During the inventory work carried out for the Flood Directive in 2011, we selected a total of 176 major floods in France since 1770 (see Lang and Coeur, 2014) based on the following considerations: diversity of flood types, strong flood hazard or spatial extent, and important socio-economic impacts, in addition to reference events used in planning documents (flood mapping area) or last significant flood in living memory. Using a multidisciplinary methodology, we established an evaluation grid based on three main features (Boudou et al., 2015): (1) flood intensity (score between 3.5 and 14) according to several criteria (return period of maximum peak discharge; duration of submersion; dyke breaches or log jams); (2) flood severity (score between 3 and 12), with two main indicators: flood damage (number of fatalities, economic loss) and social, media or political impacts of the event (establishing a new risk policy, calling for international solidarity to face the crisis, etc.); and (3) spatial extent of damage (score between 2 and 8). This grid allowed us to rank the 176 major floods (Boudou, 2015). Then, a second level of selection led us to focus on the nine events shown in Fig. 1 (January 1910, March 1930, October 1940, December 1947/January 1948, December 1959, January 1980, November 1999, December 2000/April 2001 and February 2010). These flood events cover all flood typologies (oceanic/snowmelt/Mediterranean floods, storm surges, cyclones, dam breaching) and are considered as some of the most remarkable in accordance with the evaluation grid. Lang et al. (2012) presented the main characteristics of these nine events (except for the 1947–48 flood).

In this study, we investigate the two oldest selected events, which took place in January 1910 and March 1930, focusing on the urban situation in Besançon and Moissac (Fig. 2). The aim is to focus on two cities that have been significantly flooded in the past and to understand how their vulnerability to flooding has changed up to the present day. A detailed inventory of documentary sources for these two events can be found in the Supplement.

The flood of January 1910 ranks fifth among the nine floods selected as remarkable according to the evaluation grid (Fig. 1). This flood event is mostly known for being the most significant flood affecting the city of Paris, with a return period of about 100 years for several rivers of the Seine Basin. After a very wet end during the year 1909 (450 mm of rainfall in 3 months), the Seine Basin received a large amount of rain and snow in January 1910 (about 300 mm in the upper part, 110 mm in the central part, and 280 mm in the downstream part). The water level at Paris Austerlitz was 8.66 m, the second highest historical level after the flood of February 1658 (8.80 m) (Champion, 1858–1864; Goubet, 1997). There was a relatively small number of direct fatalities (seven deaths) plus nine indirect deaths (collapsing of several cavities), but the impact within the Paris region was extremely high, with 150 000 persons affected and economic losses of about FRF 400 million (EUR 1.5 billion, 2015) (Picard, 1910). Despite the fact that a large part of northern France was also affected, most of the attention of society and recollections of this event have been focused on Paris. To demonstrate the remarkability of this event, not only for the Seine catchment area but also for more rural regions, we concentrate our study on the Doubs Basin where the flood of January 1910 remains one of the most significant historical floods, with the highest water level being recorded in the city of Besançon (see Fig. 3, e.g. Z = 245.55 m at Poterne, Place la Revolution). While the flood event across the Seine Basin was characterized by a clustering of several oceanic rainfall events, the flood event in the Doubs Basin was triggered by an episode of heavy rainfall from 18 to 21 January (between 150 and 250 mm), plus the presence of extensive snow cover after a wet winter which led to significant snow melting. A large part of the old city of Besançon was flooded, with huge damage. Many shops, houses and their basements were inundated, causing important losses of furniture. The streets were also particularly badly affected due to the high flow velocity. In total, the cost of the flooding at Besançon is estimated at around FRF 2 million (DREAL Franche-Comté, 2010), corresponding to EUR 7.7 million in the present day.

According to several documentary sources (Allard, 1910; Ministère de l'Ecologie, 2011), it appears that the hydro-meteorological conditions of the event (peak discharge at Besançon of about 1750 m3 s-1, with a return period of about 100 years; catchment area of 4379 km2) cannot explain why the flood level was so high throughout the old city. Such exceptional water levels in the city centre were the consequence of energy losses at the bridges of the town. These energy losses were larger than usual (cf. Fig. 3, in comparison with the 1882 and 1896 flood events) due to a log jam (about 35 000 m3), resulting from the inundation of a paper factory a few kilometres upstream of Besançon, contributing significantly to a raising of the water level.

