Precipitation variability and change in the Calabria region (Italy) from a high resolution daily dataset

Authors


Abstract

The present study aims at improving data availability and quality for the last 80–90 years for daily precipitation in the Calabria region (southern Italy). First, the original database was homogenised and the gaps filled in for 129 daily rain gauges for the 1916–2006 period. Then, precipitation variability and change were evaluated at an adequate spatial resolution. Monthly and annual total precipitation (P), number of wet days (WDs), and precipitation intensity (PI) were calculated for each series. With regard to the monthly total precipitation a general negative trend, albeit not everywhere significant, was detected, in particular for the autumn–winter period, while in summer the tendency was toward an increase in total precipitation. The monthly behaviour of WDs was not very different from that observed for P: a diffuse negative trend was detected in most months, particularly evident and significant in January, with the exception of April and the summer months, for which the tendency was toward an increase. Regarding the PI, a general negative and often significant trend was found for the entire region and for almost all the months, except summer. Attention was also focused on tendencies in the different PI categories, revealing negative trends in high-intensity categories, especially coming from the winter season.

Finally, running trend analysis revealed that the previously discussed tendencies were not persistent throughout the series length, but depended on the period examined. This important aspect should be taken into account when different results based on different time windows are compared. Copyright © 2010 Royal Meteorological Society

1. Introduction

One of the most noticeable consequences in a changing climate is water cycle modification (Allen and Ingram, 2002; Huntington, 2006) and precipitation plays a key role in these processes (Mariotta et al., 2002). Thus, research into precipitation has increased in recent decades not only for scientific reasons but also because precipitation is one of the most important climate elements directly affecting human society, economic activities, and natural systems (IPCC, 2007).

Long-term trends have been observed in precipitation amount over many large regions. Significantly increased precipitation has been observed in the eastern parts of North and South America, northern Europe, and northern and central Asia. Drying has been observed in the Sahel, the Mediterranean basin, southern Africa, and parts of southern Asia (Hess et al., 1995; Sharma et al., 2000; Hamilton et al., 2001; Boyles and Raman, 2003; Liu et al., 2008; Lebel and Ali, 2009).

Nevertheless, these global analyses are not useful at local scale or for detailed spatial analysis; in fact, many studies of precipitation patterns have shown high spatial and temporal variability at global (New et al., 2001), regional (Lawrimore et al., 2001; Klein-Tank et al., 2002) and sub-regional scales (De Luis et al., 2000; Brunetti et al., 2006a,b; Norrant and Douguedroit, 2006; González-Hidalgo et al., 2011).

For example, the main characteristic of precipitation around the Mediterranean basin is its extraordinary variability both on a temporal and spatial scale (Lionello et al., 2006; Norrant and Douguédroit, 2006). Such variability is caused, among other reasons, by its latitudinal position between tropical and mid-latitude climates, the influence of weather conditions over the Atlantic Ocean, and also topographical factors. It is marked by a strong annual precipitation cycle oscillating between dry and wet seasons (Dünkeloh and Jacobeit, 2003).

Several studies have contributed to the analysis of precipitation variability over the Mediterranean basin. The results from these studies indicate a precipitation decrease, not always significant, or the lack of a linear trend (De Luis et al., 1998; Lana and Burgueño, 2000; Klein-Tank et al., 2002; Douguédroit and Norrant, 2003; Norrant and Douguédroit, 2003; Brunetti et al., 2006a; González-Hidalgo et al., 2011). Most of the Mediterranean region has experienced decreasing winter precipitation during recent decades (Piervitali et al., 1997; Palutikof et al., 1999). In particular, precipitation decrease is evident in large parts of the eastern Mediterranean area, where Schönwiese et al. (1994) reported a pronounced significant trend towards a drier winter climate for the period 1961–1990.

A general drying is also discernible over most of the southern Mediterranean basin. A decrease in winter precipitation over Greece was found by Xoplaki et al. (2000) and Feidas et al. (2007), although significant only over the northern and eastern parts and in the western mountainous regions. Similar tendencies have been reported for the Mediterranean coast of Turkey (Türkes, 1996; Kadioglou, 2000).

In Spain, Mosmann et al. (2004) revealed an increasing trend in precipitation during the summer season of the period 1960–1980. González-Hidalgo et al. (2009) showed that in Spain trends for winter months (December–January–February) are dominated by an East–West gradient with a latitudinal temporal shift. Positive trends are mainly located in coastal areas, whilst negative ones predominate inland.

