Three-Century Climatology of Cold and Warm Spells and Snowfall Events in Padua, Italy (1725–2024)

Abstract

Regular meteorological observations in Padua started in 1725 and have continued unbroken up to the present, making the series one of the longest in the world. Daily mean temperatures and precipitation amounts have recently been homogenized for the entire 1725–2024 period, making it possible to add new measurements without further work. Starting from the temperature series, the trends of cold and warm spells are investigated in this paper. The ongoing warming that started in the 1970s is extensively analyzed on the basis of the variability of the mean values and a magnitude index that captures both the duration and intensity of a spell and by investigating the frequency of extreme events by means of Intensity–Duration–Frequency curves. The periods with the greatest deviation from the climatological average are analyzed in detail: February 1740 and April 1755, the months with the largest negative and positive temperature anomalies, respectively, in the 300-year-long series. Moreover, the analysis of snow occurrences extracted from the original logs, together with the pressure observations from the long series of London and Uppsala, made it possible to evaluate the most typical synoptic situations leading to snow events in Padua for the whole period.

1. Introduction

The long series of meteorological observations in Padua (Northern Italy) [1,2], covering the last three hundred years, offers the rare opportunity to study the evolution of the local climate in the transition from the pre-industrial to the modern era. The original logs, which date back to 1725 almost unbroken, contain readings of temperature, precipitation, pressure, and sky conditions. The temperature series has been recently homogenized for the entire period, providing a continuous time series of the daily mean temperatures since 1725. The availability of this dataset opens the possibility to explore the variability and trend of cold and warm spells in Padua.

A cold or warm spell (CS and WS from here on) is a prolonged period of extremely high or extremely low temperature for a particular region. The physical drivers and the mechanisms responsible for the persistence of temperature extremes are quite complex. CSs and WSs are caused by air circulation patterns in the atmosphere. In general, cold spells result from northerly cold advection, whereas warm spells are caused by either adiabatic warming (in summer) or warm advection (in winter).

CSs and WSs frequently result in adverse effects on agriculture, energy demand, human health, and infrastructure [3,4]. In recent years, numerous high-impact events have occurred worldwide linked to unusually persistent surface temperature anomalies. In particular, the mean temperatures of the years from 2022 to 2024 were the three highest ones of the entire Padua series. The impact of the spells is related not only to their intensity but also to their season of occurrence, duration, and spatial coverage. For example, CSs in spring or late winter can be particularly dangerous for vegetation, and multi-week WSs are more damaging than short-term events [5,6].

CSs in winter are sometimes associated with snowfall. Even though snow in Padua has become quite rare over the last decades, with fewer than five snowy days per year from 1995 to 2024, in the past centuries this number was twice as large. Snowfall in Padua requires particular conditions, as the Alps to the north offer efficient protection against the stormy cold winds coming from Northern Europe, which are forced to enter the Mediterranean basin through the Rodano Valley, France (“Mistral” wind, from NW), or from the Balkans (“Bora” wind, from NE). Furthermore, the spatial distribution of the snow in the “Padana” plain, Northern Italy, is strongly influenced by the effective trajectory of the front in transit. In fact, a spatial difference of even a few hundred or tens of kilometers would cause different interactions with the mountains, amplifying or attenuating the front effects in a way that is difficult to predict even a few hours before the event.

In general, the study of the snowfall characteristics in various regions of the world on a long-term time scale is limited by the fact that the meteorological series are usually short. Regarding Italy, the longest series of snow events dates back to the year 1681 for Parma, 1741 for Rome, 1784 for Turin, and 1866 for Naples [7,8,9]. Not all the series provide measurements of snow accumulation; nevertheless, even just the information on the snow frequency is extremely useful for climatic studies. The reconstructions of snow events and accumulation are mainly concentrated in the mountain regions [10,11], as snow in these areas constitutes a precious water reservoir also for the surrounding plains.

In the pre-instrumental period, written sources and visual evidence provide valuable direct and indirect information about the occurrence of harsh winters: the presence of ice in wells, in the rivers, and in the Venetian lagoon, and also the freezing of wine and the death of animals and trees are documented. For example, it was established that abundant snowfall occurred in northeastern Italy during winter in 1443 (“it snowed continuously for 12 days”, “the Po River in Ferrara was frozen over, supporting carriages with horses”), 1491 (“all the Lagoons around Venice were frozen […] because of the terrible cold and snow”), 1608 (“due to an exceptional snowfall, it was impossible to walk in the streets or go out through the door”), 1684 (“the Lagoon froze. It snowed continuously for 10 days”), and 1709 (“the frost was so severe that both the Venice Lagoon and the Po River were frozen over […]. Heavy snowfall everywhere, which caused severe famine. Frost and snow lasted till the end of February”) [12]. Critical work is necessary to assess the correct dates of the events because of the different dating styles.

