Monthly storminess over the Po River Basin during the past millennium (800–2018 CE)

Reconstructing the occurrence of diluvial storms over centennial to millennial time-scales allows for placing the emergence of modern damaging hydrological events in a longer perspective to facilitate a better understanding of their rate of return in the absence of significant anthropogenic climatic forcing. These extremes have implications for the risk of flooding in sub-regional river basins during both colder and warmer climate states. Here, we present the first homogeneous millennium-long (800–2018 CE) time-series of diluvial storms for the Po River Basin, northern Italy, which is also the longest such time-series of monthly data for the entire Europe. The monthly reconstruction of damaging hydrological events derives from several types of historical documentary sources and reveals 387 such events, allowing the construction of storm severity indices by transforming the information into a monthly, quantitative, record. A period of reduced diluvial storms occurred in the ninth and tenth centuries, followed by a stormier period culminating in the eleventh and thirteenth centuries. More complex patterns emerge in the fourteenth to sixteenth centuries, with generally wetter and stormier conditions than during other centuries. From the seventeenth century onwards the number of damaging hydrological events decreases, with a return in recent decades to conditions similar to those prior to the thirteenth century The flood frequency tended to increase for all seasons during periods of low solar irradiance, suggesting the presence of solar-induced circulation changes resembling the negative phases of the Atlantic Multidecadal Variability as a controlling atmospheric mechanism.


Introduction
There has been increasing interest in the hydroclimatic response to large-scale environmental changes such as ongoing global warming (e.g., Luo et al 2018, Orth and Destouni 2018, Ljungqvist et al 2016, 2019 and largescale urbanization (e.g., Debbage and Shepherd 2018, Paul et al 2018), including active construction sites (e.g., Ricks et al 2020). Water is not only a valuable natural resource and societal asset, but has also the potential to inflict damage on human society and the environment, causing large disasters and financial losses (Hoeppe 2016, Brown et al 2018. However, the millennium-long evolution of the regional distribution, and timing, of damaging hydrological events (DHEs) are yet poorly known (Benito et al 2015). At present, we have, at best, an incomplete picture of past hydroclimatic extremes as a backdrop to the frequency and severity of present-day extremes , Stamatopoulos et al 2018. Identifying temporal and spatial changes and trends in precipitation-especially of extreme hydrological events-is difficult and poses a major challenge both A Monthly Storm Severity Index (MSSI) was developed to verify the scale-invariance in the relationship between the number of events larger than the storm strength events and the same strength events, i.e., to verify the completeness of the 'catalogue' of extremes both for the entire data set over 800-2018 CE and for three subperiods (800-1249subperiods (800- , 1250subperiods (800- -1849subperiods (800- , and 1850subperiods (800- -2018, resulting in an Annual Storm Severity Index Sum (ASSIS), i.e., the yearly sum of MSSI values.

Data
The Po River Basin contains the longest river in Italy (652 km), the Po River, flowing eastward across northern Italy in the administrative regions Aosta Valley, Piedmont, Lombardy, Emilia-Romagna, and Veneto (figures 2(a)-(b)). The Po River Basin has an area of ∼74,000 km 2 , situated almost entirely in Italy, with at present a total population of ∼16 million. The Po Delta Regional Park in the Emilia-Romagna region was designated in 1999 as a World Heritage Site by UNESCO. The Po River Basin is also where the most disastrous floods in the Italian history are recorded (Luino 2013).
The Po River Basin climate is rather complex and differentiated, given the geographical position that the basin occupies, together with the different morphology of the sectors that compose it (figure 2(b)). In this way, the climate of the Po River Basin is situated in the transition zone between the sub-continental climate of Central Europe (Alpine and Boreal) and a Mediterranean climate (Warm Temperate). (See also Köppen's classification in Strahler and Strahler (2000).
A large spectrum of environmental variables and phenomena is associated with cyclones in this region (Lionello et al 2006). In particular, northern Italy is affected by Atlantic storm tracks, which otherwise more directly affect western and northern Europe. The influence of Atlantic storm tracks in northern Italy quite frequently subjects the region to sudden events of extreme and adverse weather conditions, frequently causing considerable societal and economic impacts. The western-most part of this region is particularly stormy (Nikolopoulos et al 2013), though DHEs are common in some of the eastern regions of northern Italy too. Maximum daily rainfall is still capable of inflicting significant damages even in the less stormy eastern hydroclimatic areas (figure 2(c)).