In Italy, a decrease in total precipitation has been highlighted in Brunetti et al. (2006a) over the past two centuries from a dataset of 111 homogenised precipitation series. This decrease becomes stronger, even if less significant, in recent decades. The signal presents some differences from place to place, with the strongest negative trend in central Italy.

Besides total precipitation, Italy also experienced a decrease in the number of wet days (WDs) and an increase in precipitation intensity (PI) (Brunetti et al., 2000, 2001a,b, 2004). The entity and the significance of these signals present some differences in the different areas into which Italy has been divided. This is mainly due to the high spatial and temporal variability of precipitation and the number of stations that in some cases is too low to discriminate some important spatial peculiarities.

The Fourth Assessment Report (IPCC, 2007) also suggests focusing on detailed sub-regional studies, with particular attention to those areas where water is a scarce resource and demand is high. In particular, given the extreme variability, a dense and robust daily precipitation dataset is needed (Moberg and Jones, 2005; Huntington, 2006). However, generally there are problems with data availability and a dense network of observatories worldwide, with long records of reliable quality, is not always obtainable. There are large areas for which appropriate records do not exist, or where data are not yet available from national archives. However, in recent decades an increasing number of researchers has been involved in gathering available meteorological records for specific regions (López-Moreno et al., 2010).

Besides data availability, data quality is also a fundamental task in climate change analyses. Several studies over the past two decades of the last century have demonstrated that climate variability research is not possible without clear knowledge about the state of the data in terms of homogeneity, the real climate signal in the original series being hidden behind non-climatic noises (Peterson et al., 1998). Homogeneity testing and the adjustment of climatic time series for non-climatic variations are a fundamental part of any climate change analysis (Alexandersson, 1986; Karl and Williams, 1987; Peterson and Easterling, 1994; Jones, 1995; Heino, 1997; Peterson et al., 1998).

In this context, the present study aims at improving data availability and quality for the last 80–90 years for daily precipitation in the Calabria region (southern Italy). First, the original database was homogenised and the gaps filled in for 129 daily precipitation stations, then precipitation variability and change were evaluated at an adequate spatial resolution, with special focus on tendencies in the different PI categories.

2. Study area and data

Calabria is a region in the furthest south of Italy with an area of 15 080 km2 and a perimeter of about 818 km. Calabria has a typical Mediterranean climate in the coastal zones with mild winters and hot summers with few precipitation events. The Ionian side in particular, which is affected by air masses coming from Africa, has high temperatures with short and heavy precipitation; the Tyrrhenian side is influenced by western air masses and presents milder temperatures and many orographic precipitation. In the inland zones there are colder winters (with snow) and cool summers (with some precipitation) if compared to the costal zones (Caloiero et al., 1990). In Calabria high-quality datasets have been collected from 1916 by the former Italian Hydrographic Service. The data used in this study are a set of daily precipitation series registered in Calabria and relative to the period 1923–2006. When the number of years of observations was too low for statistical purposes or there were too many gaps in the series (<50 available years of data), or when the series ended before the year 2000, the station series were discarded from the dataset. As a result, 173 rainfall series over a total of 311 were finally selected, with a density of one station per 87 km2 (Figure 1).

Figure 1.

Map of the stations (circles indicate stations which passed the homogenisation procedure; triangles indicate stations rejected by the homogenisation procedure). Black dots indicate the grid cells

3. Database set-up

3.1. Homogenisation

In the last decade, there has been widespread agreement that time series of meteorological data cannot be used for climate research without clear knowledge about the state of the data in terms of homogeneity. The scientific community has become aware of the fact that the real climate signal in the original series of meteorological data is generally hidden behind non-climatic noise caused by station relocation, changes in instruments and instrument screens, changes in observation times, observers, and observing regulations, algorithms for the calculation of means, and so on.

At present, among the various methods to solve homogeneity problems, there is not a single objective one, and the choice of the most suitable procedure is strictly related to the dataset characteristics (metadata availability, station density, and so on) and to the examined region (Peterson et al., 1998; Aguilar et al., 2003). The importance of this topic led in 2007 to the setting up of COST Action (COST Action ES0601 ‘HOME’: Advances in homogenisation methods of climate series: an integrated approach) with the aim of achieving a general method for homogenising climate and environmental datasets.

