Geo-historical analysis of flood impacts in Alpine catchments

Abstract. In France, flooding is the most common and damaging natural hazard. Due to global warming, it is expected to globally exacerbate, and it could be even more pronounced in the European Alps that warm at a rate twice as high in the Northern Hemisphere. The Alps are densely populated, increasing exposure and vulnerability to flood hazard. To approach long-term evolutions of past flood occurrence and related socio-economic impacts in relation to changes in the flood risk components (i.e. hazard, exposure and vulnerability), the study of historical records is highly relevant. To this aim we analyze the newly constituted database of Historical Impacts of Floods in the Arve Valley (HIFAVa) located in French Northern Alps and starting in 1850. This database reports for the first-time flood occurrences and impacts in a well-documented Alpine catchment that encompasses both a hydrological and societal diversity. We analyze past impacts in regard to their characteristics and evolution in both time and space. Our results show an increasing occurrence of impacts from 1920 onwards, which is more likely related to indirect source effect and/or increasing exposure of goods and people rather than hydrological changes. The analysis reveals that small mountain streams and particularly glacial streams caused more impacts (67 %) than the main river. While increase in heavy rainfall and ice melt are expected to enhance flood hazard in small Alpine catchments, this finding calls to pay a particular attention to flood risk assessment and management in small catchments.



45
On the mainland French territory flood is the most common and damaging natural 46 hazard in terms of economic cost and number of municipalities concerned (Ministère de la 47 Transition écologique, 2020). In highland regions, these events can be caused, among others, 48 by summer thunderstorms, rain on thaw saturated soils, rain on snow or glacial lake outburst 49 (Merz and Blöschl, 2003). The topography induces flood events with highly contrasted 50 dynamics; from sudden events with large sediment transport in the upstream small catchments 51 to multi-day events flooding large parts of the valley floor. This diversity of hydrological 52 dynamics adds to the complexity in flood risk management. Furthermore, climate change is 53 expected to increase extreme precipitation (Min et al., 2011) that could in turn increase flooding 54 (Gobiet et al., 2014;Blöschl et al., 2020). This is especially the case for the European Alps 55 where an increase in summer heavy rainfalls (Giorgi et al., 2016;Ménégoz et al., 2020) may 56 threaten densely populated mountainous valley especially exposed and vulnerable to climate 57 extremes (IPCC, 2019). With its long history of flooding, the densely populated Arve valley 58 located in the Northern French Alps is indeed prone to experience the latter effects of global 59 warming in the future. 60 Historical records constitute a source of reliable data to characterize past hydrological 61 events because they contribute to give a comprehensive representation of these events and 62 of their changes over long time scales in spite of the lack of instrumental data (Garnier and 63 Desarthe The historical analysis of past events is useful for the study of catastrophe as we can 67 hypothesize that these remarkable events are etched in the community's memory 68 (Papagiannaki et al., 2013). Indeed, it is because these events have impacted the society that 69 they are recorded in the historical records, i.e. have left a "social signature". Those high impact 70 events can come close to the notion of a catastrophe as they can lead to a societal upheaval 71 (Soanes and Stevenson, 2009) sometimes deleterious but also beneficial (behavioral change 72 promoting prevention) (Garnier, 2017). High impact events are by nature rare, often resulting 73 in a lack of available data (e.g. description of the event, time, extent, damages caused etc.). 74 However, historical approach allows a social and spatio-temporal contextualization of the data 75 (Giacona et al., 2017), making the reconstruction (date, impacts) of major flood events possible 76 (Barriendos et al., 2003(Barriendos et al., , 2019 and attesting the social apprehension of the phenomenon (Gil-77 Guirado et al., 2016). 78 Numerous historical databases were built to document past flood occurrence and 79 magnitude, such as the Prediflood database (Barriendos et al., 2014), and some, as the 80 database from Thoumas (2019), allow to analyze the climatic fluctuations. In contrast to these 81 latter databases focusing on hydrological events, some databases collected the 82 socioeconomic impacts of floods such as the APAT database (Lastoria et al., 2006), the press 83 database on natural hazards and climate change from Llasat  and management tools and discuss the trajectories of vulnerability. 89 Floods, as natural hazards, are physical phenomena naturally occurring and can, when 90 certain conditions are met, cause harm to societies. They can be interpreted as a social 91 construction (Beck, 1992)  to record them, provides a subjective measure of the events that were considered worth 96 reporting for various reasons across historical periods. Flood impacts result from the 97 interaction between the natural phenomenon and the dynamics of exposure and vulnerability. 98 As vulnerability we understand the inclination to damage of various exposed goods, activities 99 or people constituting a given territory (Leone and Vinet, 2006). We consider the vulnerability 100 as a dynamic system articulated to numerous physical and societal factors (Antoine, 2011).