The main Po River tributaries (Ticino, Adda, Oglio, and Mincio) provide the maximum water supply at the time of the thawing of alpine snow and ice (May-June), while numerous other tributaries descending from valleys of the Apennine contribute with meltwater and rain especially in spring and autumn, assuming in summer and winter a torrential character with low flow rates. The Po River Basin region can thus be viewed as the culmination of a complex hydrographic system (Guidoboni 1998). This regime of river hydrography includes rare waves of flooding of the Po's tributaries on both its right and left banks. When this occurs, the danger of disastrous and extensive floods is very high (Brandolini and Cremaschi 2018).

Documentary sources
The amount of surviving documentary sources in the Mediterranean region is sufficient back into medieval times to construct quantitative or quasi-quantitative climatic indices (see Pfister et al 2018). These sources also provide information about social vulnerability and impacts of climate extremes that allow for direct comparison with contemporary climatology (Wilhelm et al 2018). The study of the most famous fluvial floods affecting urban settings in the Middle Ages offers a conceptual reference to overcome too rigid a distinction between natural disasters and disasters produced by human agency. For many ancient sources, or sources referring to now lost original documents, information is available in Latin, and there are several words for the term storm and its damaging hydrological events (DHEs): diluvium, inundatio, excrescentia, fluminum and related composite locutions such as magnae pluviae, aqua maxima, tanta aquorum, inundation abundavit, impetu & aquarum multitudine.
Most historical climatology research in Europe focuses on the Early Modern Period (c. 1500-1800). It is research for this period that established the methods and procedures that have become standard in the discipline of historical climatology (Pfister et al 2018). In northern Italy, Milanese and Parmensi documents (Cantarelli 1882), and a variety of sources from the Po River Basin (Bottoni 1872, Cazzola 2010, Baldini and Bedeschi 2018) are qualified for reconstructing time-series of extreme events over the past millennium or more.

Methods
In addition to extensive primary archival research and research in Italian historical libraries, literary sources were accessed by web-engine search (https://books.google.com), generating ∼100,000 bibliographic records. From this first massive bibliographic search, only ∼1,000 records met the criteria of including the keywords abundant rainfall, storm, downpour, diluvial, flood and alluvial (piogge abbondanti, tempesta, nubifragio, diluvio, inondazione, alluvione), in addition to some Latin locutions (e.g., magnae pluviae, aqua maxima, diluvium, excrescentia fluminum, inundatio), which were chosen for careful reading. Useful data from the documentary sources were retrieved by transforming the information contained in narrative accounts into numbers on an index scale (see Appendix and Scoring system with monthly data [MSSI] in supplementary information is available online at stacks.iop.org/ERC/2/031004/mmedia). A procedure, called weather hindcasting (Pfister et al 2018), was applied to become familiar with well-documented anomalies within the instrumental period prior to analysing analogous cases in the pre-instrumental past. In this way, a scoring system was applied to grade the MSSI, equal to 0 (normal), 1 (stormy), 2 (stormy with a few floods), 3 (stormy with large floods) and 4 (extraordinarily stormy with very large floods) (following Wetter et al 2011). In this way, a scoring system was used to grade a MSSI (which is defined as the period of the year between December and November), equal to 0 (normal), 1 (stormy event), 2 (very stormy event), 3 (great stormy event) and 4 (extraordinary stormy event): Normal means average or storm passed unobserved, without comments about its severity or its impacts on society and economy.
Stormy means an event is considered stormy if intense rainfall occurred with only limited damage, and no floods, recorded.
Very stormy was classified when intense rainfall occurred with some floods.
Great stormy refers to an extreme diluvial event, with severe and large floods and agricultural works are suspended, and urban communications are interrupted.
Extraordinary stormy is characterized by sporadic very extreme events, with a low centennial recurrence rate. These extreme diluvial events affect, at the same time, several river basins, killing people, animals and felling trees.