The problem is even more relevant if we consider the time series of meteorological parameters at a daily resolution. In fact, few methods have been proposed in the scientific literature to deal with the homogenisation of daily series (Vincent et al., 2002; Brunetti et al., 2004; Kuglitsch et al., 2009; Della-Marta and Wanner, 2006).

One of the biggest problems concerning daily precipitation data is that the series could be affected by two kinds of inhomogeneities: (1) in the precipitation amount and (2) in the number of rainy days. The latter can obviously generate an inhomogeneity in the precipitation amount series. In fact, a period with some non-indicated missing data might be interpreted as a period with an underestimation of total precipitation and might be corrected badly by increasing each single rainy day and originating some erroneous extreme events. So, we decided to check both total precipitation and the number of rainy days separately.

As far as the inhomogeneities due to the number of WDs are concerned, we observed that their signal often disappeared from the homogeneity test if a different threshold to define a day as rainy was used (1 mm rather than 0.1 mm). This is due to instrument resolution: sometimes improvement in precipitation gauges leads to a higher number of rainy days with low precipitation amount (Nicholls and Murray, 1999). Therefore, the choice of a threshold to define the rainy days is not a negligible problem in this kind of analysis. In all the statistical analyses, precipitation events lower than 1 mm were discarded and considered as 0 mm events. Some other inhomogeneities in the number of rainy days were adjusted by the precipitation homogenisation procedure, because, by correcting total precipitation, some days with below threshold amount were increased to above threshold and vice versa, and the break signal in the test curve of the number of rainy days disappeared.

Checks of both rainy days and total precipitation were considered, and it was decided to correct only inhomogeneities in precipitation amounts corresponding to periods without problems in rainy days or with problems that could be solved by total precipitation homogenisation. Stations affected by inhomogeneities in the number of rainy days not homogenisable with one of the previously discussed procedures were discarded from the dataset.

The homogenisation approach used in this work was the same as that discussed in Brunetti et al. (2006a), but adapted to the daily resolution. Instead of using one single reference series (obtained, to say, as an average series of the neighbouring stations), each series was tested against other series, by means of a multiple application of the Craddock test (Craddock, 1979), in sub-groups of ten series. The test is based on the hypothesis of the constancy of precipitation ratios. The break signals of one series against all others are then collected in a decision matrix and the breaks are assigned to the single series according to probability. This system avoids the maintenance of the trend and an inadmissible adjustment of all series to one or a few ‘homogeneous reference series’.

Once it has been decided to correct one break, the series used to estimate the adjustments are chosen among the neighbouring series that result homogeneous in a sufficiently long sub-period centred on the break year, and that correlate well with the candidate one (the series with the highest correlation coefficients were chosen from the group of 10 reference series). Several series are chosen to estimate the adjustments to improve their stability and to avoid unidentified outliers in the reference series from producing bad corrections. Moreover, it often happens that homogeneous sub-intervals between two detected breaks are so short that the signal-to-noise ratios of the adjustments obtained with only one reference series are very low. So, using more than one series allows us to correct a large number of short sub-periods that would have to be left otherwise unchanged.

The adjustments from each reference series are calculated on a monthly basis and fitted with a trigonometric function in order to obtain daily corrections without discontinuities from 1 month to another. The final set of daily adjustments is then calculated by averaging all the yearly cycles, excluding from the computation those stations whose set of adjustments shows an incoherent behaviour compared with the others. When a clear yearly cycle is not evident, the adjustments used to correct the monthly data are assumed as constant throughout the year and are calculated as the weighted average of the monthly values, where the weights are the ratios between monthly mean precipitation and total annual precipitation.

From the results of the homogenisation procedure, 87 of the total of 173 daily precipitation series resulted homogeneous with regard to both total precipitation and number of WDs, 42 were homogenised, and 44 were discarded because of very low quality or because of unsolvable problems in the number of WDs. So, the final dataset useful for data analysis consisted of 129 stations (Figure 1).

Finally, a total of 104 breaks were corrected in the whole dataset, with 2.5 breaks per homogenised series that decreased to 0.8 breaks per series if we consider all 129 stations. In 18% of the corrections no annual cycle was evident, and corrections were assumed as constant through the year.