101
This system can evolve in time and space (Cutter, 2003). Major natural disasters, such as 102 floods, are often displayed as unforeseeable events whereas the historical facts give evidence 103 of the contrary (Garnier, 2016(Garnier, , 2019). Yet the society's vulnerability may increase as the past 104 disasters are forgotten, leading to a "society of risk" (Garnier, 2019). Historical approach allows 105 to explore the trajectories of hazard and vulnerability in response to changes in climate, land 106 use and flood risk management (Gil-Guirado et al., 2016). 107 108 The present paper introduces a newly constituted database of flood impacts of the Arve 109 River and its tributaries (Northern French Alps). The database called "Historical Impacts of 110 Floods in the Arve Valley" (HIFAVa) covers all impacts caused by hydrological events that 111 occurred since 1850. 112 The study of this database, the first one documenting a mountainous catchment, ultimately 113 aims at analyzing the interactions between social and natural dynamics engendering flood 114 impacts. In this paper we analyse the impacts with respect to their nature and evolution in both 115 time and space. 116 117 118 2. Study area: the Arve River.

2.1.
Description of the physical setting of the Arve River.

121
The Arve River is located in the Northern French Alps (Figure 1), flowing from the high 122 elevations of the Mont-Blanc summit (4810m a.s.l.) to the Swiss lowlands (330m a.s.l.), where 123 it flows into the Rhône River. The surface area of its catchment is 2164 km² with the largest 124 part higher than 1000m a.s.l.. The main tributaries of the Arve River are the Giffre, the Borne, 125 the Menoge and the Foron Rivers. Since 1850, i.e. the start date of the studied period, the 126 Arve River is already almost completely embanked (Mougin, 1914;Gex, 1924;Peiry and 127 Bravard, 1989). 128 Due to large difference of altitude between high and lowlands, the Arve flows can be defined 129 by two hydrological regimes following an upstream to downstream continuum: 130 -The upstream part of the catchment (down to the city of Sallanches; Figure 1), has a 131 glacio-nival regime due to the numerous glacial tributaries lowing from the Mont Blanc 132 massif (Viani et al., 2018). Low flows occur in winter and early spring (December to 133 March) and the high flows in summer (maximum in July and August) because of the 134 strong contribution of ice melting (Bernard, 1900). Floods mainly occur in summer due 135 to the synchronicity of both ice melt and intense subdaily rain storms. In this part of the 136 catchment, the flood plain is narrow and the slope inclination is high.

137
-At lower elevations (i.e. downstream Sallanches) the regime becomes more and more 138 nival downward. Low waters mainly occur in winter and reach the highest levels 139 between late spring and early summer with the snowmelt. Between Sallanches and 140 Bonneville, floods mainly occur in summer and autumn due to the conjunction of 141 intense daily rain storm, snow melt and, in a lesser extent, ice melt contribution.  Highway the Arve valley is a major trans-Alpine route connecting France and Italy.

161
The socioeconomic setting of the valley follows an upstream-downstream distribution 162 pattern. The period from 1850 to 1913 experienced a great touristic development (thermal bath 163 of Saint-Gervais and mountaineering in Chamonix). The economy around Chamonix is 164 essentially based on mountain tourism. In 1921, 250000 tourists were visiting Chamonix each 165 year (Gex, 1924). In 2015 the lodging capacity in the valley reached around 416400 equivalent 166 touristic beds. This part of the valley has undergone a rapid urbanization. In 1804, the 167 discovery and exploitation of spring water for hydrotherapy in Saint-Gervais (Gex, 1924)

Material and methods of the HIFAVa database.
179 180

3.1.
Collecting data from historical archives.

181
A first historical study has been carried out to draw a flood chronicle of the Arve valley  182 between the 18 th and the 21 th century, with a particular focus on the hydro-meteorological 183 circumstances of the flood events (Mélo et al., 2015). As sources are more abundant over the 184 last 165 years (1850-2015), this period was defined as the studied time frame of the HIFAVa 185 database. Only events mentioned in at least two sources were integrated in the database. 186 Since 2015, data have been further collected to complete the preliminary dataset from Mélo et 187 al. (2015).

