3.1. Data Preparation
The original data were available in NetCDF format. The NetCDF file has a vectorial structure: there are three vectors of attributes (latitude and longitude to set a point in the space, plus a time vector that is a sequence of integers counting the number of days since 1 January 1950) and a very long vector of numbers that represents the specific climate variable of interest. Once the data were structured in a tabular format, the next step was to develop an Extract, Transform and Load (ETL) transformation using the Pentaho Data Integration tool (Available online:
http://community.pentaho.com/projects/data-integration/ (accessed on 8 May 2012)). The main scope of the transformation was to discretize monthly mean temperatures from a set of meteorological stations (grid points) of two specific Italian regions (i.e., Lombardy and Campania) for studying the changing seasonal temperature distributions using variants of correspondence analysis [
22]. The analysis of climate change in Italy is based on data from the European observation gridded dataset (E-OBS). This dataset is a free set of reference climatic data widely used for the assessment of climate modelling in Europe. The E-OBS dataset contains daily observations of precipitation, temperature and pressure at sea level across Europe from 1 January 1950 to 31 August 2016, based on European Climate Assessment & Dataset (ECA&D) information (Available online:
http://www.ecad.eu/ (accessed on 31 December 2020)). Specifically, the E-OBS dataset contains gridded data for 5 climatic features: daily mean temperature, daily minimum temperature, daily maximum temperature, daily precipitation total and daily average sea level pressure (available since version 14.0 of E-OBS). In particular, we recovered values of mean daily temperature in degree Celsius for Italy in 1986 and 2015. For the sake of brevity, our analysis of the data will be confined to the regions of Lombardy and Campania. In
Table 1 and
Table 2, we summarize the monthly average temperature in 1986 and 2015 of Lombardy and Campania, respectively; in bold are the most relevant temperature increases in 2015. In particular,
Table 1 shows that, in Lombardy in 2015, the largest increase in average temperature (an increase of 2
C from the mean temperature recorded in 1986 was statistically significant with a
p-value = 0.111) occurred over January, February, April, July and December.
Figure 1 shows the temperature averages over 42 meteorologic stations during the 12 months of the year 1986 (left side of
Figure 1-plot a) and 2015 (right side of
Figure 1-plot b). Observe that the highest temperature values are recorded during August 1986. Different from 2015, the highest temperature values are registered in July 2015.
Table 2 summarizes the mean monthly temperature in Campania and shows that the largest increases were observed in 2015 during June, July, and December. In 2015, an increase of 1.9
C from the mean temperature recorded in 1986 was statistically significant with a
p-value = 0.104.
Figure 2 shows the average temperature over 23 meteorologic stations during the twelve months of 1986 (left side of
Figure 2—plot a) and of 2015 (right side of
Figure 2—plot b). Observe that, also in Campania, the highest temperature values are recorded during August 1986, while in 2015, the peak of average temperature values are registered in July. Comparing
Figure 1 and
Figure 2, we can observe that the distribution of average temperature values over the respective meteorologic stations has a different shape, showing more variability in Lombardy than in Campania, which is not surprising given the morphological diversity of the two regions.
To further investigate the issues concerned with climate change in these two Italian regions, we considering classes of temperature that have been observed monthly in 1986 and 2015, and perform some variants of ordered correspondence analysis.
Note that the extremes of temperature classes for both regions was determined using the quantiles of the temperature distribution (where the quantiles are cut points dividing the range of a probability distribution into continuous intervals with equal probabilities). A two-way contingency table was then constructed for each Italian region, which cross-tabulates the temperature classes and the months, for two specific years (1986 and 2015). From this contingency table a super-indicator table was created to perform ordered multiple correspondence analysis (OMCA).
By first performing OMCA, we are able to classify the meteorologic stations according to the temperature classes. In doing so, we can highlight the increased number of meteorological stations in 2015 with respect to 1986 for a particular range of temperature values; see
Table 3 and
Table 4. Secondly, we perform doubly ordered correspondence analysis (DOCA) for identifying those months that observed, for a particular temperature class, an increased number of meteorological stations in 2015 when compared with their number in 1986.
The temperature distribution is slightly different for the two regions due to differences in the specification of the data. The two correspondence analysis variants, OMCA and DOCA, were implemented using the
R packages
MCAvariants and
CAvariants [
36], respectively; these packages are available from the Comprehensive R Archive Network (CRAN) at
https://CRAN.R-project.org/package (accessed on 31 December 2020)).
3.2. Lombardy Region
The Lombardy region, owing to its peculiar geography (extended mountain areas, steep valleys and high-flow rivers) and to its socio-economic features (employment, health, agriculture, industry), is subjected to a high vulnerability to the impacts of climate change. Past trends and future projections suggest a marked rise in mean temperature, as highlighted in
Table 3.
The Lombardy region includes 42 meteorological monitoring stations so that in every 12 month period we have a total of 504 observations. For this, the temperature is divided into the 19 classes, detailed in
Table 3. The column variable,
Month, comprises 12 months, from January to December. By using OMCA, we first calculate the indicator super-matrix (with
number of ordered variables) according to the ordered temperature classes and the month of reference. As a result, 504 meteorological stations were objectively classified in 19 clusters.