Archive sources (especially administrative reports produced by the chief engineer of the Ponts-et-Chaussées, Serial S, Doubs departmental archives) also reveal some major failures of flood warning during the event. Surprised both by the arrival and the intensity of the flood, the local authorities did not succeed in setting up temporary protective structures at the different open city gates (“postern gates”), which directly contributed to the inundation of the city (Fig. 4).

At the end of February 1930, an intense Mediterranean rainfall event occurred in the south-west of France, with hot and moist air from the Mediterranean Sea penetrating deep into the Massif Central highlands. From 25 February to 4 March, a large area was affected by heavy rainfall (e.g. more than 200 mm over 6000 km2 during 4 days), with a maximum of 694 mm in 7 days at Saint-Gervais-sur-Mare (spring of the Orb river). The very serious adverse consequences of this rainfall event can be explained by at least two factors. From October 1929 to February 1930, high rainfall totals were observed (e.g. 1177 mm at Lodève, 840 mm at Florac), thus favouring a strong reaction of the basins which were already saturated. Moreover, a warming in temperature associated with intense rainfall was causing a large amount of snow melting (20–100 cm) above 600 m.

Due to its intensity and unusual date of occurrence (at the end of a wet winter) the rainfall event triggered an exceptional flood event (Pardé, 1930). The following flood hazard intensity can be judged exceptional for the downstream part of the Tarn catchment (8000 m3 s-1 at Moissac, 15 400 km2; mean annual discharge 230 m3 s-1), with a return period of about 250–300 years (Dreal Midi-Pyrénées, 2014). Between 210 and 230 fatalities were recorded during this Tarn River flood (resp. Bichambis, 1930, and Boudou, 2015), which represents one of the most destructive flood events ever recorded in France and surely the most significant during the 20th century. The economic loss for the entire surrounding region was estimated at around FRF 1 billion, which corresponds to EUR 570 million in 2015 (Journal Officiel de la République Française, 1930).

One of the striking features of the disaster can be found in the concentration of damage in the town of Moissac (120 deaths out of a total of 210). Reconstructing and mapping the flood chronology using historical sources provides us with a better understanding of the circumstances of the disaster (Fig. 5). On 3 March 1930, the flood arrived in the town. Before 18:30 LT (local time) the Tarn River was already overflowing the main channel, on both the south and north banks. Fortunately, the town centre was protected by three main dykes and the railway line embankment. From 18:30 to 23:00 LT, the water level rose and the flood extent covered the area between the main dikes at the eastern part of the town. Around 23:00 LT, at the time of maximum discharge (estimated at around 8000 m3 s-1), three breaches suddenly appeared along the railway embankment. These breaches led to a sudden outburst of the dykes and final inundation of the town.

According to the locations of fatalities and the feedback of information on the disaster, the explanation of the high death toll is twofold. Firstly, the rapid influx of water into the city due to the flash flood and dyke failures induced a surprise effect on the inhabitants of Moissac. Secondly, the collapse of more than 600 houses was related to the typical kind of housing in this region, being built of raw bricks especially vulnerable to flooding and sustained contact with water.

One of the main requirements of the Flood Directive is to identify areas with a potential high level of flood risk, based on historical floods that would have significant adverse consequences if they occurred again. As the consequences are dependent on the flood hazard as well as the personal, social and economic assets located in the flood risk zones, one of the main concerns is to assess the changes in local vulnerability of city centres as a function of time. In both case studies, the main casualties and/or economic losses within the catchment were located in a single municipal area. But some aggravating factors are time dependent, such as woody debris upstream of bridges at Besançon or dyke failures to the east of Moissac. Other aggravating factors are related to social vulnerability, such as failure of flood warnings at Besançon or vulnerable building materials at Moissac.

To obtain a better understanding of the local disaster process, our study aims to monitor changes in flood vulnerability, comparing the past and present situations. Several questions have to be addressed. Is it possible to assess correctly the changes in vulnerability over time according to the available sources? Does the mapping of land use provide enough information to identify indicators of vulnerability? Can we establish scenarios concerning the impact of a future flood based on a historical flood?

After a preliminary analysis that involves georeferencing historical information in the present-day context, we then consider the mapping of land use and estimating the population at risk, while comparing the past and the present situations.

A preliminary step of this study consists of carrying out dynamic mapping with a spatial display of the previously collected historical information. The historical corpus made up of various document formats and sources is included in a GIS by locating the information available. However, some place names have changed since the date of the flood event, thus requiring supplementary treatment of the data.