188
Information on floods and related impacts has been collected from various sources. Primary 189 sources range from handwritten archives like municipal acts to departmental archives (e.g. 190 reports of the Préfecture and of town councils). Secondary and tertiary sources are 191 respectively made of published documents (newspapers, reports, books) and pre-existing 192 databases. The database of historical records providing a chronological and synthetic layout 193 of the data is composed of ( Table A1) 1914. They mostly correspond to analyses of the regional hydrology (Mougin, 1914;204 Rousset-Mestrallet, 1986) but they can also focus on single hydrological events (  sometimes provide instrumental data and illustrations. They also inform about the 216 public authority response and the past ongoing discussion. 217 Other records:

218
-The national database of historical flooding (Base de Données Historiques sur les 219 Inondations : BDHI) that gathers floods events considered as "remarkable" in the 220 French territory (Ministère de la Transition Ecologique, n.d.; Boudou, 2015) has been 221 used. HIFAVa. The RTM database was built to assist the management of small tributaries.

230
-A movie realized in 1990 by the RTM is also mentioned as a source.

231
-Some records are from the Syndicat Mixte d'Aménagement de l'Arve et de ses 232 Affluents (SM3A), which is the institution in charge of the management of the Arve River 233 and its tributaries since 1994. 234 235

3.2.
Characteristics of the HIFAVa database. 236 The database has been built using the free and open-source relational database 237 management system PostgreSQL and is accessible through its package pgAdmin. 238 HIFAVa contains 916 distinct flood impacts caused by 321 flood events. The primary key is 239 the impact ID. Therefore, each impact is recorded as a unique line and described through 240 various variables (Table 1 and Table A2). The river that triggered the flood is mentioned when 241 possible (94% of cases). For instance, no river name has been attributed to the impacts related 242 to overland flow in January 1979. The accuracy of the impact location varies from specific 243 addresses (house, bridge, neighborhood) to the municipality scale. When the source is not 244 accurate enough to distinguish distinct locations of several impacts, they are all referenced 245 under a unique impact ID. In other words, sometimes an event caused numerous impacts 246 registered under distinct ID because it was possible to localize each impact precisely (at the 247 hamlet scale) and sometimes we can only localize the impacts at the municipality scale so they 248 are register under the same ID. The severity of an event can not be estimate by the number of 249 ID registered in the database. The most recent sources are often highly informative, allowing 250 impacts to be more precisely located. 251 Impacts occurring the same day on a given river are expected to be caused by the 252 same flood event. As a result, the date is the key used to connect each impact to a flood event. 253 This "flood event" definition has been extrapolated to impacts occurring the same day in 254 different catchments, assuming that two impacts occurring the same day can be caused by the 255 same hydrometeorological event given the moderate surface area of the Arve catchment. The 256 accuracy of the date is rated on a certainty scale (hour, day, month, year). Based on 257 information contained in the records, we distinguish when possible the hydrometeorological 258 events (e.g. rainfall, intense and short rainfall, melting of snow, frozen soil, glacial outburst, 259 wet period before the event) which caused the flood and the different flood types (e.g. river 260 flooding, overland flow, sediment transport) leading to the impacts. 261 262

263
The flood impacts of the HIFAVa were categorized through a text quantitative content 264 analysis with the KH Coder software (Higuchi, 2015). The description of the impacts comes 265 from comments contained in the records. The most frequent words have been gathered in 266 order to determine representative categories of the database content. A category is made of 267 several words assigned to a coding rule. Categories have been inspired by the flash flood 268 impact severity scale of Diakakis (2020). This analysis has led to the following seven 269 categories with example of the assigned words: 270 -Transport network: e.g. "road", "bridge", "railway", "street".

