2.2.1. General statistics

A synthesis of the principal data and associated uncertainties contained in the database is presented in Table 2. Among these data, one should essentially notice:

• 6427 events are contained in the EVTSIRENE table, 5743 of them are qualified as “true events”. Others are considered either doubtful or false events;
• among the 5743 true events reported in the EPCSIRENE table, only 3131 of them come along with a quantified epicentral intensity “Io” (54.5%), of which 880 are known through a single IDP (Qpos = I “Isolated”). This means that solely 2251 events (39.2%) of the database have been characterized with more than 1 IDP;
• 605 true events highlight epicentral intensities equal or higher than VI MSK (i.e., events having potentially produced structural disorder on buildings);
• 12532 documents (written archives) allowed determining a total of 108832 MDP. 107592 of them are associated with true events and 86346 have quantified intensities (IDP);
• In addition to the information contained in Table 2, one may note that a total of 1235 earthquakes are qualified as foreshocks (256 events) or aftershocks (979 events).

As a consequence, one should be aware that at least half the database is composed of events that could be qualified as poorly characterized. On the other hand, taking into account an uncertainty of A and B (see Table 1 for details) onto the estimation of the epicentral intensity, around 1100 events may be qualified as properly known.

The evolution of the database since 2004 [Scotti et al. 2004] is presented in Figure 3 (and Table 2). While the number of new documents (∼213) and new events (∼36) discovered per year is relatively constant over the time period, it is not the case concerning the number of MDP (IDP and Felt). From 2004 to 2007, a mean number of 4586 MDP/yr were added to the database, and 451 MDP/yr afterward and until 2017. This difference could be explained by the end of the integration within SISFRANCE of the BCSF surveys in 2007. Since then, the rate of new MDP/yr is that corresponding to historical events strictly speaking and remains globally stable over time. This provides a relatively clear idea of what we may expect to find annually with the work of an historian dedicated to this project (note that the person in charge at BRGM was dedicated 30% of his time working on this task).

2.2.2. Focus on epicentral parameters

An important point when it comes to feed a database is that the methodology should be as homogeneous and reproducible as possible and when subjected to expert opinion, that such interpretation can be traced in order to exploit them in a seismotectonic framework. Indeed, as already pointed out by Ambraseys [1983], a seismotectonic approach to hazard assessment requires good quality epicentral locations and precise epicentral intensity determinations. The methods used to determine epicentral parameters and their use to define seismogenic sources (zones or faults) are a matter of debate [Richter 1958; Cecic et al. 1996]. Cecic et al. [1996] present a review of these issues across Europe.

In SISFRANCE, the epicentral location of earthquakes is determined prior to their epicentral intensity which depends on it. Few simple rules are in general followed to establish the location of events:

• When a single MDP is available, the epicenter of the event corresponds to the location of this data point, hence qualified as isolated (QPOS = I; 2521 events, Figure 4). The epicentral intensity is that of the single available MDP, often unknown (felt—1641 events), but it happens that events having only one MDP have a quantified Io (i.e., 880 events, 41 of them having an Io ⩾ VI). This is the simplest case, corresponding to a significant part of the database;
• When a few MDP are available, the location of the earthquake may correspond either to (1) the location of maximal intensity when characterized by a single IDP; (2) the barycenter of the strongest IDPs, weighted by their quality (Qiobs); (3) the barycenter of weaker IDP’s, when they are better represented in number and spatially; (4) the barycenter of felt data points when no IDP is available;
• When a significant number of MDP are available, allowing to draw isoseismal lines, it enables the determination of the macroseismic epicenter as the barycenter of the highest intensity area [see Levret et al. 1994]. In such cases, the epicentral intensity is not always equal to the maximum intensity observed locally [owing to site effects due to local amplification, Levret et al. 1994].

However, additional information contained in the archives also led the contributors of SISFRANCE to constrain the epicentral locations with more subjective criteria. A recurring criteria being, for example, to consider the occurrence of foreshocks or aftershocks felt very locally in order to guide the choice of an epicentral location. Considering that part of these epicenters was proposed on such subjective criteria, it was decided at the beginning of the 2000s that any modification of an earthquake of intensity greater than VI (i.e., impacting most the seismic hazard) would be accompanied by an internal justification note. This approach having been lately established, a given number of earthquake epicentral parameters might need reinterpretation. In this regard, a future improvement of the database may consist in establishing and applying a more systematic definition and application of criteria leading to the definition of the epicentral position of historical earthquakes, even though it can be questionable in a seismotectonic perspective.

