Precipitation Regime Changes at Four Croatian Meteorological Stations

: The article analyses the values of daily, monthly and annual precipitation measured during the period 1948–2019 at the following four stations: (1) Split, (2) Hvar, (3) Lastovo and (4) Zagreb. The ﬁrst three stations are located in a Mediterranean climate, while the station in Zagreb is located in a continental climate. The aim of the performed analyses is to detect non-stationarity (trends, jumps, and seasonality) in the precipitation regime at three-time scales (day, month, and year) over the period of the last 72 years (1948–2020). Numerous previous analyses at all four stations showed statistically signiﬁcant increases in air temperature, which were particularly ampliﬁed in the late 1980s by the effect of global warming. Expressed as a percentage of the total annual precipitation at all four analysed locations, the presence of an increasing trend was calculated. The analyses carried out in this work showed that there was a redistribution of precipitation during the year, a decrease in the number of days with precipitation and an intensiﬁcation of precipitation in both climatic regions. Over the past 73 years, the number of days with precipitation per year has slowly decreased. The number of days with intense precipitation, P ≥ 32.0 mm, has become more frequent.


Introduction
In recent decades, scientific interest in the study of various aspects of the precipitation regime as a consequence of global climate change has greatly increased. It is of particular importance to identify in each region and/or location possible changes in the characteristics of extreme precipitation, changes in the redistribution of precipitation during the year, and to determine the presence of precipitation trends at different time scales from day to month and season to year [1][2][3][4][5][6][7][8][9][10][11][12][13]. Precipitation, especially extreme precipitation, affects the hydrological regime, agricultural production and the occurrence of floods, as well as almost all environmental and social aspects on the planet. On the other hand, their absence leads to droughts, with disastrous consequences. Due to the reduced precipitation and increased temperatures in Mediterranean Southern Italy, karst spring discharges have decreased by 15 to 30% since 1987 [14]. Understanding and estimating the future evolution of various precipitation characteristics is not only of scientific but also of practical interest.
A review of numerous results presented in the literature shows that various characteristics of precipitation behave very differently in different regions and even in nearby locations. The differences in the behaviour of precipitation are much more pronounced than in the behaviour of temperatures, for which there is generally an increasing trend everywhere.
From the numerous literature, only a few examples that show the different behaviour of precipitation in different regions and places are given below.
An analysis of the annual precipitation of 57 stations in Portugal during the period of  showed that none of the trends, increasing or decreasing, were statistically significant since they were at the 5% level [3].
Mathbout et al. [4] found that extreme and heavy precipitation events showed a globally statistically significant decrease in the Eastern Mediterranean and, in the southern parts, a significant decrease in total precipitation.
Analyses of daily precipitation data in 70 weather stations in central and Western Europe during 1961-2012 showed significant increasing precipitation trends over the 20th century, dominantly in winter for both average precipitation intensity and moderately strong events [5].
Of special interest are the conclusions published in [6]: "The IPCC climate models predict, for the Maghreb countries, lower rainfall and increased aridity. Current observations in the three countries of central Maghreb (Morocco, Algeria, and Tunisia) are not consistent with these predictions. The climate change observed during the last years is characterised by a rainfall return but with a far greater intensity".
The torrential rainfall in Italy exceeding 128 mm/day has increased percentage-wise by a factor of 4 during 1951-1995, with strong peaks in El-Nino years. In Spain, extreme categories at both tails of the distribution (light: 0-4 mm/day and heavy/torrential: 64 mm/day and up) increased significantly. No significant trends were found in Israel and Cyprus [7].
Analyses of 71 stations across Turkey for assessment of the long-term changes in weather extremes from 1961 to 2016 revealed decreasing trends in the number of precipitation days and the volume of precipitation. A small percentage of stations experienced significant increasing trends for the average of total precipitation and very wet days, especially over the southeast coast of the Black Sea [8].