Few other locations in the world can claim a three-century-long meteorological series, comprehensive of temperature, precipitation, pressure, and sky observations, even though early measurements are affected by higher uncertainty than the modern period (1.1–1.5 °C for the mean daily temperatures of the 18th century [2]), that reflects on the investigation of the spells. Another factor that has an effect on the CS and WS analysis is the urban heat island (UHI) effect, which impacts the long-term trend of the temperature.

The aim of this work is to investigate the characteristics and variability of the CSs and WSs in Padua in terms of their frequency, intensity, and duration. To the best of the authors’ knowledge, there is no detailed statistical analysis in the literature on the variability and trend of warm and cold spells in Italy for the covered period, i.e., from the beginning of the 18th century to the present day. Some works report evidence of warming in recent decades but the few focused on Italy cover just the last decades [13,14,15,16,17].

After the Introduction, Section 2 presents the datasets and methodology used to identify CSs and WSs. Moreover, the frequency evolution of snow events over the last three hundred years is also studied, as they are sometimes associated with CSs in winter. In Section 3, the main features of the CSs and WSs in Padua over the 1725–2024 period are discussed and the most remarkable events are identified. At the same time, the European synoptic situations concomitant with the presence of snowfall in Padua and their trends are investigated. Conclusions are drawn in Section 4.

2. Materials and Methods

2.1. Datasets

The daily mean temperature series of Padua (45.40° N, 11.88° E) from January 1725 to December 2024, reconstructed in previous works [1,2,18], is considered in this study. Daily precipitation [19,20,21] and pressure [22] series are also used. The snowfall frequency has been extracted from the same logs recording the other meteorological variables. In total, 109,562 days are available, with no gaps for the temperature, 1 day missing for the pressure, and 89 days missing for the precipitation. The mean annual temperature for the 1901−2000 period is 13.4 °C, the lowest minimum daily temperature is −16.4 °C recorded in January 1985, and the highest maximum one is 39.8 °C in August 2003. The mean annual precipitation is 871.0 mm and the highest daily amount is 156.6 mm, recorded in September 2009.

ModE-RA paleo-reanalysis data are used to assess monthly anomalies of the geopotential at 500 hPa in the European domain [23]. EURO-CORDEX (Coordinated Downscaling Experiment–European Domain, at ~12.5 km resolution [24]) climate projection data referred to the pixel nearest to Padua were considered to provide temperature scenarios up to the end of the 21st century.

2.2. Methodology

2.2.1. Cold and Warm Spells Characterization

There is no unanimous consensus on the definition of CS and WS, nor on their duration, the parameters to be considered, or their thresholds (e.g., [25,26]). In this study, mean daily temperatures were analyzed, as they were available from the beginning of the series in 1725. A day was classified cold/warm depending on whether its mean temperature was below/above the 10th/90th percentile (computed on a daily basis from the 1901–2000 climatology).

Typically, the daily percentile thresholds and averages were smoothed considering windows of more than one day: for example, 5 days [27], 11 days [28], 15 days [29,30,31,32], 21 days [33], and 31 days [34]. Unfortunately, there is no clear recipe on how to choose the length of this window, and the choice deeply affects the outcomes. A possibility is to evaluate the “elbow plot” of the lag-1 autocorrelation coefficient of the 10th/90th percentile series vs. the window length (Figure S1a). The autocorrelation coefficient rapidly reaches an asymptote as the window length increases (for both the 10th and 90th percentiles). However, the number of total spells found in the 300-year time series does not stabilize in the same way, as they decrease almost linearly with increasing window length (Figure S1b). Therefore, whether the choice of the 17-day window seems reasonable looking at the elbow plot (Figure S1a), this choice will alter the total number of spells by 3–5%, which corresponds to tenths of events gained or lost. An alternative way to smooth the data is the use of local polynomial regression fitting (LOESS). This method does not require the selection of a window, and it improves the estimate of the seasonal cycle of daily thresholds [35]. It has been applied to the daily percentiles of the Padua series by means of the ”loess” function in R [36]. The only parameter to be set is the “span”, which controls the degree of smoothing. Figure S2a shows the lag-1 autocorrelation coefficient vs. the span parameter; an asymptote is once again reached. In this case, the total number of spells is quite stationary at low values of the span (Figure S2b), and then it increases starting from the value 0.30. The choice of 0.10 as the span parameter, right after the elbow in Figure S2a, is reasonable because it provides enough smoothing and it lies in a region where the number of spells does not vary much as the span changes (0.5–0.8% in the 0.08–0.30 range).