Once the epicentral position of an event is defined, the epicentral intensity can then be deduced. In most cases, the epicentral intensity is equal to IOBSmax, especially when the epicenter is defined at the location of a punctual observation, which is the case with isolated events, but not only. The intensity Io also corresponds to IOBSmax when the macroseismic observations of maximum intensity are close in space to each other. When this is not the case (i.e., the observation points are spaced from each other and the epicentral location is at the barycenter of this distribution), a calculated intensity (QIE = K) is established on the basis of an attenuation law describing the decrease of intensity with distance following a Sponheuer [1960] relationship [Levret et al. 1994], or an arbitrarily intensity (QIE = E) is defined when the data points are too scattered and/or too far from each other.

A cartographic illustration of the number of MDP that allowed establishing epicentral parameters is presented in Figure 4. Similarly, the evolution over time of the number of MDP collected for each event is presented in Figure 5, while the temporal distribution of seismic events by intensity class is presented in Figure 2b. These different representations allow appreciating the representativeness and completeness of historical data:

• Most of the historical database contains events for which the number of MDP is equal or lower than 10 (4550 event over the 5743 of the true events, Figure 4). They obviously mainly correspond to events of unknown epicentral intensity (Figure 5), but also to non-damaging events and even some damaging events, going back through time (Figures 4 and 5). Indeed, a simple linear regression on the different defined classes of events (Figure 5) shows that the number of MDP we may expect for a given earthquake decreases as a function of time and epicentral intensity;
• Figure 2b suggest that the chances of finding the mention of an earthquake within the historical archives increases with increasing intensities going back in time. It somehow provide hints concerning the completeness of the dataset for each epicentral intensity class (discussed hereafter), and rather clearly shows that our knowledge significantly increases starting from the 15th century;
• Looking at the spatial repartition of the best-known events (>10 MDP), we observe that the number of low intensity events (in blue and yellow, Figure 4) in the northwestern part of France is over-represented in comparison to what is observed in the Pyrenees and in the Alps, where observed seismicity rates are higher [Mazzotti et al. 2020]. This is difficult to interpret directly because different factors may come into play, such as the completeness of the historical archives, the habitat distribution in plains versus mountainous regions, or even regional differences of intensity attenuations with distance [Bakun and Scotti 2006; Baumont et al. 2018]. As an example, the Provence region highlights a number of significant events for which few MDP have been collected. One explanation may come from the fact that in this region of written/Roman law (i.e., archived documents are strongly regulated), where ancient narrative sources are rarer than elsewhere in France [Giordanengo 1988], annotations relating natural phenomena are rare throughout the middle ages.
• Concerning events with more than 100 MDP (Figure 4), they mostly correspond to earthquakes occurring since the 19th century, with a strong increase starting from 1921 (birth of BCSF, Figure 5);
• A significant number of events for which an epicentral intensity has not been defined are characterized by more than one MDP (Figures 4 and 5). If this is normal for the oldest events, because few descriptions are provided in ancient archives (i.e., MDP often qualified as “felt”); it is more surprising concerning the younger ones for which an effort to precise their epicentral characteristics should be performed;
• By construction, all the events characterized by a single MDP are located onshore (Figure 4), which does not mean that part of these could be situated offshore. More generally, the location of earthquakes in the marine domain is associated to strong uncertainties [e.g. Larroque et al. 2012];

However, such a global overview of the database does not allow retrieving enough information that would allow at the national or more regional scales to directly address the notion of completeness as needed for the definition of seismicity rates [Rydelek and Sacks 1989; Stucchi et al. 2004]. Indeed, historical knowledge varies and evolves regionally due to geographic, socio-economic, and political conditions, which implies that the completeness of historical data must be assessed specifically for each single region, before it can even be extrapolated to the entire country. Such an observation is all the more true concerning “small” earthquakes for which the impacted area is small, but it is also true when it comes to focusing on major events that could be missing, missed, or underrated [Valensise and Guidoboni 2000].

The take-home message for any end user interested in using the epicentral dataset is to be aware that epicentral intensity and location are the two most expert-opinion-driven parameters in historical databases, derived from a second level of interpretation [Scotti et al. 2004]. While these data are crucial in order to plot the seismicity, perform seismotectonic, and seismic hazard analyses, they must be handled with care and consciousness, keeping in mind the historical and geographical context of the data they derive from. The uncertainties provided in SISFRANCE can definitely help exploiting these data [e.g. Provost and Scotti 2020], but they may not to be always sufficient for specific studies. As an example, the great number of events associated with an undefined epicentral intensity, even within the most recent times, recalls that the knowledge provided by archives is intrinsically limited.

