2.1 Study area: The city of Seville and its relationship with the floods of the Guadalquivir River

The Guadalquivir River is the most important river in the southern part of the Iberian Peninsula (657 km long). It crosses Córdoba and Seville (Fig. 1) and discharges in Sanlúcar de Barrameda, next to Doñana National Park. The Guadalquivir River is situated in the southwest of Spain, where there is a Mediterranean climate influenced by the Atlantic Ocean (^{García-Barrón et al., 2013}).

The city of Seville (37.38° N, 5.97° W) is built on an alluvial plain. Historically, it was a walled city surrounded from the northeast to the southwest by the Guadalquivir River (Fig. 2). This river separates the traditional neighborhood of Triana from the remainder of the historic urban center of Seville. Therefore, the extramural neighborhood of Triana was connected to the intramural area of the city of Seville by a floating (or pontoon) bridge that crossed the Guadalquivir River (Fig. 2). This floating bridge, originally built in 1171 A.D., consisted of a set of tall boats anchored to the bottom, joined together by iron chains, with wooden columns that held the platform of the bridge. By observing old lithographs, it is possible to deduce that the distance between the water surface and the platform of the bridge was no more than two and a half meters. It was reported that the bridge was damaged several times and even ripped and broken apart by the current in 1794 (Palomo, 1878).

Fig. 2 View of Seville with the Guadalquivir River in the 17th century. This map shows the floating bridge or bridge of boats at the lower left side and the Torre del Oro at the right end of the harbor. Source: detail of the enlarged facsimile version of Braun and Hogenberg, Civitates Orbis Terrarum (1588). It can be found online at the National Geographic Institute of Spain: https://www.ign.es/web/catalogo-cartoteca/resources/html/023677.html (accessed on April 16, 2020). 

The walls of the city of Seville have historically served two defensive purposes: military and hydrological. The initial construction of the wall dates from the period of Julius Caesar (1st century B.C.). Since the Roman Age, it has been rebuilt and expanded. In the late Middle Ages, the wall was 7 km long and had 166 turrets, with 19 gates and wickets. Since the 16th century, its military purpose has been less important, and it was maintained to protect the city against the flooding river. It is the fundamental protective element of the city. During flood events, its gates were externally reinforced with caulked planks inserted in lateral guides. It is important to highlight that in 1868, after great political debate, the gates and walls were torn down with the justification that this would favor the expansion of the city. However, this would also influence the impact of floods on the city, as described in subsequent sections.

Similarly, the drain spindles of the city of Seville were closed during floods to prevent them from working in the reverse direction (i.e., introducing water from outside of the city into the walled urban center). In the river plain of the opposite riverbank, next to the neighborhood of Triana, cloisters and monasteries were built, which appear frequently in historical records as being affected by river flooding. After floods, a serious additional problem was the permanent ponding of the low areas of the river plain, where poor quality water bodies formed, with pests and the risk of disease transmission, as was reported for different periods by ^{Palomo (1878)}.

Two tributaries flowed through the west side of the city, which are currently channeled and hidden: the Tamarguillo, currently a ring road for traffic, and the Tagarete, closer to the walls (Fig. 2), which crossed under the bridge that provided access to the Xerez Gate in the southwest of the city. This location is very relevant since the height of the old city wall is a reference used by some chroniclers to alert about the danger of floods. Both tributaries increased the destructive flood effects on the city. Moreover, Guadalquivir River is influenced by the rhythm of the tides at its mouth; high tides oppose the flow of the river current, which hinders river discharge during floods.

In the 7th century, the old canal of La Vega was opened to serve as a natural drain that prevented the risk of overflow into the city and that was maintained and improved during Muslim rule. However, with the Christian reconquest (13th century), the maintenance of the canal was neglected, and its depth decreased, which increased the flood risk. During the 17th-19th centuries, important containment and diversion works were proposed, although most of them were never implemented.

In 1776 the sewerage system was enhanced and a pier was built that improved the defense of the city against floods. In 1816, different reformations were carried out along the course of the Guadalquivir River between its mouth and the city of Seville to facilitate navigation throughout the river and reduce the risk of flooding.