Following Lavaysse et al. (2018), a group of days containing at least three cold days (i.e., below the 10th percentile threshold) is considered a CS. In this group of days, not-cold days can be present, but they must be single-day breaks, not consecutive. These days are called “pool days” [27]: they do not interrupt a spell, but they are not considered in the calculation of its duration and intensity. The same definition has been applied to WSs, considering the 90th percentile as the upper threshold. The definitions of CS and WS are thus not based on arbitrary fixed thresholds but on values that vary according to the location and season. An example of these definitions is shown in Figure 1. The 2nd and 3rd of April are cold days, as their temperatures are lower than the 10th percentile threshold, but they do not constitute a CS. Instead, a CS occurred on 6–11 April. These cold days cannot be merged with the previous cold days, creating a unique CS, as the two groups of cold days are separated by more than one not-cold day (4–5 April). The days 15 and 17 April are warm days, but they do not form a WS, while the period from 23 to 28 April does: 4 days are above the 90th percentile threshold broken by single days, 24 and 27 April (which are pool days).

A magnitude index, $MI,$ was created to capture both the duration and intensity of a spell. This index is derived from the daily Heat Wave Magnitude Index [34] and is defined as $={\sum }_{\mathit{d}}{\displaystyle \frac{{\mathit{T}}_{\mathit{a}}-{\mathit{T}}_{25\mathit{p}}}{{\mathit{T}}_{75\mathit{p}}-{\mathit{T}}_{25\mathit{p}}}}$, i.e., the sum of all the daily magnitude values evaluated over the cold or warm days of a CS or a WS, with $T_a$ being the daily mean temperature and $T_{25p}$ and $T_{75p}$ representing the 25th and 75th smoothed percentile values of the time series composed of the 100-year annual daily mean temperatures of the 1901–2000 period.

Wavelet and red-noise spectrum analyses of the number of CSs and WSs were performed by means of the R packages “wt” and “dplR” [37,38]. Intensity–Duration–Frequency (IDF) curves are a tool useful for associating a return period to a spell. The methodology foresees the following steps [39,40]: (i) for each spell, the average temperature and anomaly are computed for all the sub-intervals of duration D_i, from D₁ = 1 day to D_max, where D_max is the duration of the spell; (ii) for each duration D_i, the minimum for each year is calculated for the CSs and the maximum for the WSs; (iii) the best fit distribution functions are evaluated for all the annual time series obtained in the previous step (the normal distribution was chosen based on the Anderson–Darling statistic); (iv) for each D_i, the cumulative distribution function and the probability of the occurrence of a year without spells with that duration are calculated; and (v) the return periods (RPs) are estimated as described in [40].

Finally, the expected number of spells and their duration according to the annual number of cold/warm days was simulated. Boolean values are associated with the cold (or warm) and normal days, and random synthetic time series can be created with the same autocorrelation function of the Boolean time series derived from the observations. The generation of random time series was repeated 420,000 times and loess quantile regressions were evaluated to estimate the 5th, 50th, 95th, 99th, and 99.9th percentiles of the number of CSs and WSs and their durations.

2.2.2. Snow-Related Synoptic Situations

To investigate the synoptic situations in Europe linked to snowfall events in Padua, the London (51.48° N, 0.46° W) [41] and Uppsala (59.85° N, 17.63° E) [42,43] pressure daily observations were utilized (Figure 2). These data, jointly with the Padua pressure series [22], were clustered using the k-means algorithm computed with the “kmeans” function in R [44]. Since the algorithm is sensible for the initial conditions, it was repeated 10,000 times with different random initial conditions and only the days that exhibited coherence, i.e., belonging to the same cluster in at least 90% of the 10,000 runs, were retained. The other days, 345 out of a total of 1894 snow events over the last three hundred years, could not be attributed to a particular cluster and were excluded from the analysis.