This kind of understanding is exemplified in the form of table (appendix A), incorporating monthly and annual values and the relative sources for exemplary years. The creation of an annual index was so designed to summarise the sum of MSSI of each month, as the Annual Storm Severity Index Sum (ASSIS). The study was based on the systematic and critical analysis of data about the above-mentioned phenomena offered by Italian documentary sources for a period covering 800-2018 CE. For most of the information, it was possible to make an 'event check' by considering more than one documentary source to the same event. It was also possible to contextualise the storm events with other types of historical events (e.g., social, agricultural, religious). In this way, the reliability of information can be assessed by trying to shed some light on the issue of climate relations and extreme events, looking beyond the quantitative data, and seeking alternative information from diaries, chronicles, and local stories.
Completeness and robustness of the reconstructed extreme hydrological events In the documentary data, we have revealed the presence of 387 extreme hydrological events occurring in northern Italy from 800 to 2018 CE, of which 14% have an unknown date. Sorting these events by class of severity results in 83 stormy events, 219 very stormy events, 65 great stormy events, and 20 extraordinary stormy events. However, our historical hydrological database relies on several types of heterogeneous sources, including accounts which might be an exaggeration of what it was, uncritical reference to previous sources, misprints in documentation and natural records of environmental parameters (proxy data), all that can be different factors of uncertainty in 'cataloguing' storm records (Pavese et al 1994). It is well established, for example, that there is a tendency to underestimate smaller events with isolated storms, especially from remote places or during summer, when more localized storms can be common. To resolve some of these uncertainties in our database, we established a reasonable criterion with respect to the recorded MSSI events. This was done by defining a partition of the time-series in three climatic sub-periods-the Medieval Warm Period (MWP; here 800-1249), Little Ice Age (LIA; here 1250-1849) and Modern Period (MP; here 1850-2018 following Diodato et al 2019)-and verifying for each climatic sub-period (figure 3(a)), and for the entire dataset ( figure 3(b)), the scale-invariance in the relationship between the number of events larger than the storm strength events and the same strength events. The complete analysis was formalized with the relationship between the cumulative number of events (CEN) and MSSI values within the range 1MSSI4, as follows (10):  where MSSI is the monthly storm severity index by severity class (i) and sub-period ( j). The negative slopes in all the climatic sub-periods, and for the entire dataset, reflect the principle of a progression towards less frequency as storm events become larger. A Pearson's correlation coefficient r=1 is expected, with probability to reject r=0 equal to p<0.01.
The 'catalogue' concerning the Modern Period 1850-2018 represents an exception. For this period, a distinct set of more complete information is evident ( figure 3(a 3 )). In this way, the storm events in the period 800-2018 CE can be assumed significantly scale-invariant and therefore complete only for the 304 events that, within the range 2SSI4, are described in qualitative terms as very, great and extraordinary storms. The remaining 83 events with MSSI=1 (red points in the scatterplots) are not fitted by the regression lines drawn in figure (3). Their number is much smaller than required by equation (1), probably because many of these lowenergy events escaped detection in the past. Events with MSSI=1, classified simply as stormy events only, have been discarded from our temporal analysis because they are not representative of the entire 'catalogue' within the investigated time-period. The methodology adopted not only ensures a great homogeneity, and a reduced uncertainty in the whole time-series 800-2018, but also in each climate sub-period (the Medieval Warm Period, the Little Ice Age, and the Modern Period).

Results and discussion
Po River Basin-wide reconstruction of extreme hydrological events The influence of large-scale climate variability on the occurrence of DHEs in the Po River Basin since 800 CE, i.e., at the beginning of the MWP, through the Little Ice Age, until the warming phase of the Modern Period (MP), is summarised in the graphical representation of figure 4. Changes in DHEs can be visualised and explained by repeated occurrences of an annual storm severity index sum (ASSIS). In order to identify possible trends and oscillations, the reconstructed time-series was filtered using a 11-year low-pass Gaussian filter to compute a continuous time-series of ASSIS(GF), which is meant to remove the high-frequency noise in the reconstructed discontinuous data ( figure 4(a), blue curve). Filtering reduced the range of ASSIS values (0ASSIS9) to between 0 and 3.7.