In
Table 3, we highlight in bold the highest percentages of meteorological stations recorded during 2015. It is clear that there was an increase (higher than 1%) in the number of meteorological stations in 2015 compared with 1986 for only some temperature classes. In particular there was an increase in the number of stations that recorded temperatures that lie within the following
critical classes (3.5, 4.5], (4.5, 5.7], (8.2, 9.5], (9.5, 10.9], (12.2, 13.4], (13.4, 14.6].
In order to determine those months that were mainly characterized by these temperature changes, we perform doubly ordered correspondence analysis (DOCA) using the two-way contingency tables for 1986 and 2015.
The total inertia, or strength of association between the temperature and month is formally tested using the statistic. Its value is 1589.29 with 198 degrees of freedom, and thus, there exists a statistically significant association between the two variables. While we have not shown it here, all polynomials for both variables are statistically significant for modeling the association.
To portray the association between the temperature classes and the months, we construct polynomial biplots. On the left and right side of
Figure 3, plot a and plot b show the polynomial biplots for the year 1986 and 2015, respectively. Plot a of
Figure 3 highlights the strong association between the temperature class (19.5, 20.9] and the months of July and August, while, plot b of
Figure 3 shows that, in 2015, the association between the temperature classes and months has changed. In particular, plot b of
Figure 3 shows that the temperature class (19.5, 20.9] is strongly associated with the months of June and August. Furthermore, the cold months of January and February are associated with the warmer temperature class (2.3, 3.5] (instead of (−1.6, 1] in 1986). With respect to the
critical temperature classes, plot b of
Figure 3 shows a very strong association between the temperature class (3.5, 4.5] and December, and between (12.2, 13.4] and April. In agreement with the findings discussed by [
11], the average temperature increase recorded in Lombardy of about 2
C in April, and of about 2.2
C in December shows statistically significant results.
Understanding those critical months that experienced a climatic change can allow for an assessment of the vulnerability of the socio-economic and natural systems. Doing so paves the way for the implementation of a regional adaptation strategy.
3.3. Campania Region
The region of Campania, which lies in the southern part of the Italian peninsula, has an area of about 13,500 km
, and a coastline along the Tyrrhenian Sea. The landscape is dominated by the Apennine mountain ranges, reaching altitudes of 1000–2000 m, and accounts for 32% of the land area. The region of Campania has a Mediterranean climate, affected by the Azores, Siberian and South African anticyclones and the Aleutian and Icelandic lows, with hot, dry summers and moderately cool rainy winters. For Campania from January to December, temperature values are divided into the 20 classes detailed in
Table 4. The Campania region includes 23 meteorological stations collecting data during the 12 months of the year for a total of 276 observations in a year. We consider again 1986 and 2015 as our reference years. By performing OMCA, with
ordered categorical variables (the temperature and the months of the year) we have 276 observations that were automatically classified. Of course, this classification does not give the same result for both years, so that a comparison of the distribution differences are of particular interest for defining good climate adaptation practices. In
Table 4, the percentages of meteorological stations higher than 1% in 2015 with respect to 1986 are highlighted in bold. Additionally, in Campania, the increase in the number (percentages) of observations in 2015 with respect to 1986 concerns only some classes of temperature; those being (8.7, 9.6], (9.6, 10.4], (11.2, 12.2], (14.5, 16] and (25.1, 28.8], of course, these
critical classes of temperature of Campania are not the same of Lombardy.
To identify what months were strongly associated with these critical temperature classes that recorded an increase in the number of meteorological stations over the period, we perform DOCA.
The total inertia is 1092.1 for the two-way contingency table formed by cross-classifying temperature class and month for Campania in 2015 with 209 degrees of freedom, there exists a strong association between the temperature class values and the months. To visually represent the association between the two variables in 1986 and 2015, we consider the two plots of
Figure 4. Plot a of
Figure 4 shows the visual summary of the association for 1986. It shows that there exists a strong association between the temperature class (21.8, 22.9] and the month of July, while plot b of
Figure 4 shows, for 2015, the association between July and the warmer temperature class (25.1, 28.8]. These conclusions are in agreement with the findings discussed by [
11], the average test of temperature recorded in Campania shows an increase of about 4
C in July and December that is statistically significant. To analyze the correspondence between those months that were mainly characterized by these temperature changes, we perform doubly ordered correspondence analysis (DOCA) using two-way contingency tables for 1986 and 2015. Furthermore, we observe the association between the
critical temperature classes in Campania (which recorded an increased number of observations in 2015) with the months; plot b of
Figure 4 shows a strong association between (8.7, 9.6] and January, (9.6, 10.4] and March, (11.2, 12.2] and April, (14.5, 16] and October and (25.1, 28.8] and July.
Note also that, in 2015 (plot b of
Figure 4), the variability of the warmer temperature values (see class (25.1, 28.8]) is lower when compared with 1986 (plot a of
Figure 4) being the polynomial coordinate along the second axis lower. Conversely, the variability in the colder temperature values (see class (1.7, 5.8]) has increased relative to the temperatures recorded in 1986. These results highlight a shift in the temperature distribution and is in agreement with previous studies on temperature extremes in Italy [
12].