3. Results and Discussion

3.1. Cold and Warm Spells

The year 1740 was the coldest of the entire Padua series, from 1725 to the present day [2,39] and, according to Brönnimann et al. (2024) [45], the coldest for the whole of central Europe over 600 years. Furthermore, for Padua, the coldest winter, spring, summer, and the third coldest autumn belong to that year. It is possible that in Northern Italy, the winter of 1709 was even colder, as reported by several chronicles and archive documents, but no sound instrumental documentation exists. A few contemporary Italian sources mentioned some temperature values, e.g., −18 °R, but thermometers were rare, and it is possible that measurements were taken elsewhere [12]. Indeed, February 1740 has the largest negative monthly anomaly of the series, −9.2 °C with respect to 1901–2000 climatology. On the other hand, April 1755 has the largest positive monthly anomaly, +5.6 °C (Figure 3a). The lowest anomaly value is particularly impressive, as it is 0.9 °C lower than the subsequent minimum. Contrariwise, the two highest anomaly values have a difference of only 0.1 °C.

The warming of the last decades is evident from Figure 3. Gaussian distributions have been used to fit the observations as their normality has been confirmed by Kolmogorov–Smirnov tests, which provided p-values > 0.05 for all the cases, rejecting the null hypothesis that the data do not follow a Gaussian distribution. Indeed, the probability of having an anomaly below −2 °C was 23% in the earliest period, 1725–1754; it dropped to 4% in 1965–1994 and to 0.5% in 1995–2024, 8 times less than the previous period (

Table 1. Probabilities derived from the Gaussian distributions shown in Figure 3a of finding a monthly anomaly below −2 °C or above +2 °C, with respect to 1901–2000 climatology.

Monthly Anomaly	1725–1754	1965–1994	1995–2024
<−2 °C	23%	4%	0.5%
>+2 °C	7%	10%	40%

). The most recent month of the entire series below the −2 °C anomaly threshold was September 1996, and in the 21st century, only May 2019 approached it. On the other hand, the probability of finding an anomaly above +2 °C was 7% in 1725–1754; it increased to 10% in 1965–1994 and to 40% in 1995–2024, nearly 4 times more than the previous period. Figure 3b visualizes the temporal evolution of temperature anomalies with monthly resolution. Extremely warm months also occurred in the past, well before the present-day warming, e.g., the already-mentioned April 1755 but also December 1825, January 1804, January 1845, and others. However, the frequency of such events was much lower than in the last decades.

The extreme months of February 1740 and April 1755 are the result of opposite large-scale circulation patterns. In February 1740 (Figure 4a), an enhanced anticyclonic circulation at 500 hPa was likely located above the North Atlantic and extended up to above the Arctic Circle, conveying continental cold air masses from northeastern Europe to the Mediterranean areas, also causing extensive snowfall. In April 1755 (Figure 4b), the anticyclonic circulation was located over southern France and northern Italy, transporting warm subtropical air, which caused frequent clear-sky conditions and possibly temperatures near or above +30 °C, typical of summer. Indeed, for Padua, this was also one of the driest Aprils of the entire series (only 1.0 mm of precipitation was recorded, whereas the 1901–2000 average was 77.2 mm). Comparing the calendar days of each year in the 1725–2024 period, for 8 out of 30 days of April, the highest mean temperatures still belong to 1755. The exceptionality of this warm month is confirmed by the observations in Basel (northern Switzerland, ~400 km northwest of Padua) and Bologna (northern Italy, ~100 km south of Padua). The Basel temperature series started in 1755 [46], and that month had a mean temperature anomaly of +4.9 °C (with respect to 1901–2000 climatology), making it the warmest April of the whole time series until 2007 (the year to which the second warmest April for Padua belongs). The Bologna series started in 1715 [47] and the mean temperature anomaly of April 1755 is +3.7 °C, among the highest values of the pre-industrial era for the city.

A different way to investigate the temperature change is by looking at the trend of CSs and WSs. Figure 5a,b show the time trend of the number of CSs and WSs for each 30-year interval, respectively, in the cases of at least 3, 4, and so on up to 7 days. The number of CSs decreases with time: spells of at least a 3-day duration decreased by −69% from 1725 to 2024, and spells of at least a 7-day duration disappeared almost completely in 2023 (−93%). On the other hand, the number of WSs increased considerably in the entire period, with an increase of +357% and up to +437% for the spells of more than 3- and 7-day durations, respectively.

Figure 5c shows the time series of the duration of the spells. The duration of the CSs decreased from 8.1 days in 1725–1754 to 4.1 days in 1995–2024, while the WSs reached a maximum of 7.0 days in the last period, with a minimum of 4.9 at the end of the 19th century. The duration of the longest CS is 52 days, from 14 February 1808 to 5 April 1808, while the longest WS is 55 days, from 8 July 2024 to 4 September 2024. Finally, Figure 5d shows the $MI$ values during the CSs and WSs. The most intense CS occurred in 1740, as previously discussed, from 28 January to 12 March ($MI$ = −78.1), while the most intense WS ($MI$ = 112.3) is once again from 8 July to 4 September 2024. Other dates can also be provided by looking at the most extreme mean temperature daily values or anomalies. The results are reported in

Table 2. The most remarkable cold spells and warm spells in Padua, 1725–2024: dates, duration, $MI$, most extreme daily temperature anomaly, and mean values.