3.2. Gap-filling procedure

Besides the data homogeneity issue, the handling of missing values in meteorological time series is also of importance to climate research. Lack of data affects the early instrumental period just as much as it affects the most recent one, because of, for instance, occasional interruptions in automatic stations, instrument malfunctions, and network reorganizations. Excluding periods with missing values from data analyses may disregard valuable information and induce biases in many climate investigations. For this reason, a number of interpolation techniques have been developed to estimate missing observations in climatic time series, mainly on a monthly or seasonal basis. However, methods for filling-in missing values on a daily timescale are scarce and show marked errors, especially when precipitation is of concern, because of its large space and time variability (DeGaetano et al., 1995; Xia et al., 2001). The problem in this case is twofold, as both time location and rainfall amount of each single-day event must be reconstructed. Overestimation of the number of rainy days is a typical drawback of commonly used filling-in techniques, such as weighting-based spatial interpolation methods (Shepard, 1968; Young, 1992; Daly et al., 1994; Lloyd, 2005; Garcia et al., 2008), and their several variants and adds-on (Teegavarapu and Chandramouli, 2005; Suhaila et al., 2008). Multi-linear regression outperforms most of the available filing-in methods (Eischeid et al., 2000) but is plagued by the same limitation. Furthermore, regression-based approaches generally underestimate intense precipitation events, thereby modifying the rainfall probability distribution on a daily time scale.

To obviate the above biases, the estimation of missing data in this study was performed by exploiting a recently developed two-step procedure (Simolo et al., 2010) which preserves both the correct event time location (wet/dry days) and the statistical properties of daily precipitation series. The method is based on the fit to daily data of the Gamma distribution, a widely used guess in modelling daily precipitation (Bradley et al., 1987; Groisman et al., 1999; Jones et al., 2004) which was further validated by the Kolmogorov–Smirnov test (Simolo et al., 2010).

The algorithm first estimates rainfall occurrence in the target series in terms of probabilities, using a weighted average of synchronous values from a cluster of reference series. The weighting factors are functions of the relative distance, elevation, and angular distribution of the reference stations with respect to the target one. Actual precipitation events are selected using a time-dependent threshold which discards those low-probability values which most likely correspond to dry events. This enables us to recover the correct number and time location of wet/dry days in the target series, thereby drastically reducing the error due to false alarms.

Second, the rainfall amount of wet-classified days is reconstructed by a multivariate fit with ordinary least squares, and the generated values are subsequently rescaled to recover the daily probability distribution of the original series. Imposing the correct time-dependent distribution in wet-classified days significantly improves the reconstruction of intense precipitation events, thus obviating their systematic underestimation induced by multi-linear regression.

The completion procedure was applied to the final dataset of 129 station series. The reconstructed period spans from the first available year of each series up to June 2007, in which period a total amount of 11.4% of data was missing and required a reconstruction. The final database dimension is shown in Figure 2.

Figure 2.

Development of the final homogenised precipitation network

4. Data analysis

Monthly and annual total precipitation (P), number of WDs and PI (PI—evaluated as the ratio between P and WDs) were calculated for each series.

To facilitate the spatial analysis of precipitation we decided to interpolate our dataset onto a regular grid.

The grid resolution was chosen near to the mean interstation distance of the selected 129 rain gauges, which is about 8 km. Then, a 1/10 of degree resolution gridded version of the monthly dataset was constructed. Before interpolating the station data on the grid cells, each series was converted into anomalies (multiplicative anomalies for P and PI, additive anomalies for WDs), normalising each monthly value by its average estimated over the 1961–1990 period. This choice is justified by the fact that absolute values present important spatial gradients, and anomalies avoid the different starting year of each series to bias the interpolation result, as suggested by Mitchell and Jones (2005).

A radial weight with a Gaussian shape was used for the stations involved in each grid point's estimation. The radial weight of the i-th station for the evaluation of the (x, y) grid cell is as follows:

equation image(1)

with

equation image(2)

where i runs over the stations and di(x, y) is the distance between the station i and the point (x, y) for which the local record is being estimated. With this choice of the parameter c, we get weights of 0.5 for station distances equal to from the point (x, y), where is defined as the mean value of the distances among each grid point of the data set and its closest neighbour (i.e. about 10 km). The computation is performed by considering all stations within a distance of 2. In this way, the sets of stations involved in each grid cell computation are completely independent every four grid cells. This choice, together with the high resolution of the grid, avoids the undesired exchange of information among different climatic regions and, in particular, between the different sides of the most important mountain chains that play a key role in the separation of different precipitation regimes.