The dynamic consultation of historical information is not only of interest for correctly locating the various sources of information on flood vulnerability, it can also be used to develop risk awareness and risk culture on an exposed territory. As an example, the high-water mark inventory developed for the Seine River catchment (www.reperesdecrues-seine.fr/carte.php) provides dynamic mapping which is easily understandable and interactive for the general public, in contrast to the maps resulting from hydraulic or hydromorphogenic modelling (de Moel et al., 2009).

In this section, we address the exposure and susceptibility to flood risk (Fig. 6) using simplified descriptors which remain consistent with the level of data availability and accuracy of historical information (Barnikel and Becht, 2003; Barnikel, 2004).

Firstly, the exposure analysis is based on the changes in the population living per building and provides information about the evolution of built-up areas. Secondly, susceptibility analysis based on land-use classification provides relevant information to evaluate the nature of buildings affected during flooding. Historical information is required which at least describes the land cover on different dates. For example, historical maps and aerial photos often depict the built-up territory for a specific year.

To perform a spatial analysis of historical maps, it is necessary to integrate them into a GIS. Three steps are executed: scanning, georeferencing, and digitization supported by a spatial reference system (Fig. 6a) (Rumsey and Williams, 2002; Levin et al., 2010). A set of historical maps and aerial photographs produced by the French National Institute of Geographic and Forest Information (IGN) are used to depict the extent of built-up areas at the scale of a block of houses. A total of seven topographic maps (from 1911 to 1988) are used for Besançon and 26 aerial photographs for Moissac (from 1947 to 1983). Aerial photographs are favoured in the case of Moissac because of the inconvenient representation of the town on topographic maps, which is split between four map plates. These raster data are then imported and georeferenced. A spatial database (BD TOPO) produced by the IGN, describing the present French territory and its infrastructures, is used to select control points and evaluate distortions during the digitizing step. During this last step, information from topographic maps is vectorized into a unique “historical layer”. In this way, each object is given a spatial reality (via the GIS representation) and a temporal reality (by associating a temporal field to indicate its existence for a specific year). Consequently, the “historical layer” allows us to obtain “temporal snapshots” (Langran and Chrisman, 1988; Gregory and Healey, 2007) of the urban fabric: the space is discretized based on information available at the time of the event.

Subsequently, the description of “historical layer” objects provides information on the nature of building exposure. A land-use classification is drawn up based on a nomenclature adapted from the Urban Atlas of the European Environment Agency (http://www.eea.europa.eu/data-and-maps/data/urban-atlas), according to historical information constraints (Fig. 6b). A first geomatic processing step is performed to discretize the residential buildings on a 0.25 ha grid. A density criterion is applied in each grid cell, based on the percentage contribution to the building footprint, leading to a distinction between dense and sparse areas. To enhance the classification, a second processing step is carried out using a proximity criterion for each building based on the number of buildings within a 200 m radius (continuous and discontinuous buildings). Local information is then added related to the location and nature of non-residential constructions. BD TOPO data are used to describe the current situation, and a point-in-time layer is built with our “historical corpus” information for earlier historical periods.

General information is provided by the evolution of population at the scale of the municipality. Figure 7 presents the data derived from several population censuses during the 20th century. It shows than the number of inhabitants has grown by about +100 % at Besançon (from 57 978 to 116 914, between 1911 and 2010) and +60 % at Moissac (from 7814 to 12 354, between 1911 and 2006). As only part of the built-up area was affected by floods, especially in the case of Besançon, it is necessary to cross two layers of information: the number of inhabitants per small block and the spatial extent of the historical flood (1910 or 1930 floods at Besançon and Moissac, respectively).

Human exposure is taken into account by census or an estimation of the resident population. The aim here is to distribute the raw demographic data throughout the blocks of houses by following its evolution at different scales (Wu et al., 2008). The maps so produced can shed light on the evolution of human exposure within the area affected by the flood.