Analyses of trends in precipitation concentration and extremes in the Mediterranean Penedès-Anoia Region (Spain) indicate an increase in precipitation in winter and summer and a positive trend of concentration in autumn, with a higher number of extreme events separated by longer dry periods [9].
Trend analysis of the long-lasting precipitation time series  in north-central Ethiopia showed a statistically significant declining trend for annual rainfall with a rate of 15.03 mm per decade [10].
Investigation of trends of seasonal precipitation in Hunan Province (China) over the 1960-2013 period showed different behaviour in five subregions for winter and three subregions for spring, summer and autumn. Winter and summer precipitations experienced an increasing trend in all of the divisions. Winter precipitation in southeast-central and southern Hunan and summer precipitation in southwest-central Hunan especially exhibited a statistically significant tendency, while spring and autumn precipitations show a nonsignificant decreasing trend in all of the divisions [11].
Using different methods and concepts, numerous papers have estimated changes in precipitation regimes by the end of the 21st century.
Analysing the future changes in the occurrence of extreme precipitation events in the eastern Mediterranean, Oikonomou et al. [12] derived the following conclusions: "A general future tendency was found towards drier Eastern Mediterranean, with reduced rainfall intensity. Longer dry spells are expected in all seasons, except autumn, with the largest increase in the southern part of the area. Extreme wet spells will shorten everywhere during all seasons, except autumn. Precipitation intensity was found reduced for all seasons and mostly for summer in South Aegean Sea".
Hertig et al. [13] calculated changes in total precipitation, extreme precipitation, and dry periods in the Mediterranean area until the end of the twenty-first century. Their conclusion are different from the previously mentioned paper: "The results mostly point to reductions of total and extreme precipitation over the western and central-northern Mediterranean areas in summer and autumn and to increases in winter. In contrast, over the eastern Mediterranean area widespread precipitation increases are assessed in summer and autumn, whereas reductions dominate in winter. In spring, total and extreme precipitation decreases prevail over the whole Mediterranean area. Total and extreme precipitation decreases mostly come along with increases of the maximum dry period length. Vice versa precipitation increases are commonly accompanied by a shortening of the maximum dry period length." Changes in the precipitation regime at different time scales are influenced by many global [15,16], regional and local factors. From the extensive literature, one can infer very different behaviours of precipitation in different locations and climates during the 20th and early 21st centuries [1][2][3][4][5][6][7][8][9][10][11][12][13]. The impact of urbanisation on the nature of precipitation in Toronto has been analysed by Gough [36]. Yang et al. [42], using Beijing as an example, conclude that urbanisation reduces the frequency of light rainfall. Thompson and Green [37] analysed the connection between Mediterranean precipitation and its relation to sea level pressure patterns.
The high variability of precipitations, especially in a Mediterranean climate, from year to year and within each year, makes it difficult to assess changes that could be associated with the future development of climate change [9]. The most reliable estimates of the evolution of the precipitation regime in the future are obtained by analysing their characteristics at stations that have a long and reliable series of measurements. This is exactly what has been done in this work. A data series for a period of 73 years (1948-2020), measured at four of the main meteorological stations in Croatia, were analysed. Three stations (Split, Hvar, and Lastovo) are located in the Mediterranean climate, and one (Zagreb) in the continental area ( Figure 1). One of the aims of this work is to investigate the differences in the behaviour of trends and intra-annual distributions at relatively nearby stations located both in identical climatic regions and in two different climatic regions-Mediterranean and continental. These analyses should help to estimate the future behaviour of precipitation more reliably in order to find practical solutions to mitigate possible negative consequences.

