Spatial and temporal characterization of climate at regional scale using homogeneous monthly precipitation and air temperature data: an application in Calabria (southern Italy)

In recent years, with growing concerns about climate change impacts, precipitation and air temperature have become the most investigated meteorological variables because of their influence on all natural systems and human activities, including agriculture, water quantity and quality, ecosystems, and human health (IPCC 2013). In particular, decreasing water quality and changes in its spatial and temporal distribution have become one of the most crucial issues for their environmental and socio-economic implications (Incerti et al. 2007; Helldén & Tottrup 2008; Zhou & Wang 2011).

As regards the temporal distribution of rainfall, some decreases in annual precipitation in the Mediterranean area have been detected (Xoplaki et al. 2006; Feidas et al. 2007; del Rio et al. 2011). In Italy, the existence of a decreasing trend in annual precipitation has been identified by many regional studies (Palmieri et al. 1991; Brunetti et al. 2006b, 2012; Pavan et al. 2008; Caloiero et al. 2011a, b), while other regional studies did not detect any significant trend at all (Von Hardenberg et al. 2008; Fatichi & Caporali 2009). On seasonal and monthly time scales, in northern and central Italy, Brunetti et al. (2006a) found a decreasing rainfall trend more evident in spring, while Ventura et al. (2002) and Vergni & Todisco (2011) showed that decreases occur only during autumn and winter. As regards winter precipitation, Toreti et al. (2009) detected a positive trend in central Italy and a negative trend in northern Italy. In southern Italy, and particularly in the Calabria region, different studies (Buttafuoco et al. 2011; Caloiero et al. 2011a, b; Brunetti et al. 2012; Capra et al. 2013; Ferrari et al. 2013) highlighted a decreasing trend in winter and autumn rainfall, and an increasing trend in spring and summer.

As to what concerns temperature, the increase in mean values over the last century has been largely demonstrated in many studies focused on different regions of the world on a large spatial scale (Klein Tank & Können 2003; Vose et al. 2005; Klok & Tank 2009) or at national and local spatial scale (Brunetti et al. 2004, 2006a; Toreti & Desiato 2008; del Rio et al. 2012). The magnitude of trends differs according to the area and the period studied, especially when comparing the rate in increase of maximum and minimum temperatures. In particular, Klok & Tank (2009), for the whole of Europe, Sánchez et al. (2004), for the Mediterranean, and Toreti & Desiato (2008), for Italy, have highlighted the fact that the maximum temperatures have increased at a higher rate than the minimum ones. Donat & Alexander (2012) used a global dataset of daily temperatures to show that both day-time (daily maximum) and night-time (daily minimum) temperatures have become higher over the past 60 years: changes were greater for minimum than for maximum values.

Precipitation and temperature vary more or less continuously in a geographical space and could be regarded as regionalized variables (Matheron 1971). Geostatistics provides a valuable tool for the study of the spatial correlation in data by creating mathematical models, commonly expressed by semivariograms. The interpolation technique of the variable, known as kriging, provides an unbiased linear estimate of a regionalized variable in an unsampled location (Chilès & Delfiner 2012). Examples of geostatistical studies on precipitation and temperature can be found in Hudson & Wackernagel (1994), Diodato & Ceccarelli (2005), Grimes & Pardo-Iguzquiza (2010), Buttafuoco et al. (2011), Aznar et al. (2013), Bajat et al. (2013) and Diodato et al. (2013).

This investigation presents an analysis of the spatial and temporal behaviour of monthly precipitation and temperature in the Calabria region (southern Italy). In the ‘Methodology’ section, the procedures used for the analysis are presented, with particular concern as regards the geostatistical techniques. In the ‘Study area and data’ section, the geographical and climatic features of the Calabria region and the homogenized databases of rainfall and temperature are described. In the ‘Results and discussion’ section, firstly the spatial distributions of the monthly mean values of both precipitation and temperature in the study area for the whole observation period (1916–2010) have been described. Then, in the same section, in order to better investigate the link between the two variables, the Péguy climograph (Péguy 1961) was drawn for different ranges of the station elevations. Moreover, the tendencies of rainfall and temperature at monthly and seasonal scales were investigated separately through the application of the non-parametric Mann–Kendall test. Finally, in order to test the temporal variability of the climatic features on the study area, the Péguy climographs were built with reference to three sub-periods (1921–1950, 1951–1980, 1981–2010).