279
A more in-depth analysis will be conducted later on to define severity classes based on the 280 nature of the impacts. This future analysis will be based on the work of Barriendos

4.1.
Evolving sources over time.

292
During the studied period, the diversity and the quantity of sources in which mentions 293 of impacts have been found fluctuate (Figure 2). Among the existing databases used, the BDHI 294 database for instance continuously covers the studied period but was sporadically informative 295 since it only contains two mentions of flood impacts in the Arve catchment, respectively in 1892 296 and in 1987. By contrast, the SM3A database appears later (1979) in the studied time frame. represents 58% of the sources describing the impacts. 308 One of the evolutions of the sources is the increase in the newspaper articles mentioning flood 309 impacts. Following the 1881 press freedom French law, the 1880-1889 decade marks the 310 emergence of articles recording natural hazards, such as flood impacts (Ferenczi, 1996). 311 However, the 1855 flood in Bonneville was already reported by the swiss newspaper, Le 312 Journal de Genève. 313 Although a few sources (e.g. the municipal archives of Sallanches) remain to be examined, 314 most of them have been analyzed in order to constitute the database. Hence, we consider that 315 we have a comprehensive view of past flood impacts since 1850 over the whole Arve 316 catchment. 317 such a large number of impacts become more frequent (1930, 1940, 1944, 1968, 1979, 1987, 322 1990, 1996, 1997 and 2007) and the total amount of impacts per year reaches 54 in 1996 323 (Figure 3). The decennial moving average of the impacts' number highlights an overall 324 increase over the 165 years, punctuated by periods with less frequent impacts (in 1910-1923, 325 1950-1960 and 1975-1980). 326 Besides, the number of recorded flood events stays relatively stable between 1.5 and 3 events 327 per year on average until 1990, then it rises up to 4.5 events per year. Therefore, the overall 328 increase in recorded impacts seems partly disconnected to changes in flood occurrence. Only When analyzing the spatial distribution of the flood impacts, we can see that they are 338 spread over the entire catchment ( Figure 4). They are, however, mainly gathered in the Arve 339 valley around Chamonix and Bonneville (24 and 12,5% of total impacts recorded in the Arve 340 catchment). These high numbers may be due to the fact that these towns are both among the 341 most densely populated and the closest towns to the Arve River. The impacts caused by the 342 Arve River floods represent 33% of all recorded impacts, and its two main tributaries, the Giffre 343 and the Borne Rivers, have only caused 8% of the recorded impacts. In fact, most impacts are 344 due to small torrential streams (53%). Among them, almost a third is related to glacial 345 tributaries, while these tributaries are localized only in the uppermost part of the catchment 346 near Chamonix. For instance, small torrential tributaries such as the Arveyron, the Grépon (left 347 bank tributary close to Chamonix) or the Bon Nant have caused alone more impacts than the 348 Borne River itself.  1900 1905 1910 1915 1920 1925 1930 1935 1940 1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995  category "non attributed" corresponds to the part of impacts for which it was not possible to 353 attribute a river, either because events are related to overland flows or because of the location 354 between two or more rivers.

356
The Arve tributaries produced disasters characterized by numerous and major flood damage. 357 Among them, the 1987 Borne River flooding in its uppermost part washed away the municipal 358 campsite of the village of the Grand-Bornand causing 23 casualties and heavy economic 359 losses (Meunier, 1990). In addition, the 1892 glacial lake outburst from the Tête Rousse glacier 360 in the Bon Nant River (which literally translated means "Good Stream") swept away the thermal 361 bath of Saint-Gervais ( Figure A1) and 33 houses causing at least 175 casualties. The glacier 362 was drained in 2010 and is today closely monitored to avoid such a brutal and disastrous 363 natural event (Garambois et al., 2016). 364 All these high impacts events are due to sudden, highly-dynamic summer floods of tributaries, 365 often aggravated by large sediment transport. Some towns located along the Arve Riversuch 366 as Sallanchesare more prone to tributary's floods because embankments have been built 367 and efficiently prevent impacts from the Arve flooding. In contrast, there are very few impacts 368 recorded in high altitude, probably due to the sparse human settlements. To decipher the potential source effect in the increase in impacts since 1960, maps of 378 the impacts by sources have been drawn for the periods before and after 1960 ( Figure 5). For 379 the first period (1850-1959), three main sources describe 64% of the impacts (literary records 380 28%, RTM 18% and departmental and municipal archives 18%) and for 29% the information 381 comes from more than one source. The impacts are mainly gathered along the Arve and the 382 Giffre Rivers, especially in the valley of Chamonix and between the towns of Cluses and 383 Bonneville.