It is worth mentioning the severe floods of 1892 and 1895, when the city was not protected by the wall. In the 20th century, several overflows also occurred, generally with small impacts, of which there is extensive journalistic and graphic documentation. During the 19th century and especially the 20th century, hydraulic works were carried out in the river and in the flood plain, such as the cutting of meanders and canals or reservoirs (^{García-Martínez and Baena, 2006}), which have substantially modified the natural dynamics of the river and the risk of flooding.

2.2 Historical documentary sources

The present study is fundamentally based on the documentary proxy data obtained from the historical monograph by ^{Palomo (1878)} (Fig. 3), who gathered data on the historical floods of Seville and their effects on the city between the 13th and 19th centuries, and for all years with floods, he compiled all the known documentary sources. Francisco de Borja Palomo was a scholar and bibliophile, professor of jurisprudence at the University of Seville and official receiver of the city hall of Seville in the late 19th century. In addition to the detailed critical compilation of annual floods, he also included notes, biographical comments of distinguished figures, descriptions of relevant and catastrophic events (earthquakes, hurricanes, epidemics, famines, etc.) and a description of some monuments. He provided paintings with a view of the city and its relationship with the river, as well as images of the old gates of the city wall. His complex book is divided into two volumes. The first volume covers the period between the Christian Reconquista (13th century) and 1800 A.D. and compiles and expands upon the articles that were previously published by the author. The second volume is focused on the 19th century and was published later (1884), incorporating other manuscripts. In each of the years when there were reports about floods, the book includes comments on the primary documents written by chroniclers who were contemporary to the event or had direct references. Similarly, the second volume includes texts of historians who highlighted the effects of the floods with a broader perspective. We have consulted primary sources in files and historical records of different institutions that corroborate the reliability of the corresponding comments by ^{Palomo (1878)}. As an interesting example, we highlight the dissertations presented at the Regia Sociedad de Medicina y Ciencia of Seville after the floods of the 18th century, focused on the medical damages that floods can cause.

Fig. 3 Original cover of Historia crítica de las riadas o grandes avenidas del Guadalquivir en Sevilla, vol. I, published in Seville in 1878. 

Data of ^{Palomo (1878)} have been used in studies that analyze changes in the geomorphology of the Guadalquivir River during the Holocene (^{Uribelarrea and Benito, 2008}), the sinking of the flood plain as a consequence of the growth of the city of Seville (^{Ruiz-Constán et al., 2017}) and the flood risk of the Guadalquivir River (^{García-Martínez and Baena, 2006}), as well as the fact that it has been cited as a source of studies on historical floods in Andalusia in the 16th century (^{Pfister et al., 1999}) and throughout the 20th century (^{León et al., 2020}).

One of the most laborious tasks involved in acquiring information related to historical climatology and hydrology is consulting multiple documents with varied content in archives and libraries of diverse ownership. ^{Palomo (1878)} gathered over 400 works, most of them related to the history of Seville and the floods of the Guadalquivir River: records, annals, memorials, ephemerides, relations, appendices, speeches, dissertations, etc., of numerous authors throughout history. Among this vast number of documents, he highlighted, based on the number of citations and the authority granted to its comments, the Ecclesiastical and secular annals of Seville by D. Diego Ortiz de Zúñiga (1638-1680). This author gathered, in his memoirs, data on the period between 1246 and 1671 A.D. According to ^{León et al. (2020)}, the book of ^{Palomo (1878)} is a bibliographic source of inestimable value for its clarity and detailed information. It is also cited in the National catalogue of historical floods of Spain (^{DGPCE, 2019}).

2.3 Links of floods with meteorological instrumental records

Similarly, the Royal Institute and Observatory of the Navy of San Fernando (Cádiz, southwestern Spain) has uninterrupted pluviometric records from 1805, although with homogeneous monthly series from 1837. It is considered as a reference observatory due to the homogeneity of its temporal series, which is the longest series for southern Spain (^{Rodrigo, 2002}; ^{Sousa et al., 2010}). A recent study, which analyzed 25 423 pluviometric observatories in 32 geographical areas worldwide, highlighted that the observatory of San Fernando has the longest uninterrupted daily series (^{Morbidelli et al., 2020}). San Fernando is not located in the Guadalquivir Valley (it is 30 km away from the mouth of the Guadalquivir River), although it is in the same drainage basin, i.e., the Andalusian South Atlantic basin. Different authors have applied the evolution of monthly precipitation series of San Fernando as an indicator for the entire southwestern region of the Iberian Peninsula (^{Sousa et al., 2009}, ²⁰¹⁵; ^{García-Barrón et al., 2013}, ²⁰¹⁸). During the 19th century, reports about floods partially overlapped with the instrumental rainfall records of the observatory of San Fernando.