Stations and Data
The article uses data on annual, monthly and daily precipitation observed at the main meteorological stations of the State Hydrometeorological Institute (DHMZ) during a period of 73 years, from 1948 to 2020: (1) Split-Marjan (hereafter Split), (2) Hvar, (3) Lastovo, and (4) Zagreb-Grič (hereafter Zagreb). Figure 1 shows a map with the positions of these four stations, whose basic characteristics are listed in Table 1. At the locations of the meteorological stations Split, Hvar and Lastovo, the climate is Mediterranean with hot summers [43]. According to the Köppen-Geiger [44] climate classification, it belongs to the Csa class. At the location of the main meteorological station Zagreb, the climate is moderately warm, humid and with warm summers [43]. According to the Köppen-Geiger [44] climate classification, it belongs to the Cfb class. Stations on the islands of Hvar and Lastovo are not affected by urbanisation. Stations in the cities of Split and Zagreb are located on the tops of hills and are therefore not exposed to heavy urbanisation, which is especially true for the station in Split. Thus, although they are located in rapidly developing cities, the effects of the urban heat island effect are significantly mitigated in their case.
The precipitation regime at the four stations analysed in this paper has been studied in numerous papers [17][18][19][20]45]. The general conclusion is that no statistically significant trend of increase or decrease in precipitation was observed at the stations analysed in these works. It is especially important to emphasise that regarding the precipitation of different durations (day, month, year) during the period 1862-2017 measured at the Zagreb Observatory, neither climate change nor the process of urbanisation influenced the emergence of trends [20].
As for the variations in air temperature, the situation is distinctly different. It should be emphasised that statistically significant trends in the increase of mean annual, as well as minimum and maximum annual temperatures during the twentieth century, were observed at all stations [21][22][23][24][25]. Particularly strong increases in mean annual temperatures have been observed in Split (0.043 • C per year) and Hvar (0.024 • C per year) since 1998, in Lastovo (0.039 • C per year) since 1992, and in Zagreb (0.053 • C per year) since 1988.

Stations and Data
Detailed analyses of daily, monthly and annual precipitation time series of the four stations in the 1948-2020 period were made using different statistical methods.
The probability of the occurrence of extreme minimum and maximum values were analysed using the GEV method (generalised extreme values) [26,27]. The distribution function of the random variable x, the three-parameter distribution of extreme values, is given by the expression: where ξ is a location parameter, α a scale parameter and k a shape parameter. The location parameter ξ describes the displacement of the distribution in a particular direction on the horizontal axis. The scale parameter α defines where the bulk of the distribution lies and how strongly it is distributed, while the shape parameter k is a parameter derived from skewness and represents where the bulk of the data lies. It strictly affects the shape of the distribution by creating a tail distribution [46]. Given the value of the shape parameter k, the GEV includes three standard distributions of extreme values: Frechet, Weibull, and Gumbel. When k = 0, the distribution is of type I or Gumbel; when k > 0, of type II or Frechet; and in the case of k < 0, of a type III or Weibull distribution. In this paper, the statistical analysis of data and the prediction of extreme values for a specific return period was performed by determining the maximum likelihood estimates for the GEV parameters (Figures 2 and 3).  Linear and nonlinear regression and correlation methods were used for the time series trend analysis.
The statistical significance of linear series trends was determined using the Mann-Kendall (M-K) test [28][29][30]. This test is now considered the most reliable procedure for assessing the statistical significance of monotonic linear and nonlinear trends (upwards and downwards) in an analysed time series. One of its major advantages is that it does not require fitting the analysed data to a distribution curve, i.e., the M-K test is nonparametric. As such, it has the widest application in the analysis of a time series of climatological parameters, especially of temperature and precipitation [31][32][33][34]. In this work, the package pyMannKendall for Python was used [35]. The null hypothesis is that there is no monotonic trend in the analysed time series. An alternative hypothesis is that the trend exists. In this work, the probability value p < 0.05 was used as the criterion for accepting the alternative hypothesis (the existence of a statistically significant linear trend).