2.2.3. Focus on damaging earthquakes

A focus on damaging events reported in SISFRANCE is provided in this section. 605 earthquakes have an epicentral intensity compatible with the occurrence of structural damage on buildings (Io ⩾ VI, Table 2), of which 336 are located within the French territory. The others are mainly distributed near the French border, but some of the strongest events occurring at a great distance and having been felt in France also appear in the database, such as the great Lisbon earthquake of 1755.

A cartographic representation of these events, associated with their uncertainties is presented in Figure 6. A distinction is made between events qualified as “fairly well known” (QPOS = A, B, C and QIEPC = A, B, K, see Table 1) and those for which the uncertainty about the location or the epicentral intensity is greater (QPOS = D, E, I; QIEPC = C, E, I). Of the 412 events qualified as “uncertain”, practically half are located in France (202 events). Likewise, out of the 193 well-known events, a much larger part of them is located in France (134 events). We then notice that a majority of events located abroad, but also at sea are of “uncertain” quality. This is obvious concerning offshore events, for which fishes, unfortunately, cannot witness. But for those located beyond the borders, this is explained by the fact that research was carried out with the objective of prioritizing the collection of MDP in France and also that the data from foreign databases are not integrated into SISFRANCE. Again, it underlines the benefits from future cross-border collaborations on past events characterization.

Beyond this spatial aspect, the birth of the BCSF 100 years ago (i.e., in 1921) allowed an important leap forward in terms of knowledge concerning, in particular, the stronger events. Indeed, one may notice that, of the 134 best-known events reported in France (Io ⩾ VI), 75 took place after the creation of the BCSF, and therefore since the establishment of almost systematic macroseismic surveys. Surprisingly, since the onset of macroseismic inquiries, 14 events located in France are still qualified as uncertain, the last of them dating back to 1982.

Of the slightly more than 100,000 MDP associated with real events in SISFRANCE, ∼50% of them correspond to events of epicentral intensity greater than or equal to VI (51003 MDP), of which 15045 were acquired before 1921. Between 1921 and 2007, SISFRANCE data corresponds primarily (70% of events) to a reinterpretation by SISFRANCE of the macroseismic surveys and questionnaires from BCSF and BRGM. Macroseismic data for events after 2007 are only available at the BCSF website (www.franceseisme.fr).

The number of IDP for each level of intensity exceeding VI is presented in Figure 7. The highest intensity points (I = IX–X to X–XI) are only found abroad. In France, the strongest IDP are 2 points of intensity IX recorded following the Lambesc earthquakes in 1909 (Io = VIII–IX) and the Ligurian earthquake of 1887 (Io = IX). Regarding intensity levels VII–VIII to VIII–IX, we note that they were mostly recorded before 1921 (73 out of 87 IDP and 68 out of 71 IDP, respectively), with the occurrence of few strong earthquakes since then. The 14 points of intensity VII–VIII and the 3 points of intensity VIII observed since 1921 are associated with the earthquakes of Viella in Spain (1923, Io = VIII), Chasteuil (1951, Io = VII–VIII), St Paul sur Ubaye (1959, Io = VII–VIII), Corrençon en Vercors (1962, Io = VII–VIII), Arette (1967, Io = VIII) and Arudy (1980, Io = VII–VIII). Since 2007, only the Le Teil earthquake reached such levels of intensity [see Schlupp et al. 2021]. Concerning IDP’s below VII–VIII in France, we note a significant increase in the number of municipalities affected by these intensity levels since 1921. This is of course not related to an increasing rate of damaging events in recent times, but to a more complete view of the impact of earthquakes provided by the macroseismic surveys of BCSF. On the contrary, the number of municipalities impacted by intensities above VIII drops drastically in France, occurring only before the birth of BCSF (7 IDP of intensity VIII–IX, 2 IDP of intensity IX). This could be explained (1) by a return period of strong earthquakes equal or longer than the period of time covered by the database, or (2) by the incomplete knowledge we have concerning the stronger events, especially for the older ages. Both hypotheses are only valid when considering stable seismicity rates over time. We favor the second hypothesis considering the poor historical knowledge we have for the older events. For instance, the Basel earthquake of year 1356 is an interesting example. This event of epicentral intensity Io = IX, is characterized by only two IDPs = VIII–IX, and 42 IDP = VIII, coming from the fact that we most probably miss information mentioning cities or villages located the closest to the epicenter.