After the summer period, in which the monthly precipitation of July and August is frequently null, the precipitation of the beginning of autumn is incorporated into the subsoil; therefore, except for occasional torrential rains, the runoff of streams and the flow in the tributaries of the Guadalquivir does not present a notable increase until the end of autumn.

3.1 Methodology used to estimate the severity of the floods between the late 13th century and the 18th century

To calculate the severity of floods reported between the 13th and 18th centuries, so that the severity values can be compared, we must standardize the information obtained from documentary sources of different origin (by author, date and content). Thus, we established source-contrast systems that allowed the subsequent classification of very different flood events, selecting the following main criteria: (a) the height reached by the overflow on the wall or on one of the gates of the city, (b) the damage caused by the collapse of buildings, and (c) the number of people drowned.

Other secondary criteria to standardize the range of severity of the floods were: (a) the evacuation of exterior buildings (private homes and cloisters), (b) livestock mortality in the flood plain of the Guadalquivir River in the vicinity of Seville, (c) effects on the floating bridge or on the ships of the harbor, and (d) rogations and other religious ceremonies.

Apart from these criteria, and with the aim of establishing the severity of the flood events, we used a triple filter that modulated the classification assigned to them by different documentary sources, in order to compare their severity levels:

Appendix 1 in the supplementary material shows, in a comparative manner, the main and secondary criteria employed to assign the overflow level and impact of different flood indices, as well as the filters applied to modulate the classification of the severity of floods.

Guided by these criteria, we assigned an ascending index of flood severity as a function of the level of the overflow and its impact:

Each of the testimonies about specific events highlights some of the above-mentioned aspects, without the possibility of establishing a common typology. Thus, it was necessary to conduct a comparative evaluation of the information provided in each case. We recognize that the application of the previous criteria to each year with a registered flood is subject to personal valuation margins of the authors of the original documentary sources. To control this subjectivity, the descriptions of the different documentary sources were transformed into numerical tables based on the flood indices (see Appendix 2 in the supplementary material). Other documentary sources consulted can be seen in Appendix 3 in the supplementary material.

3.2. Methodology used to calculate the severity of the floods reported in the 19th century and their relationship with instrumental meteorological records

The second volume of ^{Palomo (1878)} is focused on the floods that took place in Seville during the 19th century. The information provided by different documents compiled by this author is not uniform throughout the 19th century. The description about the effects of the overflow cannot be directly compared to those in previous centuries, mainly due to the large construction works carried out in 1776 with the building of a pier in front of the walls, and later in 1816 with the construction of a canal that reduced the risk of flooding. Thus, these elements modified the severity of the impacts that served as classification criteria in this study for previous centuries.

The average of the instrumental series from the 19th century shows that in the Guadalquivir River Valley, the intra-annual distribution of precipitation increases in autumn until it reaches a monthly maximum in November-December and decreases progressively in spring (^{García-Barrón et al., 2013}). The interannual variability of the seasonal changes of the precipitation regime is associated with the North Atlantic Oscillation (NAO) in the Iberian Peninsula (^{Trigo et al., 2004}), which, in its negative phase, favors the entry of storms in the Guadalquivir River Valley (^{Gallego et al., 2006}; ^{García-Barrón et al., 2018}). This current behavior can be extrapolated to previous centuries (^{Luterbacher et al., 2002}).

During the first half of the 19th century, the classification of the severity of floods of the Guadalquivir River was based on the description of the effects caused by the floods. However, from 1858 to the end of the 19th century, we generated a quantified series of river overflows that allowed us to establish the possible degree of synchronous correspondence between the temporal series of both climatic manifestations: precipitation vs. overflow. For each flood, we identified the height of the river over the usual level. In the Guadalquivir River basin, floods are generally due to persistent rainfall and not to very short and intense rainfall events. This period of persistent rains precedes the more intense rainfall that ends up leading to the overflow of the channel. According to the revised historical documentation, in the surroundings of Seville, the floods used to last several weeks. For this reason, we used the data for the month in which the maximum overflow occurred and also the data for the previous month (bimonthly). For that reason, the meteorological variable used for the correlation test was the excess of bimonthly precipitation at San Fernando with respect to the average of the period 1837-1890.