Meteorological time series cannot be used for climate research without clear knowledge of data homogeneity. Actually, the real climate signal can be hidden behind non-climatic noise caused by changes in instruments and algorithms for statistical calculation, station relocation, changes in observation procedures, and so on (Brunetti et al. 2012). In fact, homogeneity testing and adjustment of climatic time series for non-climatic variations are a fundamental part of any climate change analysis (Alexandersson 1986; Karl & Williams 1987; Peterson & Easterling 1994; Jones 1995). In this work, a homogenization approach similar to that discussed in Brunetti et al. (2012) for precipitation was adapted to the temperature. Due to the low number of temperature stations available for this analysis, as described in the following section, each series was tested against other ones, in sub-groups of five series, by means of a multiple application of the Craddock test (Craddock 1979). When an inhomogeneity (break) was identified, the series used to estimate the adjustments were chosen among the neighbouring series that were homogeneous in a sufficiently long sub-period centred on the break year, and which correlate well with the candidate one. The adjustments from each reference series were calculated on a monthly basis.

Then, the mean precipitation and temperature data for each month of the year were modelled as an intrinsic stationary process using a geostatistical approach, where each datum z(x_α) measured at x_α (x is the location coordinates vector and α the sampling points = 1, …, N) is interpreted as a particular realization of a random variable Z(x_α) (Goovaerts 1997; Wackernagel 2003; Webster & Oliver 2007; Chilès & Delfiner 2012). The set of random variables Z(x₁), Z(x₂), …, constitutes a random function. The set of actual values of Z that comprises the realization of the random function is known as a regionalized variable z(x_α). Interested readers should refer to textbooks such as Goovaerts (1997), Chilès & Delfiner (2012), Wackernagel (2003), Webster & Oliver (2007), among others, for a detailed presentation of the theory of the regionalized variables. An objective of geostatistics is to quantify and exploit spatial pattern of regionalized variables. The quantitative measure of spatial correlation of the regionalized variable z(x_α) is the experimental variogram γ(h), which is a function of the distance vector (h) of data pairs values ⁠. A theoretical function, called variogram model, is fitted to the experimental data to analytically estimate the variogram for any distance h. The variogram model generally requires two parameters: range and sill. The range is the distance over which pairs of regionalized variable values are spatially correlated, while the sill is the variogram value corresponding to the range.

The optimal fitting is chosen on the basis of a cross-validation. The goodness of fit is evaluated by the mean error (ME) and the mean squared deviation ratio (MSDR). The ME proves the unbiasedness of estimate if its value is close to 0. The MSDR is the ratio between the squared error and the kriging variance (Webster & Oliver 2007), and if the model for the variogram is accurate, its value should be 1. The fitted variograms for both the measured mean monthly precipitation and temperature data were used to estimate their values at ungauged locations using a variant of kriging. Precipitation was predicted using Ordinary Kriging (OK) (Webster & Oliver 2007), whereas, since monthly mean temperature data are available only at few locations, and temperature is generally well correlated with elevation, a variant of kriging known as Kriging with External Drift (KED) was adopted using elevation as secondary information (Hudson & Wackernagel 1994).

In this study, the values of precipitation and temperature were estimated at the nodes of a 250 × 250 m interpolation grid, and all geostatistical analyses were performed using the software Isatis®, release 2013.3 (http://geovariances.com).