3.1.1. Before the renaissance (XVIth century)

In France, until the Middle Ages, historical sources have an eminently religious character (annals), while civil chronicles are rarer. Thanks to the increasing centralization of powers, accompanied by an increasing documentary richness over time, civil chronicles also became better and better archived and preserved (Figure 8). Given the relatively limited amount of ancient archives available, it is now possible to believe that the work carried out for years has enabled the vast majority of earthquakes that have been reported for this period to be identified. In particular, Alexandre [1990] established a critical catalog of the narrative sources at a European scale, thus discarding many documents whose little historical value had been demonstrated. For example, following the “pneuma theory” of Aristotle [Poirier 2008], confusion could often be made with the occurrence of storms which, until the 16th century were often referred to as “terrae motus”, hence classified as “false earthquake” within SISFRANCE. The earthquake annotations recorded during the Middle Ages are, with the exception of major events such as the Basel earthquake in 1356 [Lambert et al. 2005], short and lacking in detail, even when damages are mentioned. Then, uncertainties regarding both the location and intensity of ancient earthquakes remain most often strong. However, notarial records, which appear from the beginning of the 14th century, and which have been little explored to date, could still provide indirect details concerning a potential impact of earthquakes on people and properties. These handwritten archives are, however, difficult to access, although their exploration yielded interesting results for a more recent period [i.e., for the Manosque earthquake in 1708, Quenet 2001]. Finally, exceptional discoveries can still be made on the occasion of research in archival records accessible only to a few historians, as was the case in the Vatican archives concerning the earthquake of 1186 near Uzès [Castelli et al. 2012]. Overall, the period of the Middle Age is marked in SISFRANCE by a significant increase in the number of MDP for the strongest events and from the 12th century, with a steady increase afterward (Figure 5).

3.1.2. From the renaissance until the mid-XVIIIth century

From the middle of the 16th century, the renaissance in general and more specifically the development of printing press (e.g., birth of the Gazette de France by T. Renaudot in 1631 and other Gazettes across Europe), and the keeping of parish registers made compulsory by the ordinance of Villers-Cotterêts of August 1539 [Foyer 1989], contributed to considerably increase the number and the nature of archives, thus increasing the likelihood of finding records, which could mention earthquakes (Figure 8). Also corresponding to the appearance of the notion of “disaster”, this period saw the appearance of the quantification of human and material tolls caused by disasters (including earthquakes), in connection with the development of the figures of the victim and of the hero [Labbé 2017]. Even though the effects of earthquakes on people and buildings are then frequently reported, they are unequally detailed depending on the authors and the purpose of their writings, which do not yet have an established scientific approach. Moreover, the historical context, whether national or more local, is also largely influencing the quantity and quality of the information available. Indeed, long periods of “silence” are observed in the archives depending on whether periods of war, famine or other major events naturally take precedence over the occurrence of “minor” seismic events. We may, however, consider that the strongest events are identified for that period and, with a few exceptions, relatively well characterized. On the other hand, the hope of enriching knowledge for this period is widely permitted, in particular, because private and administrative archives (e.g., municipal deliberations, notarial archives, etc.), and in general, handwritten archives have not been explored much in comparison with the printed archives. As the increase in the number of MDP per event noted in SISFRANCE since the end of the Middle Ages continues, events of lower epicentral intensity are beginning to be more systematically reported (Figure 5).

3.1.3. From the mid-XVIIIth century until the modern era (XXth century)

From the second half of the 18th century, the development experienced during the Renaissance period continued and intensified. Press organs are multiplying and the general interest to science is taking off with the multiplication of the number of learned societies. Concerning earthquakes, in particular, the Lisbon earthquake of 1755 (Io = X–XI), widely felt in France, deeply marked the people during this period of enlightenment [Dynes 2000], and marks a real turning point in terms of available earthquake related information (Figure 5). However, as quantity is not always a guarantee of quality, the circulating information is often repeated and sometimes quickly interpreted, distorted, amplified, leading to a necessary effort to interpret and cross-reference important data. This is all the more true in the 19th century when the press exists at all levels of society and, as it does today, often relays second hand information. In parallel, the scientific literature is developing with the aim of being readable by the general public [Metzger 1934] and the first syntheses of historical and contemporary seismicity emerge [e.g. Pallassou 1815; Perrey 1845; Montessus de Ballore 1906], as well as the first surveys aimed at locally quantifying the impact of earthquakes [see Vogt 2003; Camelbeeck et al. 2021 and reference therein]. A very marked peak in the number of events and associated MDP is observable following the Lisbon earthquake of 1755, then followed with a strong and steady increase during the 19th century. In this very rich period, the quantity of archives available is such that their exploitation is today incomplete and that many new information can still be discovered, with the possible exception of the strongest events for which in-depth research has often already been carried out.