	Main Feature	Dates	Duration (Days)	$\mathit{M}\mathit{I}$	Extreme Daily Anomaly (°C)	Extreme Daily Mean Value (°C)
Cold spells	Max duration	14 February–5 April 1808	52	−69.0	−11.7°	−2.4°
	Max intensity	28 January–12 March 1740	44	−78.1	−12.8°	−8.4°
	Min daily anomaly	28 January–26 February 1929	29	−50.9	−15.4°	−11.1°
	Min daily value	12 December 1788–10 January 1789	28	−50.4	−14.0°	−11.2°
Warm spells	Max duration Max intensity	8 July–4 September 2024	55	112.3	+7.5°	30.4°
	Max daily anomaly	31 March–12 April 2011	13	29.7	+10.7°	22.9°
	Max daily value	2–30 August 2003	27	63.0	+8.4°	32.1°

. Some other notable spells are reported in Table S1 according to their $MI$. Apart from the 1808 winter cold spell, it is worth mentioning that during the summers (JJA) of the years from 1812 to 1816, some notable cold spells occurred. In particular, in 1815 and 1816, two events that are among the most extreme summer CSs of the Padua series occurred. In fact, the largest $MI$ value for a summer cold spell of the whole series is −13.0, which happened both in 1725 and 1919, and the 1815 and 1816 events were characterized by $MI$ = −12.9 and =−11.9, respectively. In the 1810–1820 period, the overall summer temperature anomalies were very low, and 1813, 1815, and 1816 were the coldest summers after that of 1740.

In Figure 6a,b the yearly numbers of CSs and WSs are grouped according to the $MI$ range values. While in Figure 5d, the mean value of $MI$ for each 30-year interval was shown, here, the contribution of the spells with different magnitudes has been made explicit. The few CSs over the last 30 years were characterized almost entirely by an $MI$ higher than −5 and lower values were very rare. CSs with an $MI$ lower than −20 disappeared, while in the previous decade, they occurred once every few years. Conversely, the number of WSs increased, particularly the most severe ones. An $MI$ higher than 10 has become common, and nearly every year in the last decades has had WSs with an $MI$ higher than 20. In 2022, even the 40 threshold was surpassed four times (see Table S1). After a certain yearly number of cold or warm days (shown in Figure 6c,d), the number of spells does not increase anymore, because with fewer “normal” days (i.e., not cold nor warm) available, the duration of the spells increases and their total yearly number decreases. This number, ~180 days (see Figure S3), will likely be reached by warm days in the next decades if the increasing trend of Figure 6d continues, saturating the number of WSs at an average of 16–17 per year (14–21 is the 5th–95th percentiles range). Instead, the maximum annual duration of the WSs will increase monotonically with the increase in the number of warm days and will reach 1 month, on average, with ~180 warm days, but WSs with a duration of over 2 months will also be possible.

From a seasonal perspective, the analysis of the yearly number of cold and warm days can be performed for winter (DJF), spring (MAM), summer (JJA), and autumn (SON), as illustrated in Figure S4 for each 30-year period. The lowest numbers of cold days belong to the 1995–2024 period for all the seasons and it is worth noting that in the early period, the variability between the seasons was higher than in the last 30s. On the other hand, warm days (Figure S4b) have remarkably increased in the last decades in all seasons, in particular in summer, and to a lesser extent in winter.

An interesting aspect is to analyze the variability of the occurrence of the first and last CS and WS considering as thresholds the 10th and 90th percentiles of the daily temperature calculated for each calendar day over the reference period 1901–2000. These values are 4.6° and 24.7 °C, respectively. Then, for each year from 1725 to 2024, it is possible to extract the first and last dates when these thresholds were exceeded, i.e., four dates per year. Thus, these dates provide an indication for each year of the first and last winter-like days and the first and last summer-like days. The result of the analysis is shown in Figure 7, where the years have been grouped in 30s. It appears that, over the last decades, the winter has been receding (Figure 7a), as the first day with a temperature below 4.6 °C on average occurs at the end of November, advancing by 5–10 days with respect to the past, and the last winter-like day occurs between 20 and 25 February, while in past centuries, they typically happened at the beginning of March or even mid-March. Contrariwise, the summers are advancing (Figure 7b), as the first day with a mean temperature above 24.7 °C presently occurs on 5–10 June, but previously, it was at the end of June, and the last summer-like day is at the end of August, advancing by 10–15 days with respect to the beginning of the series.