To synthesize the temporal development, we generated a new series, based on the flood index, that shows the rainy sequences or periods between the late 13th and 19th centuries using the centered moving average calculated for 11-yr periods. In general, from the 16th century the number of documentary sources available for the same event has progressively increased. This means that the information available is also greater, which does not necessarily imply that the number of floods is therefore more frequent or intense.

4.1 Severity of the floods between the late 13th century and the 19th century

A classification of the Guadalquivir River floods, based on the flood severity index assigned, between 1290 A.D. and the late 18th century is shown in Figure 4. Although there is temporal continuity, to visualize the flood events in more detail, they were grouped by century. More specifically, Figure 4a represents the 13th, 14th and 15th centuries; Figure 4b shows the 16th century; Figure 4c displays the 17th century; and Figure 4d shows the 18th century.

Fig. 4 Flood index estimated for the Guadalquivir River in Seville: (a) 1290 A.D. to the 15th century, (b) 16th century, (c) 17th century, and (d) 18th century. 

4.2 Severity of the floods during the 19th century

Section I of volume II (1800-1858) of ^{Palomo (1878)} describes 10 overflows of the Guadalquivir River. The recorded documentary sources differ in the type of information they provide in different years, although they are consistent in the use of platforms (“borriquetes” in Spanish) raised above the floating bridge of the Guadalquivir River to connect to the neighborhood of Triana (1803, 1830, and 1841). In January 1823, the main catastrophic flood of this period took place, which inundated the entire neighborhood of Triana and reached the height of the previous great flood of 1796. We can assert that in the first half of the 19th century, the frequency of floods of different intensities was high. Section II describes historical episodes of the city that were not directly related to the river flooding. As a novelty, section III provides uninterrupted data from 1858 on the height (feet) reached by the river during each flooding event. Table II shows the years and months of maximum overflow and the corresponding elevation over the usual level of the river.

Moreover, as noted in the methodology section, we used the monthly rainfall series of the San Fernando Observatory (southwestern Spain) to calculate the bimonthly excess precipitation with respect to the average of that in 1837-1890, coinciding with the floods. The flood of 1876-1877 was not used in the calculations since after the demolition of most of the gates and the walls of the city in the preceding years, the river flooded the city. Therefore, without protection, there was no longer a uniform criterion to compare the effects of previous floods.

Figure 5 shows a pair diagram between the synchronous series of bimonthly excess rainfall p (l m^-2) and the overflow level d (inches). Although the number of data points is small, it can be observed that the dots are roughly aligned.

Fig. 5 Scatter plot of the elevation of the river level, measured in inches, and bimonthly excess rainfall, measured in millimeters, with the regression line indicated. 

The small number of events and the type of variables that led to Eq. (1) do not allow extrapolation to previous centuries, although R² explains more than 70% of the variance (Pearson’s correlation coefficient R = 0.84, p = 0.073, and the standard error of the estimates is 3.78).

The average of the instrumental series from the 19th century shows that in the Guadalquivir River valley, the intra-annual distribution of precipitation increases in autumn until it reaches a monthly maximum in November-December and decreases progressively in spring (^{García-Barrón et al., 2013}). The precipitation regime is associated with the NAO in the Iberian Peninsula (^{Trigo et al., 2004}) which, in its negative phase, favors the entry of storms in the Guadalquivir River valley (^{Gallego et al., 2006}; ^{García-Barrón et al., 2018}). This current behavior can be extrapolated to previous centuries (^{Luterbacher et al., 2002}). Figure 6 shows the intra-annual rainfall distribution (Fig. 6a) and the intra-annual distribution in the number of floods (Fig. 6b), exhibiting the delay in the month with the highest probability of floods (January) with respect to the maximum of the intra-annual rainfall distribution (November-December). This could be due to the fact that floods are frequently influenced not only by immediate direct rain but also accumulated rain in a drainage basin.

Fig. 6 Intra-annual distribution: (a) relative rate of monthly rainfall during the instrumental period and (b) intra-annual distribution in the number of floods. 