384
For the second period , the RTM reports 65% of the impacts, and 20% come from 385 multiple sources while departmental and municipal archives and the PPR/PPRI describe 5% 386 each. Information coming only from literature decreases substantially (122 described impacts 387 in the first period to 3 impacts in the second), SM3A records start in 1979 and only document 388 the Giffre and the Bon Nant Rivers. The distribution of the impacts is much more scattered 389 across the whole catchment than during the first period. The impacts are not anymore gathered 390 along the Arve River, since most of them result from small tributaries. Impacts described by 391 more than one source are located in the valley of Chamonix and around Bonneville, probably 392 because these economic and touristic centers arouse interest of many sources (newspapers, 393 departmental and municipal archives and RTM). The touristic specificity of valley and its 394 international stature can explain the media coverage. We can assume that, floods are more 395 likely to be reported in newspapers as when they happen in a location known by the reader. In punctually increase from year to year, somehow mirroring the population growth ( Figure 6). 457 Increasing exposure due to population growth and urbanization may then explain the 458 increasing number of impacts. One can, however, notice the decrease in impacts after the 459 disastrous 1996 flood event. This is due to the heightening of the dikes after the 1996 flood. In 460 Bonneville, the link between the number of impacts and the population growth is not clear.

464
The quantitative analysis of text content reveals the distribution of the impact categories 465 by river and illustrates the diversity of the catchment in terms of land use and economic 466 development (Figure 7). This analysis of text content is particularly relevant because it allows 467 to overcome the database scarcity of quantitative information on the severity of the flood. The number of impacts has been almost multiplied by two since 1920. Hence, the mentions of 516 impacts on natural environment for the period 1850-1930 were multiplied by more than two 517 compared to the period 1960-2015. During the first period, mentions of impacts on natural 518 environment refer mainly to forest, field and crops, while after 1960 there is no mention of field 519 or crops and most of the mentions are about gullying, deposition of sediments and banks. This 520 is in agreement with the evolution of the land use due to the demographic growth, i.e. the 521 observed vanishing of forest areas and cultivated land to the benefit of urbanization ( Figure 6). 522 Yet, mentions of impacts on urbanized areas during 1960-2015 has been multiplied almost by 523 four compared to the 1850-1930 period. This agrees with observed changes in land use 524 characterized by urban sprawl. 525 526 5. Conclusions. 527 We collected historical information from 1850 onwards to document flood impacts in 528 the Arve valley, an Alpine catchment characterized by a high hydrological and socioeconomic 529 diversity. The analysis of the HIFAVa database led to acknowledge the rise in the number of 530 impacts starting in 1920 and well-marked from 1960. This rise does not seem to be related to 531 increased flood hazard since it does not follow changes in flood occurrence, except partially 532 for the latest period . Instead, more frequent impact could be explained by 533 increased exposure related to the demographic growth (tourism and economic attractiveness 534 of Geneva) and/or by the evolution of the sources, in particular the emergence of the RTM. 535 There are two main types of flood events causing impacts in the catchment, e.g. floods related 536 to the main river and those related to the smaller mountain streams. Floods from these small 1968 flood affecting a large part of the Arve catchment is an exemplary case of this flood type. 543 The analysis of the nature of the impacts does not reveal a clear evolution over the last 165 544 years. However, further work is required to define more detailed categories allowing to 545 question the evolution of the assigned words since 1850. Additional investigations of the 546 municipal archives and interviews of local flood risk managers should also be undertaken to 547 allow answering some key questions, such as the definition of high-impacts events in this very 548 valley. The processes of risk memory transmission and the evolution of the local political risk 549 management will be queried for the purpose of understanding the evolution of the social 550 vulnerability.