Moreover, monthly data were used to outline Péguy climographs (Péguy 1961), where mean values of both mean monthly temperatures (°C) and precipitation (mm) are reported on the x-axis and on the y-axis, respectively. A triangular reference polygon, having coordinates (0 °C; 0 mm), (23 °C; 45 mm), and (16 °C; 200 mm), divides the Cartesian space in four climatic regions, correspondent to the dominant character of the month: (a) temperate, months falling within the triangle; (b) arid, months below the alignment (0 °C; 0 mm), (23 °C; 45 mm); (c) tropical, months on the right side of the alignment (23 °C; 45 mm), (16 °C; 200 mm) and (d) cold, months on the left side of the alignment (0 °C; 0 mm), (16 °C; 200 mm).

Finally, the precipitation and temperature datasets were also analysed for trend, with significance assessed through the Mann–Kendall non-parametric test, widely used since normality or linearity are not required (Mann 1945; Kendall 1962).

The Calabria region occupies the southern part of the Italian peninsula (Figure 1). It has an area of 15,080 km² and a coastline of 738 km on the Ionian and Tyrrhenian seas. To the north, it borders the Basilicata region for 80 km. The Calabria region has an oblong shape with a length of 248 km and a width ranging between 31 and 111 km. Although Calabria does not have many high summits, it is one of the most mountainous regions in Italy (Figure 1): 42% of the land is mountainous (elevation greater than 500 m above sea level (a.s.l.)), 49% hilly (elevation between 50 and 500 m a.s.l.) and only the 9% is flat (elevation lower than 50 m a.s.l.). The maximum elevation is 2,267 m a.s.l., while the average elevation is 597 m a.s.l.

Because of its geographic position and mountainous nature, Calabria has a high climatic variability (Coscarelli & Caloiero 2012) with a typically dry summer subtropical climate, also known as the Mediterranean climate. The coastal zones are characterized by mild winters and hot summers with little precipitation. In particular, the Ionian side, which is influenced by currents coming from Africa, has high temperatures with short and heavy precipitation, while the Tyrrhenian side is influenced by western air currents and presents milder temperatures and considerable orographic precipitation. In the inland zones there are colder winters (with snow) and fresher summers (with some precipitation) than those marking the costal zones (Caloiero et al. 2011a).

Daily precipitation and air temperature data, for the Calabria region, have been collected since 1916 by the former Italian Hydrographic Service, now Centro Funzionale Multirischi Arpacal. The data used in this study are a set of monthly precipitation and air temperature series registered in Calabria for the period 1916–2010. In particular, the precipitation database was extracted from the homogeneous one presented in Brunetti et al. (2012), consisting of 129 rainfall series, updated to 2010. The whole database was used only for the spatial analysis of rainfall. Further analyses were conducted using only the rainfall data given by gauges in which both rainfall and temperature data were present. As regards temperature, the homogenization procedure was applied to both maximum and minimum series having more than 50 years of observation. As a result, 16 out of 34 monthly temperature series were homogeneous, while six series were discarded due to their very low quality. A total of 12 series were homogenized, by correcting 25 and 27 breaks for the maximum and minimum temperature, respectively. Finally, the dataset used for the analysis included 28 stations, with a density of about one gauge per 530 km² (Figure 1).

Basic statistics of regional precipitation and temperature data are presented in Tables 1 and 2, respectively. With reference to the mean values, December (163.5 mm) is the rainiest month of the year, while the driest month is July (17.8 mm). January is the coldest month (8.2 °C), while August is the warmest one (24.5 °C). The skewness of the precipitation data ranges from 0.06 to 0.58 (Table 1) and no transformation was considered to remove asymmetry. On the contrary, the skewness of the temperature data ranges from −0.80 to −1.3 (Table 2) and for subsequent analyses they were transformed into a Gaussian variable, and standardized to zero mean and unit variance using the Gaussian anamorphosis (Wackernagel 2003) by an expansion of Hermite polynomials (Rivoirard 1994) restricted to the first 30 terms. The transformed data were used in the interpolation process and then the kriging results were back-transformed to the raw distribution.