To assess the current population living within the flood extent, we make use of two demographic data sets produced by the French National Institute for Statistics and Economic Studies (INSEE), applying Eq. (1) to redistribute the population data at the scale of blocks of houses. The first data set is defined at infra-municipal scale with IRIS (infra-urban statistical area) data. The second data set is based on an estimation of the fiscal population within a 200 m × 200 m grid. These data sets are distributed at the scale of residential blocks of houses, based on a volumetric method (Lwin and Murayama, 2009), in proportion to the building footprint area multiplied by the vertical density, using the building height provided by BD TOPO: developedarea=buildingheight×buildingfloorareaaveragestoreyheight. Historical information, in the form of a census or raw demographic data, is required to estimate (Ekamper, 2010) the numbers of the population exposed at the time of the disaster. General census reports are available for every French municipality (sometimes online), generally compiled every 5 years up until 1946, with some exceptions. These documents contain nominative information about the municipal population, grouped by building and street, at different dates. The comparison between past and present exposed population within the flood extent should take account of possible changes of census methodology over time.

Figure 8 displays the land use within the area affected by the 1910 flood in Besançon, based on the situations in 1911 and 2013 (resp. dates of two censuses). No significant change can be seen in terms of vulnerability, according to the spatial extent of the built-up area. Since the centre of Besançon is located within a meander of the Doubs River, with no opportunity for spatial expansion or urban densification, there has been no increase of exposure, apart from the hospital area. Although the city has experienced a spatial expansion towards the north, on the right bank, this area is located outside our zoning at a larger scale.

According to the land-use classification, we can note significant changes in the various activities. There has been a fall in military employment, in favour of an increase in administrative and public facilities. While military areas have decreased by 74 % between 1911 and 2013, administrative areas have grown by a factor of 12. A reduction of human exposure is noticeable between 1911 (the census year closest to the 1910 flood) and 2013, with a 24 % decrease in the city-centre population.

The demographic evolution is represented on Fig. 9 at the scale of a block of houses, reflecting the decrease in household size (decline in the number of inhabitants per building) and a decline in residential function (reduction of inhabited buildings within the city centre).

The flood risk mapping of Moissac yields an opposite diagnosis, with a major increase of vulnerability within the area affected by the 1930 flood (Fig. 10). Built-up areas have expanded by 122 % between 1930 and 2013. Such spatial extension is explained by new residential development (mainly housing estates) and economic buildings east of the city centre and by a progressive densification of the low-density area on the south bank flood plain.

Despite a new distribution of the population (Table 1), the human exposure has not changed significantly. The reduction of population density in the city centre is compensated by a spatial expansion (Fig. 11). The human exposure has mainly increased on the east side of the city centre, especially in the area located between the two levees. It should be noted that no general census report is available for Moissac in the 1930s. Therefore, the population exposed to flood risk in 1930 was estimated from a raw demographic data set, obtained from an internet database containing a historical population census at the municipality scale (http://cassini.ehess.fr/), which was then distributed according to the volume-based method.

These two case studies shed light on the complexity of flood-risk evolution. At the nationwide scale, it is clearly acknowledged that the increase of flood damage over the last few decades is induced by a general increase in flood vulnerability (Kron, 2002; Luino et al., 2012; Kundzewicz et al., 2014; Smith et al., 2014). At a local scale, where topographic, social, and economic contexts are crucial, it is necessary to have a more detailed analysis.

In Besançon, there has been no extension of the urban area within the old city since 1910, but significant land-use changes have led to a decrease of flood vulnerability as some previously residential areas are now used as administrative buildings. The frequency of flooding has changed in the historical centre, due to the establishment of safety measures, especially with the construction of mitigation structures such as cofferdams to close the postern gates. Some uncertainties remain for determining the flooded area in the case of an event comparable to the 1910 reference flood, since opposite effects come into play. The log jams at the bridges are not expected to be repeated, but additional hydraulic losses have been introduced by new hydraulic structures since 1910. Nowadays, the reference flood selected in the regulatory documents is a simulated flood larger than the January 1910 flood.

In Moissac, the changes in vulnerability show a more contrasted pattern. As in various other French regions, the built-up areas have grown in spatial extent since 1930, characterized by an important development of housing estates. One critical point is the development of one-storey buildings, leading to a higher human vulnerability due to the lack of a refuge floor. On the other hand, building quality has improved. During the 1930 flood, the house collapses in Moissac and the consequent fatalities were closely related to the construction materials used. To increase the resistance of the structures, new materials and building techniques were used during the reconstruction stage. Another positive change is related to the improvement of safety measures, due to progress in flood-warning decision-making as well as in regards to emergency population evacuation schemes implemented by the civil protection services. The 1930 flood in Moissac, with a return period estimated at around 250 years, is nowadays considered as the reference flood hazard for the local flood risk management strategy as well as for planning and development documents. This territory appears to remain vulnerable, especially to risks of dyke failure.