Besides the radial weight, an angular weight which accounts for the geographical separation among the sites with available station data was also considered. It has the following form:

equation image(3)

where θ(x, y)(i, l) is the angular separation of stations i and l with the vertex of the angle defined at grid point (x, y), and equation image is the radial weight as defined in Equation (1). The introduction of this angular weight avoids the undesired over-weighting of the areas with the highest station density in the evaluation of the grid cell.

The final weight is the product of the radial and the angular weight.

A total of 194 grid cells were estimated (Figures 1 and 2).

The dataset was analysed for trend, and significance assessed with the Mann–Kendall non-parametric test. Use and computation of this test has been well described by Sneyers (1990). The slopes of the trends were calculated by least square linear fitting.

Figures 3–6 show the results of the application of the trend analysis to the yearly and monthly series of P, WDs, and PI.

Figure 3.

Annual trend maps for: (a) P, (b) WDs, and (c) PI. Squares dimension indicates the significance level of the trend: large squares p < 0.1, small squares otherwise. This figure is available in colour online at www.wileyonlinelibrary.com/journal/joc

Figure 4.

Monthly trend maps for P. Squares dimension indicates the significance level of the trend: large squares p < 0.1, small squares otherwise. This figure is available in colour online at www.wileyonlinelibrary.com/journal/joc

Figure 5.

Monthly trend maps for WDs. Squares dimension indicates the significance level of the trend: large squares p < 0.1, small squares otherwise. This figure is available in colour online at www.wileyonlinelibrary.com/journal/joc

Figure 6.

Monthly trend maps for PI. Squares dimension indicates the significance level of the trend: large squares p < 0.1, small squares otherwise. This figure is available in colour online at www.wileyonlinelibrary.com/journal/joc

4.1. Total precipitation

On a yearly basis, P shows a significant negative trend in the 1923–2006 period, the general tendency of all grid points being well summarised by the interquartile range (IQR) (i.e. the values of the second and the fourth quartile of the trend distribution) whose range is from − 1.9 to − 3.4%/decade (Figure 3(a)).

With regard to the monthly total precipitation a general negative trend, albeit not significant for the entire region, was detected for the 1923–2006 period, in particular for the autumn-winter period (Figure 4).

A negative and significant trend was detected particularly in January and November, although elaborations show a high and locally significant negative trend also in June (IQR ranging from − 4.3 to − 9.7%/decade), with a maximum precipitation reduction in the eastern part of the Pollino Massif, between the Sila Plateau and the Ionian sea, and between Serre Chain and the Tyrrhenian Sea. In January, a decrease was detected in the whole region, with an IQR ranging from − 3.8 to − 6.5%/decade, with significant values in all the Tyrrhenian side, in the Crati basin and between the Serre Chain and the Sila Plateau (‘Stretta’ of Catanzaro). In November the IQR ranges from − 2.9 to − 10%/decades, with the maximum reduction identified in almost all the Ionian side of the region.

Conversely, summer precipitation shows a positive trend over the whole region, with an IQR ranging from + 2.8 to + 4.0%/decade. In July, significant trends were detected on the Tyrrhenian side, in the Aspromonte, in the low valley of the Crati basin, and in the ‘Stretta’ of Catanzaro. In August, a positive trend was identified in the Tyrrhenian side of the region, in the Sila Plateau up to the Ionian Sea, in the southern part of Aspromonte, and in the eastern part of the Pollino Massif. Locally, a positive trend was also detected in September and in April.

4.2. Wet days

On a yearly basis, WDs show a significant negative trend in the ‘Stretta’ of Catanzaro and a significant positive trend in the western part of the Pollino Massif (Figure 3(b)).

The monthly behaviour of WDs (Figure 5), in the period 1923–2006, is not very different from that observed for P. Indeed, a general negative trend, albeit not relevant, was detected: in particular, in January the WD series has a negative trend (IQR ranging from − 0.2 to − 0.4 days/decade) significant in all the Tyrrhenian side of the region, in the Crati basin, in the eastern part of the Pollino Massif, and in the ‘Stretta’ of Catanzaro.

