asking researchers to propose their own model. The proposals were then tested according to their ability to explain historical data. Eleven models were finally selected and combined with an approach similar the one used for weather forecasting, where each model is assigned a weight, based partly on performance metrics and partly on interviews with a panel of experts. “We wanted to be as inclusive as possible and to avoid biases”, explains Meletti.
The following step was to estimate possible ground accelerations in each area, using other models that describe how seismic waves are attenuated as they move away from the epicentre. “Since 2004 we have learned that the shaking is more intense than we thought near the epicentre, but it fades more rapidly with the distance”, explains Meletti.
After some years of hard work, in early 2019 the team led by Meletti and Marzocchi had a first version of the new map, named MPS192. The new model had many methodological improvements over the previous one, that led to different estimations of earthquake hazard levels used for building codes across the country.
“This difference was not homogeneous on the Italian territory”, says Marzocchi. “In northern Italy seismic hazard levels predicted by MPS19 are higher than those in MPS04, because the new catalogue contains some intense earthquakes in the Po Valley that were not present in 2004, whereas in southern Italy, especially in the Tyrrhenian coast of Calabria and in western Sicily, the hazard levels are lower.”
The map now needed to be approved by the Italian Civil Protection Department, a government body. The seismic risk experts of the Major Risk Commission, the scientific advisory body of the Civil Protection, reviewed the model in progress but the evaluation wasn’t yet complete when their mandate expired. So they were appointed to an external committee to finish the work, and in May 2019 they approved the map. “I have strongly supported the new seismic hazard map”, says Francesco Mulargia, a seismologist at the University of Bologna and the coordinator of the committee. Its most important result, he says, is that the uncertainty on the model estimates was quantified for the first time. “We finally know the extent of our ignorance,” says Mulargia.
“We had a very fruitful collaboration with the model’s authors, who addressed all our questions and incorporated our suggestions” says Paolo Bazzurro, a structural and risk engineer at University School for Advanced Studies IUSS in Pavia, who was also on the committee.
Then the problems began. In September 2019, despite the committee’s positive opinion, the new members of the Major Risk Commission decided to run additional tests on the model. “We expected the commission to simply endorse our opinion” Bazzurro adds. “This was the informal agreement on the role of our committee, and we were quite surprised when they decided to start all over again”.
A second round of evaluation was conducted by Dario Albarello, geophysicist at the University of Siena, and Roberto Paolucci, seismic engineer at the Politecnico di Milano, both members of the new Commission’s seismic risk section. After a year of testing, in September 2020 “the commission’s seismic risk section stated that the MPS19 model could not yet be considered mature for the operational purposes for which it was commissioned”, the Civil Protection wrote to Nature Italy, and recommended changes in order to make the model “more suitable” to inform the Italian building code.
Marzocchi and Meletti explain that the Commission was especially concerned about the consequences that lower hazard levels in the southern regions could have on the building code. Nature Italy contacted both Albarello and Paolucci, but they declined to comment. The INGV research group devoted 2021 to making major adjustments to the model, based on the requests of the commission. In December 2021 it delivered a new version, called MPS19.s, which contains substantial changes with respect to the one published in the Annals of Geophysics.
The Civil Protection has told Nature Italy that “the commission expressed deep appreciation for the finalisation of the model and for the commitment of the INGV to the complex revision process”, and in February approved the model. But the lead authors are not happy with the result.
In the original MPS19, historical data were used to weigh the different earthquake rate models, but the authors had chosen to rely more on the frequencies of lower-magnitude earthquakes than on high-magnitude ones. “This is simply a consequence of acknowledging the uncertainty”, Marzocchi explains, “large earthquakes are rare and thus the raw frequencies do not allow a reliable estimate of the overall seismicity rates.” In contrast, the 2004 map adopted a precautionary approach by giving more weight to the raw frequencies of large earthquakes, which in turn led to an overestimation of the rate of medium earthquakes. Marzocchi believes this single choice explains much of the difference in the outputs of the two maps in southern regions.