Until the middle of the thirteenth century, few and sporadic-sometimes moderate-DHEs affected the Po River Basin. From the ninth to twelfth centuries, catastrophic floods appear to have been rather exceptional, with only seasonal floods not considered as events worth considering. The presence of still rather extensive woodlands may have prevented rain-water from arriving downstream quickly in the form of dangerous flood waves. This is distinctly visible in figure 4(a), which shows that ASSIS(GF) is only little accentuated in this drier period in the ninth and tenth centuries. Between c. 750 and 1250 CE, southern Europe, including northern Italy, tended to receive relatively low amounts of precipitation, at least in summer, whereas the temperature, on average, was higher (Luterbacher et al 2016, Ljungqvist et al 2019. With the rapid disappearance of the woodland coverage (e.g., Kaplan et al 2009Kaplan et al , 2010, culminating in the twelfth and thirteenth centuries (owing to increased agricultural activities), the time required for river waters to flow from the most remote point of a watershed to the watershed outlet decreased, together with a reduced absorption of rainfall in the soil through vegetation cover (Cazzola 2010). During the twelfth to fourteenth century pulsing storms were observed with greater frequency and intensity, with the ASSIS(GF) index oscillating upwards from the thirteenth century. Human activities may have contributed to the impacts of rainfall events on land degradation when frequent floods occurred in this period (Brandolini and Cremaschi 2018). However, the regional storminess, which increased systematically at the beginning of the LIA, can be considered the main cause of the hydrological change occurring in this period (Baldini and Bedeschi 2018). The exception is one brief phase (1176-1178 CE) when (accompanied by a temporary lowering of temperatures; Borsato et al 2003, Luterbacher et al 2016, Wilson et al 2016 the first major floods are recorded-with the ASSIS(GF) greater than 98 th percentile (that is when it is stormy with great and extraordinarily stormy floods occurring in the same year) -and hydrographic changes in the Po River Valley, with damages also on agricultural crops (Veggiani 1986).
A possible climate mechanism governing the frequency of DHEs in northern Italy is the AMV. During a positive phase of the AMV the number of DHE are significantly lower whereas during a negative phase of the AMV the number of DHE are significantly higher (figures 4(a)-(b)). Moreover, the colder phases of the LIA were associated with the intensification of extreme precipitation in the region and correspond to periods of reduced solar activity (Steinhilber et al 2009)  The wavelet coherence spectrum (figure 4(c)) displays both high-(∼11 years) and low-frequency (∼300-years) periodicities during the LIA. The ∼11-year sunspot cycle is a main feature of solar activity  (Hammer et al 2001). Colours identify the 0.10 significance level areas. The bell-shaped, black contour marks the limit between the reliable region and the region above the contour where the edge effects occur.
variability (Wolf 1852(Wolf , 1853, while the ∼300-year cycle may reflect a quasi-periodic feature of sunspot variability, which is likely related to precipitation changes in the North Atlantic region (as determined by Ojala et al (2015) from sediment records of lakes Nautajärvi and Korttajärvi in Finland). Whereas significant periodicities less than ∼11 years occasionally occur without any relation with climatic periods, the periodicity of ∼300 years extends over the entire LIA. Other low-frequency periodicities (>300 years) are also significant but they fall above the reliable area formed by the time axis and the bell-shaped contour. On the one hand, this suggests that recurrences of storminess become less statistically predictable outside the LIA. This may demonstrate that dynamic atmospheric processes are present in generating cyclone-related precipitation extremes throughout the LIA (Raible et al 2018), departing from the Clausius-Clapeyron thermodynamic expectations of an increase in precipitation intensity associated with atmospheric warming (Pall et al 2007, Trenberth 2011, Trenberth et al 2014, Kirby 2016, Prein and Pendergrass 2019. Our study thus shows the same tendency as revealed by Ljungqvist et al (2019) for a negative low-frequency temperature-hydroclimatic coupling (i.e., warm and dry) in southern Europe. Likewise, solar-type periodicities suggest that the Sun may be one of the precursors of hydrological processes in northern Italy (Zanchettin et al 2008).
Variability, frequency distribution and seasonal patterns of the storms Knowledge of the monthly or seasonal variability of storms is important to obtain an insight into the water's destructive force. As it is a measure of how far monthly values deviate from their average value, it reflects the dangerousness of storms for land management (associated with land cover/use changes). We have presented a twelve century-long (800-2018 CE) perspective of storm regime patterns shown as the absolute frequency distribution of storm events (table 1).