A strong positive and significant trend in WDs was detected for the entire region in the summer period and in particular in the months of July, August, and September, with an IQR ranging from + 0.1 to + 0.2 days/decade for July and August and from + 0.2 to + 0.3 days/decade for September, with the maximum increase in the western part of the Pollino Massif. A positive trend was also detected in April (IQR ranging from + 0.2 to + 0.3 days/decade), with a maximum increase of the WDs in the western part of the Pollino Massif and in the coastal area of the Strait of Messina.

4.3. Precipitation intensity

As for P, on a yearly basis PI shows a significant negative trend in the 1923–2006 period, in particular in mountainous areas; thus the general tendency is towards a slow decrease (Figure 3(c)).

On a monthly basis, a general negative and significant trend was detected for the whole region and for almost all months (Figure 6).

The clearest negative trend was detected in the winter period, from November to February. In November a negative trend was detected in the Ionian side of the region, in almost all the Pollino Massif, in the Sila Plateau, in the Serre Chain, and in the Aspromonte, the IQR ranging from − 2.9%/decade to − 8.4%/decade. Conversely, in January the negative trend was identified in the Tyrrhenian side of the region. In December, the negative trend was identified in the western part of the Pollino Massif and in the valley of the Crati basin while in February in almost all the mountainous areas.

Elaborations on summer precipitation show an opposite trend to the previous ones. In fact, even if a negative trend was identified in the month of June, with a significant negative trend in the central part of Calabria, a strong positive and significant trend was detected in the months of July and August (with IQR ranging from + 1.5 to + 6.0%/decade and from + 1.4 to 3.9%/decade, respectively), with a maximum increase of the PI on July in the northern Tyrrhenian side and in the ‘Stretta’ of Catanzaro.

However, it is to be highlighted that the summer positive trend, the one concerning P and WDs in particular, does not give a large contribution to the annual trend, summer being the driest season.

4.4. Yearly regional series

To evaluate the temporal stability of the trend signal, a mean yearly regional series was estimated for each variable (P, WD, and PI) and analysed for trend with a running approach.

Temporal evolution of P, WDs, and PI average series are shown in figure 7, together with an 11-year window 3-year σ Gaussian low-pass filter.

Figure 7.

Annual average regional series for: (a) P; (b) WDs; and (c) PI. The series are displayed together with an 11-year window 3-year σ Gaussian low-pass filter. This figure is available in colour online at www.wileyonlinelibrary.com/journal/joc

The slopes of the trends were estimated within temporal windows of widths ranging from 20 years up to the entire series length. The trends obtained were plotted for better visualization in graphs where the y axis represents the window width, and the x axis the first year of the window the trend refers to. The value of the trend is represented by the colour of the corresponding pixel and the significance level by the dimension of the square: large squares are plotted for p < 0.1, small squares otherwise. These figures capture the whole possible spectrum of significant trends present in the series, thus providing the best possible detailed information.

P (Figure 8(a)) shows quite a stable signal toward a light decrease all over the examined period. Conversely, WDs and PI (Figure 8(b), and (c), respectively) present different trend signs depending on the considered sub-period: WDs significantly increase in the first 30–40 years and significantly decrease in the last 50–60 years; in a similar way, PI significantly decreases in the first part of the series and increases in the last one, with significant values only between the 1950s and 1990s. This is consistent with the results obtained by Brunetti et al. (2001a) for southern Italy in the 1951–1996 period with a lower density dataset.

Figure 8.

Running trend analysis for annual mean regional series of: (a) P; (b) WDs; and (c) PI. The y axis represents the window width, and the x axis represents the first year of the window over which the trend is calculated. Squares dimension indicates the significance level of the trend: large squares p < 0.1, small squares otherwise. This figure is available in colour online at www.wileyonlinelibrary.com/journal/joc

5. Precipitation categories

To investigate better the trend in precipitation regimes, the daily precipitation were aggregated into categories of different daily intensities based on equally spaced quantiles from the 10th to the 90th (PC01, PC02, …, PC10) along with the most extreme 95th and 99th percentiles (PC95 and PC99).

The 12 quantile thresholds were estimated for each Julian day of the year from the associated Gamma distributions fitted to the daily data.

The shape and scale parameters of the Gamma distributions were estimated for each Julian day of the year on the basis of the maximum likelihood, using the data sample provided by the precipitation events (greater or equal to 1 mm/day) within a 31-day window centred on that day and considering all the data from 1961 to 1990.