A seismic model is like a tree, with several branches based on different modelling techniques and contributing to the final calculation. To make the new map resembling the old one, the Commission asked the authors to select only the branches that faithfully reproduce the intensities observed historically in nearly 120 sites. “This gives a much bigger relevance [than in the original MPS19] to historical data in constraining the model”, says Meletti. “The Italian earthquake catalogue is the richest in the world, but it is not exhaustive, especially on large earthquakes that have a longer return period”, he adds.
Gianluca Valensise, an earthquake geologist at INGV who contributed to both the 2004 and the 2019 map, says the changes requested by the commission also led to the discarding of parts of the model that were based on the Database of Individual Seismogenic Sources, compiled since 2001 under his coordination, with about 330 earthquake sources. “Knowing the location and geometry of large faults not only improves the resolution of the model in areas that have already generated strong earthquakes historically, but sheds light on the silent areas, such as in the case of the 2012 Emilia earthquakes,” he says.
The strict selection on the model branches led to another consequence that is especially concerning for the coordinators. “We washed away all the uncertainty which originated from considering different modeling approaches”, says Meletti. “The number of branches that survived is too small to allow a meaningful quantification of uncertainty.”
The last word
Despite agreeing to make the changes, Marzocchi and Meletti stand by their initial choices. “We were asked by the Civil Protection to produce the best possible scientific model that reflected the view of an entire community and that’s what we did”, Marzocchi says. “I understand the practical need of avoiding strong variations that could be challenging for the engineering community”. Marzocchi and Meletti would have preferred to combine the old map and the new one, rather than give up entire branches of the 2019 version, as happened in the end. “I collaborated in the latest version of the model, but I do not agree with it from a scientific standpoint,” says Marzocchi.
This is not the end of the story though. In May 2021, Carlo Doglioni, president of the INGV, had informed the Commission that the model may not pass an internal validation procedure required for all scientific output of the Institute that influences regulations. This process is separate from the scientific elaboration of the map by the group led by Marzocchi and Meletti, and involves the institute’s board of directors.
According to the Civil Protection’s statement, when MPS2019.s was completed in December, Doglioni informed the Civil Protection that the validation process had not begun yet, and that it would probably take eight months.
So now the whole process hangs on the very same institute that has worked on the map for years, using 3.5 million euros in public funding, approving it through its scientific committee and board of directors. This should normally be taken for granted, and the Civil Protection stated that “it hopes that the INGV internal validation process will be carried out in the shortest possible time”
Doglioni, however, has publicly criticized the approach to seismic risk assessment followed by his own institute, as well as by most countries with an advanced seismic risk management3. In an August 2020, interviewed by the Italian TV show Presa Diretta, he said that the probabilistic approach is outdated and needs to be abandoned. Nature Italy requested comment from the INGV presidency through its press office, but it declined to comment until the internal validation process is completed.
Embracing the uncertainty
What matters most in the end is how the seismic model is translated into building codes, and experts point out that it is possible to maintain the scientific soundness of the original 2019 model without lowering safety standards. The key, once again, is uncertainty.
In Italy, engineers have to make sure that buildings can withstand the average ground acceleration that has a 10% probability to be exceeded over a period of 50 years. But the model outputs can be much richer. The map allows to consider a lower probability threshold, or to extend the time horizon, in order to account also for rarer earthquakes. And for each different choice, it computes a distribution of probabilities for different ground accelerations.
If considering only the average value was justified by the lack of uncertainty estimates in the previous map, it is less clear now. Acknowledging the uncertainty would ease the disparity between the outputs of the new and old seismic hazard maps. “In areas where the model estimates of the peak ground acceleration come with great uncertainty, it would make sense to consider values higher than the average as a support to building design,” says Bazzurro. “This would favour a precautionary approach without touching the scientific basis of the model”.
“The new Italian map treats uncertainty very thoroughly”, comments Matt Gerstenberger, a seismologist at GNS Science in New Zealand who was a referee of the article where MPS19 was presented. “I was not aware about the recent changes in the model, and I hope that the entire evaluation, including this last step, is open and transparent,” he says.