3.1.4. Modern era

Just like the earthquakes of Basel (1356, Io = IX) and Lisbon (1755) in their time, the earthquake that occurred near Lambesc in 1909 (Io = VIII–IX) marks a new turning point. This is indeed the first major earthquake for several generations [Vogt 2003] in the French territory. The Central Meteorological Bureau took matters in hand with a detailed macroseismic survey published by Angot and Lemoine [1910], which led, a few years later, to the creation of the BCSF in 1921 [Fréchet 2008; Roger 2021] and to a systematization of the macroseismic surveys during the 20th century. Resorting to other sources of information remains, however, necessary and can still clarify our knowledge, in particular, because the action of the BCSF has not been continuous and homogeneous over the century [Roger 2021; Sira et al. 2021]. The period was marked by a high and almost constant number of MDP recorded in SISFRANCE, until 2007 (Figure 5).

3.2.1. Archival researches: lessons from past and present

For each aforementioned historical period, new avenues of archival research can be considered to enrich the database (Figure 8). However, the experience gained so far shows that the depth of investigation in the archives, and hence the necessary time dedicated to it, becomes increasingly more important as the most accessible documents have already been searched for and analyzed. Between 2004 and 2017, the pace of unpublished archive discoveries (∼213 per year; see Section 2.2) was maintained by continuous research and systematic monitoring of the archives by a dedicated historian, largely facilitated by the continuous online posting of new archives on the Internet. On the other hand, research carried out on more specific events, targeted considering their potential impact on the seismic hazard for nuclear installations (i.e., in general, strong and poorly known events), have often proved to be more complicated, each new piece of data implying a substantial investment. It is important to underline again that the efforts carried out in SISFRANCE were largely oriented by the needs linked to the development of the nuclear industry in France [Roger 2021]. As a result, some regions in general and specific events, in particular, have been the subject of more in-depth study than others. This does not mean that the rest of the historical seismicity in France has been left out, continuous research all over the country allowing the knowledge to evolve regularly, but it is important to note that many earthquakes have not been reviewed since their implementation in SISFRANCE in the 1980’s.

To summarize, the constitution of a database of historical seismicity relies first and foremost on the ability to find, contextualize, and analyze original documentary sources. This work requires combining the wide range of skills of historians (knowledge of the organization of and access to archives, translation into modern language, filiation of sources, historical contextualization …) with those of historical seismologist for the translation of testimonies into values of macroseismic intensity. The retirement of J. Lambert in 2017 and the non-renewal of his position led the SISFRANCE consortium to subcontract historical researches to academics or research firms.

This approach allowed diversifying and modernizing the search methodology for original documents, with the participation of new people bringing a fresh look and new research skills. However, in order to guarantee the homogeneity and continuity of interpretations with the work carried out in the past, this phase is now completely devoted to the SISFRANCE working group. After three years of operation, this way of conducting new researches offers the possibility to target on specific earthquakes, archives or periods of interest, with requirement specifications defined upstream. However, it has the main drawback of no longer allowing wide-ranging researches and therefore precludes the continuous enrichment of the database for events not considered as a priority. In addition, the research carried out by professional historians is limited to their disciplinary field, and the training of people capable of having a critical look between history and seismology is no longer guaranteed.

In order to complement the continuous enrichment of the database performed with classical archival researches, new avenues of research are under investigation such as “data-mining techniques” aimed at automating the collection of new data from the huge amount of digitalized archives [Nayman et al. 2020], or participatory science. For the latter, the ongoing modernization of the www.sisfrance.net website is an opportunity to collect information by allowing the public, whether from academia, associations or individuals, to report new documents still unexploited. In addition, in the framework of EPOS-RESIF [Masson et al. 2021], macroseismic data from the Sisfrance database will soon be available together with data from BCSF in a unique website (www.franceseisme.fr), offering the opportunity of an enhanced visibility. It is nowadays difficult to draw consolidated feedbacks concerning these actions as we do not have the necessary hindsight. But it however appears, as shown by the last update of the database in 2017, that historical seismicity research will no longer be as effective as in the past as long as academic researches and hence the training of researchers skilled to the use of historical data related to natural events (earthquakes or other) will not be carried out in parallel to SISFRANCE.