Based on these threshold sets, the following seasonal and annual statistics were extracted from each station series:

  • 1)Total precipitation falling into each category
  • 2)Number of events per category

As for P, WDs, and PI, to facilitate the spatial analysis of the trend, all these statistics were interpolated onto a regular grid following the same procedure discussed in Section 4.

On an annual scale, the precipitation amounts falling into the highest categories show a negative trend over the 1923–2006 period. The stronger the trend the higher the category is (Figure 9), with some areas reaching decreases of about − 20%/decade for PC99. At the same time, less intense events show a generalised positive trend (with few exceptions for the lowest category in the area between the Sila Plateau and Serre Chain) even if they are statistically significant only for a few areas (Figure 9).

Figure 9.

Annual trend maps of the precipitation amount falling into the 12 precipitation categories. Squares dimension indicates the significance level of the trend: large squares p < 0.1, small squares otherwise. This figure is available in colour online at www.wileyonlinelibrary.com/journal/joc

The strongest contribution to the negative trend in high-intensity categories comes from the winter season and, to a lesser extent, from autumn (figure not shown). Conversely, in summer almost all categories, except the three highest ones above the 90th-percentile threshold (for which no clear trends can be detected), display a positive trend. This is consistent with the precipitation increase observed in summer months.

The same pattern is evident in the number of events (Figure 10). The frequency of high-intensity events decreased over the period 1923–2006 (with the strongest decrease of − 0.7 days/decade in PC10 between the Sila Plateau and Serre Chain), while the frequency of those falling into low-intensity categories increased over the same period. It should be remembered that PC95 and PC99 trends are based on a lower fraction of the total events (5 and 1% respectively, compared to the 10% of all the other categories) and their trend amounts are not comparable in magnitude to those of the 10 deciles.

Figure 10.

As Figure 8, but for the number of events falling into each category. This figure is available in colour online at www.wileyonlinelibrary.com/journal/joc

Thanks to the high spatial coherence of almost all the category trends (for both precipitation and number of events), it is possible to represent the whole region with one single (seasonal and annual) regional average series for each category, and use it to study the temporal stability of the trends in the lowest (PC01, PC02, PC03) and highest categories (PC08, PC09, PC10, PC95, PC99).

The trend slopes were evaluated over sub-periods of variable width (from 20 years to the whole series length) and variable starting year, spanning all the series length. This analysis allows us to evaluate trends over the different sub-periods of the series, but also helps in comparing our results with those of related studies when the periods of data availability are different.

The analysis highlighted an interesting temporal variability in the trend sign. This is particularly true for the low-intensity categories, for which the time series are characterised by a twofold behaviour with an increasing tendency in the first part of the series followed by decreasing tendency after the 1950s (Figure 11).

Figure 11.

Running trend analysis for annual mean regional series of precipitation falling into the lowest and the highest categories. The y axis represents the window width, and the x axis represents the first year of the window over which the trend is calculated. Squares dimension indicates the significance level of the trend: large squares p < 0.1, small squares otherwise. This figure is available in colour online at www.wileyonlinelibrary.com/journal/joc

The low values at the beginning of the series, both in the precipitation amount (Figure 11) and in the number of events (Figure 12), make the trend positive over the whole period.

Figure 12.

As Figure 10, but for annual mean regional series of the number of events falling into the lowest and the highest categories. This figure is available in colour online at www.wileyonlinelibrary.com/journal/joc

The trend sign is more persistent in high-intensity categories for which the trend is always negative with the exception of a 40- to 50-year wide window (from 1950s to 1990s) with some positive but not significant trend values, in the number of events in particular (Figure 12).

This alternation of positive and negative trends is even more evident if the relative contribution of each precipitation category to total precipitation amount is considered. This is shown in Figure 13 where a positive (negative) trend in the contribution of low (heavy) precipitation events over the whole period is clearly evident, while the opposite behaviour emerges if the period from the 1950s to 1990s is considered. This is consistent with the results previously obtained by Brunetti et al. (2001a) for the 1951–1996 period for the southern Italy area, even if Calabria region is only a small part of southern Italy and the previous results were obtained, as already mentioned, from a lower number of stations (75 stations distributed all over Italy with only a couple of stations located in Calabria).

Figure 13.