The IDF curves of Figure S5 show the 5-, 10-, 25-, 50-, 100-, and 300-year return periods for the highest temperature anomalies for the 1725–1994 and 1995–2024 periods, thus separating the most recent phase, which experienced the largest warming. For example, the RP for a WS of a 6-day duration and an anomaly of +7.0 °C is 50 years from 1725 to 1994 but less than 5 years from 1995 to 2024. All the curves decrease with increasing durations, implying that it is harder to sustain large anomalies for many days. Considering only the summer mean temperatures, the difference between the RPs related to the two periods is again evident (see Figure S6). For example, the RP of a 3-day summer WS with a mean temperature of 30 °C is over 100 years in the 1725–1994 period and only 5 years in the 1995–2024 period. On the other hand, Figure S7 shows the same analysis as Figure S6 but for CSs. Once again, the anomalies approach zero as the duration increases. In this case, the comparison between the 1725–1994 and 1995–2024 periods is more difficult, as few events in the most recent years had long CSs. Nonetheless, a shift in the curves can be seen: for example, the RP of a CS with a 5-day duration and −8.0 °C mean anomaly increased from 5 to 10 years to over nearly 300 years. The selection of the winter CSs shows the difficulty of observing severe frost in the recent period (Figure S8). An average temperature of −5 °C over 4 days happened once in 10 years from 1725 to 1994 but only once in over 50 years from 1995 to 2024. Values below –7 °C are virtually impossible nowadays for any duration.

The periodicities of the yearly number of CSs and WSs are shown in Figure 8a and Figure 8b, respectively. A periodicity of about 35 years for CSs emerges over the 95% confidence level threshold, stronger in the early period than in the modern era in which it fades or possibly shifts to lower frequencies (Figure 8a). A less sharp periodicity peaking at 35 years is also found for the WSs, which fall within the detectable frequency range (Figure 8b).

The list of daily extreme temperatures for each month for the 1725–2023 period was reported in Stefanini et al., 2024 [2]. Figure 9 shows, for each year starting from 1726, the number of times (N) when the daily mean temperature reached different lowest or highest values with respect to the observations of the previous years according to the calendar day. The beginnings of the series are noisy, as it is quite easy to have new extreme observations if a series is short. In the last decades, the cold extremes are not approached (Figure 9a), and the last lowest daily observation was recorded at the end of May 2006. Starting from the 1990s, it has become easier to reach new highest temperatures (Figure 9b), well above the expected theoretical decay proportional to 1/n (n is the number of years since the beginning of the series) [48]. Two main peaks are present in 2003 and 2011, when 28 and 26 new daily highest extremes were recorded, respectively. In the 21st century, 8–9 new highest daily values are reached every year on average, more than the 1–2 theoretically expected.

The time distributions of the lowest and highest daily mean temperatures have opposite behaviors. The yearly absolute number and percentage of the lowest and highest daily extremes for each 30-year interval, from 1725 to 2024, are shown in Figure 10a and Figure 10b, respectively. The lowest daily temperatures are more uniformly distributed than the highest ones, as 12–18% of them occurred from 1755 to 1874, while over 50% of the highest daily temperatures belong to the 1995–2024 period. These last 30 years were also characterized by the lowest percentage of lowest daily temperatures, i.e., 1.6%.

The projections of the temperatures in the Padua region until the end of the 21st century can be provided by EURO-CORDEX. In Figure 11, the 300-year temperature observations are completed until 2084 according to three different scenarios based on mitigation efforts, and the mean temperature anomaly with respect to the 1901–2000 period is represented. In the strong mitigation RCP2.6 scenario, with the CO₂ concentration at the end of the century at the same level as today (420 ppm), the temperature will stabilize at the present-day level, +1.5° above the 1901–2000 period. In the RCP4.5 scenario, with the CO₂ concentration at the end of the century at 538 ppm, the anomaly of +2.5 °C could be reached. Finally, in the no-mitigation RCP8.5 scenario, with 900 ppm of CO₂ concentration at the end of the century, the anomaly could rise to +4.5 °C. The largest warming should be in summer, with an anomaly between +1.6° and +5.3 °C depending on the scenario, and the lowest in spring, between +1.2° and +3.5 °C at the end of the century. Winter and autumn should be characterized by anomalies between +1.3 °C and +4.6 °C.