Between the starting date of the uninterrupted instrumental series of rainfall records (1837) and the year of the last reports about floods gathered by ^{Palomo (1878)}, the following is observed:

4.3 Interdecadal-scale estimation of the pluviometric evolution in the Guadalquivir River basin based on the records of floods in Seville

Frequently the floods of the Guadalquivir River are due to the accumulation of intense rains for several days in the context of a wide period of some weeks with rainy conditions. In general, this type of temporary distribution of rainfall is usually associated with the existence of west-south-west Atlantic fronts.

Using the results obtained in sections 4.1 and 4.2, it was possible to generate an inter-annual series of the flood index from the 13th to the 19th century. From the results of the annual flood indices for 1297-1796 (Fig. 4 and Appendix II), the centered moving average was calculated for 11-yr periods (Fig. 7). The values of the Y coordinate represent the temporal variation of the flood index. This allows to establish a comparable ordinal scale for more than five centuries, which estimates the historical evolution of floods in the southwestern Iberian Peninsula.

Fig. 7 Decadal-scale estimation of the flood index evolution in the Guadalquivir River basin based on the flood records for Seville between 1250 and 1800 A.D 

Figure 7 shows a non-periodic alternation of years with flooding and non-flooding intervals. During the 14th century, a moderate level of floods can be observed; however between 1580 and 1650 there is a long sequence of flooding. Finally, from 1680, three discontinuous pulses are shown with a greater amount of flooding episodes: 1680 to 1710, 1730 to 1760, and 1770 until the end of the 18th century.

^{Rodrigo et al. (1999)} presented conclusions on the temporal development of precipitation in southern Spain based on several types of historical reports in different localities between 1500 and 2000, with instrumental records for the 19th and 20th centuries. We observed a generalized likelihood between their pluviometric evolution and the one obtained in the present study for the same periods in both studies (1500-1800 A.D.). The authors highlighted the positive anomaly or the humid period from the late 16th century to the mid-17th century.

Through a documentary analysis of institutional sources and ecclesiastic pro-pluvia rogations, ^{Barriendos (2007)} used two complementary ordinal indices of precipitation and drought. This author highlighted that in Seville, two periods of floods, one from 1580 to 1620 and the other from 1760 to 1800, occurred. He also identified two drought periods: 1560-1580 and 1660-1730. Generally, his results are globally consistent with those presented in this study, with no discrepancies at the multi-decadal scale. The results obtained by ^{García-Martínez and Baena (2006)} from a geomorphological analysis of the study of floods in the lower Guadalquivir River are also in line with our results.

It seems probable that the effects of the Little Ice Age (LIA) on the Mediterranean latitudes were markedly more humid and with greater variability with respect to those on the northern latitudes of Europe (^{Sousa and García-Murillo, 2003}). ^{Pfister et al. (1999)} stated that climate change on the Iberian Peninsula could be more associated with precipitation and less associated with temperature. ^{Grove (2001)} highlighted that the same climatic conditions that induced the advance of glaciers during the LIA were also responsible for an increase in the frequency of floods and sedimentation in Mediterranean Europe. Several studies (^{Barriendos and Martín-Vide, 1998}; ^{Rodrigo et al., 1999}, ²⁰⁰⁰) detected, with different aspects, humid periods during the LIA in Andalusia (1570-1630, 1780-1800, and 1830-1870), which alternated with dry periods and with great variability (^{Benito, 2006}; ^{Rodrigo, 2018}). Similarly, different studies have detected greater aridity in the climatic conditions of the Doñana biosphere reserve, which coincided with the end of the last of the humid periods of the LIA on the southern Iberian Peninsula, and these conditions influenced the deterioration of lagoons (^{Sousa et al., 2010}), hygrophytic plants (^{Sousa et al., 2013}) and coastal streams (^{Sousa et al., 2015}).

In comparison with other studies, based on different types of qualitative sources of information and through different assignation procedures, the results of this study must be interpreted in their own context. The concept of drought can result in multiple interpretations in general, which refer to a deficit of precipitation. Therefore, although different authors that documented historic climatology used terms that, in a simplified manner, identify with dry/rainy, these do not have an unambiguous meaning, but they introduce conceptual differences depending on the applied methodology and the type of precipitation variable estimated.