As Figure 10, but for annual mean regional series of the contribution of the lowest and the highest categories to the total annual precipitation. This figure is available in colour online at www.wileyonlinelibrary.com/journal/joc

6. Conclusions

The real climatic signal in the original series of meteorological data is often hidden behind non-climatic noise, caused by various factors. In this work, to investigate the tendency in rainfall distribution in Calabria (southern Italy), prelimary checks of both rainy days and total precipitation were considered. Applying a tested homogenisation procedure, 87 over a total of 173 daily precipitation series resulted homogeneous with regard to total precipitation and number of WDs, 42 were homogenised, and 44 were discarded because of very low quality or because of unsolvable problems in the number of WDs. The final dataset useful for data analysis (129 stations) was then subjected to a reconstruction procedure, from the first available year of each series up to 2006. The final dataset was then analysed for trend and the following main results were obtained:

  • 1.On a yearly basis, total precipitation shows a significant negative trend over the 1923–2006 period, with an IQR ranging from − 1.9 to − 3.4%/decade. On a monthly scale a general negative trend, albeit not significant for the whole region, was detected, in particular for the autumn-winter period, with the largest IQR in November ranging from − 2.9 to − 10%/decade. Summer months show a positive trend (especially in July, with an IQR ranging from + 8.0 to + 11.7%/decade) over the whole region, anyway providing a little contribution to annual trend because of the low summer precipitation amount.
  • 2.On a yearly basis, WDs show a generalised decrease with few areas with significant trend (negative in the ‘Stretta’ of Catanzaro and positive in the western part of the Pollino Massif) due to different trends in the different months of the year: a general negative trend, although not always significant, during winter months, particularly evident and significant in January, and a strong positive and significant trend in April and summer months.
  • 3.PI (mean precipitation per WD) shows a general negative and significant trend both on a yearly and monthly basis, for almost the whole region, with the exception of some summer months (July and August).
  • 4.The aggregation of the precipitation into categories of different daily intensity shows a negative trend for the precipitation amounts falling into the highest categories, with an IQR ranging from − 7.5 to − 12.5%/decade in PC99, from − 4.8 to − 8.7%/decade in PC95, and from − 4.2 to − 7.4%/decade in PC10. Low-intensity categories, on the other hand, show a generalised positive trend, even if they are statistically significant only for a few areas. The strongest contribution to the negative trend in high-intensity categories comes from the winter season; conversely, in summer almost all categories displayed a positive trend. The same pattern is evident in the number of events.
  • 5.A running trend analysis, aimed at evaluating trends over the different sub-periods of the series, highlights an interesting temporal variability in the trend sign: besides a stable signal towards a slight decrease throughout the examined period in total precipitation, WDs and PI present different trend signs depending on the considered sub-period, with an increase (decrease) in the first part of the series and a decrease (increase) in the last one. The same twofold behaviour in trend sign is evident in different precipitation categories (both in the precipitation amount and in the number of events), with an increasing tendency in the first part of the series followed by a decreasing tendency after the 1950s for low-intensity categories. The trend sign is more persistent in high-intensity categories for which the trend is always negative with the exception of a 40-to 50-year wide window (from the 1950s to 1990s) with some positive, but not significant, trend values, in the number of events in particular.

Acknowledgements

This research was developed in the framework of the ‘Programma Integrato di Voucher e Borse per l’Alta Formazione POR CALABRIA 2000-2006-Misura 3.7b'. The authors also thank EU-COST-ACTION ES0601 ‘Advances in homogeneisation methods of climate series: an integrated approach (HOME)’; the project ‘Sviluppo e verifica di tecniche di downscaling e calibrazione di modelli idrologici, sulla base di una griglia termo-pluviometrica ad altissima risoluzione (1 × 1 km) per gli ultimi 150 anni, per la valutazione dell’impatto dei cambiamenti climatici sulla risorsa idrica' in the frame of the framework programme agreement between the Dipartimento Terra e Ambiente of the CNR and the Centro Euro-Mediterraneo per i Cambiamenti Climatici; and the project ‘Validazione e downscaling di scenari prodotti con modelli climatici attraverso l’utilizzo di una griglia di variabili meteorologiche ad altissima risoluzione' in the frame of the ISAC-CNR/CMCC agreement. Data were kindly provided by the former Italian Hydrographic Service, now ‘Centro Funzionale Multirischi della Calabria’ of the ‘Agenzia Regionale per la Protezione dell’Ambiente della Calabria (Arpacal)'.

Ancillary