3.2. Snowfall

In the case of enhanced winter warming, snowfall would become rare in Padua. The trend of the monthly number of snowy days, i.e., days with snow, neglecting accumulation, is represented in Figure 12. In the subsequent analysis, only the months from November to March (NDJFM) have been considered, when snowfall typically occurs in Padua. Snow can also occur in April, as shown in Figure 12, but it has become very rare over the last century, therefore this month has been excluded from the study.

The yearly number of snowy days in Padua had its maximum of 32 days occurring in the winter of 1783/1784 (see Figure 13a). Since then, it has decreased from 10.4 in the 1785–1814 period to 4.4 in the 1995–2024 period. Note that it is not uncommon to have entire seasons without any snow occurrence. The number of days that elapsed from the first to the last snow occurrence also had its peak in the late 18th century (Figure 13b). The earliest snowfall of the entire series is the event of 27 October 1946, while the latest occurred on 26 April 1784. Figure 13c shows the Gaussian fit of the frequency distributions of snowfall in Padua, from 1725 to 2024, according to the mean daily temperature.

Figure 14a shows the number of snowy days versus the mean temperature from November to March (NDJFM), similar to the calculation performed by [49]. The linear regression indicates that, on average, a snowy day is lost every time the NDJFM temperature rises by half a degree.

The role of total precipitation amount in the relation between temperature and snow has been investigated. In the scatter plot of the mean temperature anomaly vs. the normalized precipitation anomaly (Figure 14b), the values have been grouped according to the corresponding number of snowy days. Four groups (0–3, 4–8, 9–15, and 16 or more) have been considered and the mean values of each group are shown (big circles). The dependence on normalized precipitation anomalies is negligible: performing a Student’s t-test on the normalized precipitation anomaly values belonging to each pair of these four groups, no statistical difference can be found (at the 95% significance threshold). On the other hand, the temperature dependence is more robust, as the t-tests applied to the temperature values are significant (except for the pair “9–15” and “≥16”, due to the scarcity of the samples).

As previously mentioned, over the last 30 years, the mean number of snowy days in Padua is 4–5 per year, but during the late 18th and the beginning of the 19th century, it was about twice as large. The relation with temperature has been explored, as well as that with atmospheric circulation. The reanalysis of daily resolution started in 1806, provided by NOAA [51], and for the previous period, only monthly reconstructions are available [23,52,53], which cannot provide enough information on the large-scale circulation during a snowfall event. As already said, few pressure series are available at the daily level covering the whole 1725–2024 period: Padua itself, starting from 1725; London Heathrow (United Kingdom) from 1692; and Uppsala (Sweden) from 1722 (see Figure 2). The Paris series, which is the oldest, starting in 1670, has been excluded because it has gaps in the 18th century and Paris is not too far from London, whose series is complete. By using these three long series, it is possible to obtain hints about the synoptic situation for Europe. The k-means algorithm has been exploited to cluster the daily sea level pressure observations of Padua, London, and Uppsala during the snowy days in Padua. Four clusters have been selected after optimization by means of an elbow plot, which shows the within-cluster-sum-of-square values corresponding to the different numbers of clusters, and by visual inspection, a larger number of clusters would introduce similar patterns. The synoptic situations corresponding to these four clusters are shown in the generic representations of Figure 15. Cluster #1 (Figure 15a) includes the snowy days in Padua caused by the interaction of relatively warm air coming from the Atlantic Ocean and the cold continental air pushed westwards by a high pressure on northeastern Europe. Cluster #2 (Figure 15b) has low pressure in central Italy and high pressure in western Europe, conveying cold air to the Mediterranean region. In cluster #3 (Figure 15c), a large amount of low pressure is centered on northern Europe, providing cold. Finally, cluster #4 (Figure 15d) is the coldest one, conveying continental air and cyclonic cut-off systems westwards. Clusters #2 and #4 are similar to those obtained by D’Errico et al. (2022), which considered the cold and snowy spells for the whole of Italy [14].