The Year as a Time Unit of Analyses
The sequences of annual precipitation measured at the four stations analyzed during the period of 1948-2020 are shown graphically in Figure 4a.   Table 2 shows some characteristic values of the precipitation time series observed at the four stations analysed. On average, the most precipitation falls annually in Zagreb, which is in a continental climate, and then in Split, Hvar and Lastovo. At the three stations located in the range of the Mediterranean climate, the distance from the mainland plays an important role in the amount of annual precipitation. The situation regarding the maximum daily precipitation in a year is different. In the Mediterranean climate, this precipitation is much higher than in Zagreb. The standard deviations, Std, and the coefficients of variation, CV, of the precipitation series at the stations in the Mediterranean climate are significantly higher than at the Zagreb station, indicating their much greater variability over time, although the total annual precipitation is lower.
The series of maximum daily precipitations observed in each year during the period of 1948-2020 at the four analysed stations are shown in Figure 4b Table 3 shows the values of total annual precipitation and maximum daily precipitation in a year, P (mm), calculated for return periods of 50, 100, 200 and 500, T (years) using the GEV method. In Table 4, each maximum value of annual or daily precipitation over the period of 73 years that were analysed is associated with the value of the return period calculated with the GEV method. Table 3. Matrix of precipitations, P (mm), and their return periods, T (year), calculated for the four analysed stations time series of whole annual and maximum annual daily precipitations (1948-2020) using GEV distribution. This paper analyses the trends in the series of the number of days of precipitation per year, N, in the following five categories: (1) P ≥ 0.0 mm, (2) P ≥ 4.0 mm, (3) P ≥ 10.0 mm, (4) P ≥ 16.0 mm, and (5) P ≥ 32.0 mm. Figure 5a shows the sequences of the number of precipitation days per year, N, with P ≥ 0.0 mm for all four stations. Importantly, a statistically significant decreasing trend is evident at all stations. It is obvious that the number of days with the occurrence of some kind of precipitation from the trace ("precipitation trace", the unmeasurable precipitation amount in a period of observation by standard instruments; it is different than a period without precipitation) to the maximum is significantly reduced. The trend lines are plotted in the figure Table 5 summarizes all M-K test results for series of days per year, N, for the five categories of precipitation greater than a given value. The general conclusion is that the number of days, N, with precipitation, P ≥ 4.0 mm, and lower decreases at all four stations. In contrast, the number of days per year, N, with precipitation, P ≥ 16.0 mm, and higher increases at all stations except Lastovo. From the previous analysis, it can be concluded that in the last 73 years there has been a slow process of decrease in the number of days with precipitation and an intensification of precipitation in both climatic regions. The following series of sums of annual precipitation, ΣP, is greater than (1) P ≥ 4.0 mm; (2) P ≥ 10.0 mm; (3) P ≥ 16.0 mm; and (4) P ≥ 32.0 mm. Figure 6a shows four series of sums of annual precipitation, ΣP, greater than, P ≥ 4.0 mm. The figure shows linear trends and the linear regression equations with the corresponding squares of the linear correlation coefficients, R2. Linear trends at Split, Hvar and Zagreb stations are upward, while the trend at Lastovo station is downward. The trends are not statistically significant in any of the cases. Figure 6b shows the series of sums of annual precipitation ΣP, greater than, P ≥ 10.0 mm. The linear trends at Split, Hvar and Zagreb stations are upward, while the trend at Lastovo station is downward. The trends are not statistically significant in any of the cases. Figure 6c shows the series of sums of annual precipitation, ΣP, greater than, P ≥ 16.0 mm. Linear trends at Split, Hvar and Zagreb stations are upward, while the trend at Lastovo station is downward. The upward trend of the series at Hvar station is statistically significant, while for the other three stations, the trends are not statistically significant. Figure 6d shows the series of sums of annual precipitation, ΣP, greater than P ≥ 32.0. The linear trends at all four stations are upward, with none of the trends being statistically significant. This analysis also shows a redistribution of the precipitation regime and an increase in the proportion of higher intensity precipitation in the annual total.