To check the behaviour of elevation in terms of stationarity and anisotropy, monthly variogram maps were calculated for each variable. Variogram maps (not shown) for monthly mean precipitations and temperatures clearly showed the same behaviours of the experimental variograms in different directions so that an isotropic model was fitted to them. The parameters of the fitted model are reported in Tables 3 and 4 respectively. The fitted variogram models for mean monthly precipitations were nested and included mainly a nugget effect, a short-range spherical model (Webster & Oliver 2007) and a long-range spherical model (Table 3). The nugget effect implies a discontinuity in the random function and is a positive intercept of the variogram. It arises from errors of measurement and spatial variation within the shortest sampling interval (Webster & Oliver 2007).

The nested variogram with two models at shorter and longer scales describes two scales of variation, probably due to different processes.

The fitted variogram models for the Gaussian values of mean monthly temperature (Table 4) were nested and included a nugget effect and a spherical model. The goodness of fitting of all variogram models was verified by cross validation. The results were quite satisfactory because the statistics used, i.e. mean of the estimation error and variance of the MSDR, were quite close to 0 and 1, respectively.

The values of the mean monthly precipitation were predicted at the nodes of the interpolation grid using OK and mapped (Figure 2).

The values of the Gaussian mean monthly temperature were predicted at the nodes of the interpolation grid using KED. The predicted values of temperature were back transformed to the raw data through the mathematical model calculated in the Gaussian anamorphosis and mapped (Figure 3).

Monthly climatic characteristics were obtained through the Péguy climographs of the 28 stations in which both temperature and precipitation data were available (Figure 4). As a result, three groups of stations were identified with a different behaviour of the climograph. In the first one (Figure 4(a)), the thermo-pluviometric stations are characterized by arid climate for 3–5 months, while the remaining months have a temperate climate; these stations are all located near the coast with an altitude lower than about 80 m a.s.l. A second group of stations presents similar behaviour but with 1–3 months characterized by cold climate; in this case the stations elevation ranges from 80 to about 400 m a.s.l. In the third one, cold conditions are clearly shown, with 5–6 months which present cold climate and fewer months (1–2) with arid climate. The stations of the latter group are all located at the highest altitudes (>400 m a.s.l.).

As expected, the climate conditions are quite different between high and low altitude stations, mainly during cold months. This aspect plays an important role on evapotranspiration and infiltration processes. In fact, high mean temperature values, coupled with a low amount of rainfall, are the main reasons that could trigger desertification processes in the topographically low areas of the region. However, other morphological features, besides altitude (e.g. continentality and exposition), could affect climate conditions.

The results of the application of the trend analysis (statistical significance: 95%) to the monthly and seasonal precipitation series are shown in Figure 5 and Table 5. With regards to the seasonal precipitation, a general negative trend was detected for the 1916–2010 period, in particular for the autumn and winter period (Figure 5(a) and Table 5). In fact, 50 and 25% of the rain gauges show a negative trend, with a maximum decreasing rate of −57.8 mm/10 years and −32.9 mm/10 years in winter and autumn respectively. On a monthly scale, a negative trend was detected in January (35.7% of the rain gauges) and November (53.6%), with maximum decreases of −24.8 mm/10 years and −23.6 mm/10 years respectively. A negative trend was also found in September (28.6% of the rain gauges), with a maximum rate of −25.8 mm/10 years. Conversely, summer precipitation shows positive trends, in particular in July (21.4% of the data series) and August (17.9%), with a maximum positive trend of +1.7 mm/10 years and +2.3 mm/10 years respectively.

The results of the trend analysis performed on the 28 temperature series are shown in Figure 5 and Tables 6,⁷–8. In particular, the average temperature (Figure 5(b) and Table 6) shows a positive trend from May to August, with a percentage of stations between 25 and 33.3% and a maximum trend (°C/10 years) ranging from +0.23 to +0.32. A negative trend was detected in September and November, respectively for 46 and 35% of the temperature series with a maximum trend of −0.36 °C/10 years and −0.32 °C/10 years. On a seasonal scale results show a positive trend in spring (18% of the stations and +0.18 °C/10 years) and summer (29% and +0.34 °C/10 years), while in autumn a negative trend for 39% of the series was detected with a maximum reduction of −0.30 °C/10 years.