3.2.2. Improving the database

The SISFRANCE database relies on a structure developed at the beginning of the 1980s. If it has proven its efficiency, remaining fully usable after 40 years, it is indeed not flawless.

There is a clear need to improve the hierarchization of the archives in SISFRANCE as pointed out by Quenet [2001] by setting up hierarchical trees explicitly presenting the relation between primary and secondary sources [Quenet et al. 2004]. He also recommended an enhanced contribution of historians to the critical analysis of historical data. Since fifteen years, the systematic production of justification notes tracing and explaining the modification of earthquakes of epicentral intensity greater than or equal to VI allowed, at least in part, to tackle this problem. However, a retrospective review of older modifications, for which specific notes were not systematically produced, constitutes a difficult task needing the involvement of historians or people trained to the critical analysis of archives and to the specificities of macroseismic analysis.

The experience gathered within the SISFRANCE working group also makes it possible to identify a certain number of areas for improvement, which are more achievable today:

• In the SISFRANCE database, documents are related to earthquakes (through Chrono and Numevt field, Figure 1). The initial architecture of the database, conceived in the early eighties did not anticipate the possibility to link each single MDP to the documents that made it possible to establish them (see the dotted arrow in Figure 1). If this point does not pose a real problem in the case of earthquakes for which few documents and few MDP are available (i.e., the major part of the database), great difficulties arise as soon as they are more numerous. Indeed, when a modification has to be made for such or such well-known event, it is difficult today to recover the document(s) at the base of the definition of an MDP, and therefore to discuss its possible modification on the basis of a new document. OCR (Optical Character Recognition) techniques available today should help overcome this problem in the coming years by offering the possibility to relate the OBSIRENE and DOCUMENTS tables together. However, while a systematic ocerization (defined as the digitization of the texts contained in the archives) of printed documents is feasible, it is still complicated for handwritten documents and a manual work is still required. The ongoing “data-mining” project [Nayman et al. 2020], which requires the ocerization of SISFRANCE, will constitute an essential step forward, offering the possibility to overcome this problem in the coming years;
• Each year, the SISFRANCE working group has to establish a research program for the following one. Such a yearly exercise revealed the difficulty to know precisely which archives have already been investigated (where and when), and also those which gave negative results (therefore not referenced in SISFRANCE). A cartography of the investigated archives is partly possible starting from the referenced documents (feasibility is actually under evaluation), but it is clear that a complete vision is not achievable and that the consultation of archives that have already been visited in the past cannot be systematically excluded;
• The increasing availability of archives on the Internet is a major element allowing archival research to be carried out more and more efficiently. As part of these archives are also ocerized, their exploration by keywords are fast and comfortable. However, this implies the constitution of an extensive lexicon related to the phenomenology and impact of earthquakes. The simple term earthquake is, for example, not used systematically over the history. Such work has been carried out by Fradet [2016] for a limited time period and is also underway in order to supervise computerized searches during the “data-mining” process [Nayman et al. 2020, see next section].

3.2.3. The promising data-mining approach

The principle of data mining is to extract automatically useful information from various numerical databases available online. Applied to seismicity, it is an interesting methodology to improve knowledge of historical earthquakes by finding new testimonies in written archives available on the web and, therefore, to enrich the SISFRANCE database.

A dedicated method for historical documents investigation has been developed and applied to the digital library of the Bibliothèque Nationale de France (BNF), named Gallica [Nayman et al. 2020]. Given the large amount of harvested texts (3.8 millions) on this website, algorithms were designed to build and implement an efficient workflow strategy: (i) Definition of seismological ontology (6 concepts: seismic, damage, assembly, behavior, noise, divine) from SISFRANCE database which is used as dedicated dictionary to extract relevant information from the Gallica collection of documents; (ii) Semantic and knowledge enrichment of harvested documents which constitute the knowledge base; (iii) Use of advanced techniques of data mining, as for example the use of a similarity process that dramatically helps to find relevant text through the background “noise”.

After one year of investigations in the Gallica Corpus, the work led to more than 1600 documents dealing with earthquakes felt in mainland France, among which 62% are not listed in the SISFRANCE database. All this new information will be analyzed in the frame of the SISFRANCE consortium for updating the database. The approach is now being improved, mainly concerning data-mining techniques, but given the success of the first application, the exploration of other websites is already planned (RetroNews, LecturaPlus). Although the proposed methodology aims at facilitating source findings, it does not preclude the expertise of the historian for analyzing and interpreting new sources.