The synoptic situation represented by cluster #4 is the most frequent when a snowfall event occurs in Padua, whereas the one corresponding to cluster #1 is usually the least frequent, as shown by its mean number of occurrences in Figure 16a. This synoptic situation was already found to be linked with snowfall events in Italy in [54]: “snowfalls over Italy are typically associated with cold outbreaks from northeastern Europe, such as may develop with a blocking high formation extending from Spain into Scandinavia”. In the context of a general decrease in the number of snowy days since the mid-19th century, from 1936 to 1965 the number of snowy days that happened in correspondence with the cluster #4 circulation type was particularly high, though not as high as in the transition from the 18th to the 19th century. Extending the analysis to all the winter days (from November to March), it is possible to investigate any trend in the synoptic situations over the last three centuries independently from snowfall in Padua. The classification of the remaining days was based on the Euclidean distance of the daily pressure values from the pressure values corresponding to the four centroids calculated for the snowy days. Each day was associated with one of the four clusters according to its lowest distance. However, if this distance is larger than the mean distance of the snowy days of that cluster with respect to the corresponding centroid plus one standard deviation, then the day becomes unclassified. This ensures that not all the days are forced to be grouped into this four-cluster classification, as there may be different synoptic situations with respect to those shown in Figure 15. Thus, Figure 16b exhibits the mean number of days for each cluster, independently from the snowfall, for each 30-year period from 1726 to 2024. Note that these values are larger than those of Figure 16a, because these synoptic situations are not sufficient to generate snow in Padua, as other factors such as temperature near the ground and in the upper atmosphere, humidity, precipitation presence and intensity, and wind are involved. Cluster #1 has its minimum in 1966–1995 after the maximum of the period before, cluster #2 has no evident trend but it was generally higher in the past, cluster #3 has a slight increase over the last two centuries, and cluster #4 has been decreasing since the end of 19th century, reaching the minimum in the latest period.

Overall, summing all four clusters’ occurrences for each period, over the last 60 years, the number of large-scale patterns that cause the snow in Padua has reached lower values than in the past, losing about a week of favorable situations with respect to the 1936–1965 period (black line in Figure 17a). Another way to look at this is in Figure 18, which shows the yearly number of snowy days, Ns, versus the number of favorable snowfall synoptic situations, Nf, in winter; Nf has to decrease by 7 days for one day of snowfall to be lost, on average. Therefore, the general scarcity of snowfall events over the 1966–2024 period is probably due to a combination of an increase in temperature and a decrease in the number of synoptic situations that could potentially lead to snow. This can also be recognized from Figure 17b, showing Ns vs Nf for different classes of mean temperature, where the relation between mean temperatures and Nf is evidenced. For each interval of temperatures, Ns increases with increasing Nf, and for each Nf, the number of Ns increases with decreasing mean temperature.

The UHI effect can also play a role in the trend of the frequency of snowy days in Padua. If industrial areas are included, the city expanded by over 5 times between the 18th century and the present day. This expansion has been related to people’s thermal comfort and the possibility of heat-related risk mitigation in summer [55,56,57,58,59,60,61]. Noro et al. (2013) [56] found that in the 1994–2011 period during winter, the difference between the temperature in the Padua city center and in the surrounding rural areas was ~0.5 °C. Their preliminary results indicate that this difference increased by 0.3 °C over the 1994–2024 period, from +0.3° in 1994 to +0.6 °C in 2024, even though the number of inhabitants remained nearly the same. Further studies are necessary to better understand the UHI effect in winter, including the influence of soil consumption.

4. Conclusions

The availability of the 300-year series of meteorological observations in Padua allows the exploration of the entire period from the late Little Ice Age to the modern era and in the context of a long-term perspective. The results indicate that cold spells have been declining over the last decades, whereas warm spells are increasing in duration, intensity, and number. Two case studies for the most extreme spells have been explored, February 1740 and April 1755, respectively, highlighting their exceptionality. Projections derived from climate models point out the possibility of remarkable warming scenarios in the next decades, with a further increase in the number of days with positive temperature anomalies with respect to the 1901–2000 reference period.

The analysis of snowfall events, often associated with cold spells, revealed that over the last 60 years, the number of snowy days in Padua decreased, reaching a yearly frequency that is nearly halved with respect to the beginning of the 19th century. This decrease can be related to the increase in the mean temperature but also to the decrease in the number of synoptic situations in Europe conducive to snowfall events in Padua. This latter change is likely another effect of global warming, which not only causes an increase in temperature but also induces variations in large-scale circulation.

Although the results confirm at a local scale what is happening in the Mediterranean and European context, by considering a 300-year time series, they allow for a better and more robust assessment of the magnitude of the change. As the use of a long-term dataset provides a more reliable and precise basis for drawing relevant conclusions, further efforts to reconstruct long meteorological series will be made. The recovery and reconstruction of other Italian series, such as Pisa and Rome, and the addition of the missing meteorological variables in the Bologna series are in progress.