The seasonal distribution of events, grouped on a seasonal basis, showed that autumn is unambiguously dominating with most frequent flooding (49%), while the remainder of the events is almost identically distributed over the other three seasons: summer (18%), spring (17%) and winter (16%). This distribution is in agreement with the assessment of reconstructed hydrological conditions of the Central Alps derived from Wirth et al (2013). We also consider the attribution of changes in DHEs at the monthly scale. These analyses provide more insight into flood-generating processes and enable the evaluation of the conclusions we reached using data on an annual scale. Grouped by century, the seasonal DHEs appear distributed according to similar distributions (Kolmogorov-Smirnov pair-wise p-values>0.05) in winter, spring and summer (with a maximum during the 16 th century, at the climax of the LIA), while the autumn events follow a distinct distribution (Kolmogorov-Smirnov pairwise p-values<0.05) with important peaks of activity (figure 5), which make them unpredictable in any climatic period.
The yearly frequency of monthly DHEs across the three considered climatic periods: 800-1249 (MWP), 1250-1849 (LIA), and 1850-2018 (MP) is shown in figure 6. The respective histograms of frequency indicate that the three periods present distinct climatic regimes: quasi-multimodal (MWP); bimodal-continental (colder) with diluvial events roughly occurring in all seasons and strengthening in spring and autumn (LIA); and bimodal-Mediterranean (warmer) with concentration of extreme events in late spring and autumn only (MP). As shown in the maps of figure 6, these shifts of the peak positions align with the shifts in relative regional reconstructed drought (Palmer Drought Severity Index, scPDSI) in the tree-ring based Old World Drought Atlas (Cook et al 2015).
The MWP (figure 6(a))) shows only a little prominence of multi-modality, while intra-seasonal characteri.e., when the seasonality in rainfall intensities was in phase with the seasonality of previous catchment wetness -appears to be small. In these months, with relatively wet antecedent conditions, small floods are occurring quite frequently, leading to a stable flood-frequency curve. Hence, meridional circulation conditions could have affected the MWP inducing heavy convective storms throughout the year, which resulted in large variability, and thus little seasonality, of DHEs over northern Italy. This phase displays dry-to-very-dry western and northern margins of the Po River Basin (figure 6(d)), with an only marginally more humid area in the southeast.
In contrast, a strong bimodal component in spring and autumn is found during the LIA, with a relative peak of flood frequency in April-June and an absolute peak in October ( figure 6(b)). This is an important finding, indicating that small and moderate flooding in spring-summer is related to the frequency of storms. However, there may be quite unusual and extreme DHEs occurring in the autumn, which are presumably related to  southerly circulation patterns when warm and moist air is advected from the Mediterranean Sea (e.g., Parajka et al 2010). During the MP (figure 6(c)), only autumn storminess remains active. In these months, with dry antecedent conditions and large rainfall intensities (e.g., in June-August), localized storms are more frequent with a number of larger floods occurring only in October-November. The map in figure 6(f) also shows dry conditions, with alternating dry and very dry anomalies.

Conclusions
From the ninth century onwards, floods are reported more systematically than for earlier periods. Recurring storms are listed in documentary data, allowing a quantitative reconstruction of extreme hydrological events on a monthly level. It is clear that the past twelve centuries (800-2018 CE) have been marked by alternating phases of higher and lower frequency of storms and floods providing a backdrop for how to manage natural resources to protect new generations. Periods with increasing flood frequency are found to align with solar minima. Though the physical meaning of this finding merits further investigation, we tentatively suggest that solar activity likely helps to induce large-scale circulation changes resembling negative phases of the AMV acting as a controlling atmospheric mechanism of DHEs. Then, despite the presence of increasing anthropogenic climatic forcing, an expected decrease in solar activity during coming decades could possibly re-intensify the risk of frequent flooding in southern Europe. The rapidly developing sub-mesoscale convective systems tend to be responsible for the heaviest and most locally destructive storm events in the Mediterranean region as they are affecting small catchments with the most vulnerable systems to storm-driven flash floods and soil erosion hazards.