P (mm) T (Year) SPLIT HVAR LASTOVO ZAGREB
The above conclusion was further confirmed by analysing the series of percentages of annual precipitation, ΣP%,i, and, greater than (1) P ≥ 4.0 mm, (2) P ≥ 10.0 mm, (3) P ≥ 16.0 mm, and (4) P ≥ 32.0 mm. The arithmetical expression, ΣP%,i, is: where ΣP %,i denotes the sum of precipitation in the year (mm) and is greater than some of the selected values, while P year,i denotes the total precipitation that falls in the year (mm). Figure 7a shows the percentages of precipitation, ΣP%, greater than, P ≥ 4.0 mm. The figure shows linear trends and the linear regression equations with the corresponding squares of the linear correlation coefficients, R2. Trends at all stations are upward, while the trends at Split and Zagreb stations are statistically significant. Figure 7b shows the percentages of precipitation, ΣP%, greater than P ≥ 10.0 mm. Trends at all stations are upward, while the trends at Hvar and Zagreb stations are statistically significant. Figure 7c shows the percentages of precipitation, ΣP%, greater than P ≥ 16.0 mm. Trends at all stations are upward, while the trends at Hvar station are statistically significant. Figure 7d shows the percentages of precipitation, ΣP%, greater than P ≥ 32.0 mm. The linear trends at all four stations are upward, with none of the trends being statistically significant.  The overall summary of the results of the M-K tests, p, performed on the series shown in Figures 6 and 7 are presented in Table 6. From this overview, it can be seen that there is a trend of increasing precipitation at higher intensities at all four analysed stations.

The Month as a Time Unit of Analyses
The graphical representation of the average monthly precipitation at the four analysed stations during the period of 1948-2020 shows a significant difference in precipitation regime during the year between the Mediterranean and continental climates (Figure 8). In the Mediterranean climate, minimal precipitation occurs during the warm season, and the minimum value is reached during the warmest month of the year, in July. In the continental climate in Zagreb, the minima occur during the winter period, and the lowest value is reached in February. From the point of view of agricultural production, but also of the population's water supply, the annual rainfall in the Mediterranean climate is very unfavourable, especially because in summer in this area the water demand is much higher than in the rest of the year, both for agriculture and much more for tourism. In an attempt to determine changes in the precipitation regime over the months of the year, precipitation trends were calculated for each month of the year over the period of 1948-2020. Table 7 lists the results of the M-K test, p. Statistically, significant growth trends occurred only in October at the Split and Zagreb stations. It can be seen that the trends of increase and/or decrease in monthly precipitation vary from station to station. It is difficult to discern any regularity from this overview.

Conclusions
The analyses carried out in this work showed slight tendencies in the intensification of the precipitation regime at the four analysed stations, three of which are located in a Mediterranean climate (Split, Hvar and Lastovo) and one (Zagreb) in a continental climate. This intensification is expressed in a decrease in the number of days per year with precipitation greater than the selected categories and in an increase in precipitation greater than 16.0 mm.
Change in spatiotemporal precipitation properties due to the ongoing climate change can be identified on different scales and different resolutions. Even with four different measuring points, analysis on registered precipitation patterns can help to understand the significance of ramifications caused by global climate change on to hydrological regime.
In addition to global warming and belonging to different climatic zones, the four analysed stations were significantly influenced by the fact that Lastovo, for example, is located on a small island in the open Adriatic Sea. Hvar is also located on a small island, although much closer to the mainland, while Split is located on the mainland and in a relatively large city. Zagreb station is located on a hill in the centre of the big city. The influences of the sea and urbanisation have certainly contributed to different changes in precipitation at different time scales over the 73 years analysed. However, the similarities in the trends cannot be overlooked.
The Mediterranean area is recognised as a hot spot for climate change [6]. Changes in the precipitation regime in the Mediterranean are a very exciting and important topic in numerous papers [4,6,9,[12][13][14][37][38][39][40][41]. The authors hope that this work will contribute to the solution of the extremely complex and very important topic of changes in the precipitation regime, especially in the Mediterranean region.