Homogeneity and Trend Analysis of Climatic Variables in Cap-Bon Region of Tunisia

Abstract

As a semi-arid Mediterranean country, Tunisia is affected by the impacts of climate change, particularly the coastal regions like the Cap-Bon. Irregular rainfall, rising temperatures and the recurrence of extreme events are all indicators that affect ecosystems and populations and make them more vulnerable to the influence of climatic variables. Therefore, an analysis of the trends of climate variables can contribute to facilitating the development of effective adaptation strategies. In this matter, this study was conducted to assess the homogeneity and trends of minimum and maximum air temperature (Tmin and Tmax) and precipitation (P) in the Cap-Bon region. Daily data were collected from the meteorological station of Nabeul for the period of 1982–2020. Pettitt and SNHT tests for homogeneity were applied to identify the breakpoints in multi-time scales of Tmax, Tmin and P data series. The Mann–Kendall (MK) test was used to detect the change in the time-series trend. A modified Mann–Kendall (mMK) test was used to remove the autocorrelation effect from the data series. Both the MK and mMK tests were used at the 5% significant level. The magnitude of the climatic trend was estimated using the non-parametric Sen’s slope estimator. Contrary to Tmin and P, the results of the homogeneity tests revealed the existence of significant breakpoints in the annual, seasonal and monthly Tmax time series. For most cases, the breakpoint occurred around the year 2000. For Tmin, significant breakpoints were recorded in March and April, while a significant shift in the P time series was detected in December. The Mann–Kendall results show a significant warming trend in annual Tmax, with magnitudes equal to 0.065 and 0.045 °C/year before and after the breakpoint, respectively. Nevertheless, non-significant tendencies were observed in the annual Tmin and P time series. On the monthly time scale, Tmax exhibited a significant upward trend in June and August, before the observed breakpoints, with Sen’s slope values equal to 0.065 and 0.045 °C/year, respectively. Regarding the Tmin data, a significant positive trend was observed in July at a rate of 0.033 °C/year.

1. Introduction

Atmospheric carbon dioxide rates have increased significantly since the 1950s, a phenomenon that affects global and regional climate characteristics, such as temperature and precipitation [1]. The effects of climate change are evident directly in terms of extreme events, such as droughts, floods, water resource availability and agricultural production, and indirectly via increasing crop water requirements [2]. Over the last century, the average global air temperature has increased by about 0.74 °C [3]. It is worth mentioning that the observed warming, as well as precipitation anomalies, is not globally uniform [4,5]. Yue and Hashino [6] highlighted that temperature variations at regional scales may be much more pronounced and that substantial spatial and temporal variations can be identified between climatically different areas.

A long-term time-series trend analysis of meteorological parameters is a primordial step in climate change studies. Several studies have been conducted to assess trends in temperature and precipitation across the globe [7,8,9,10,11]. Alam et al. [12] assessed the trend of precipitation and the accuracy of predicted temperatures at three stations in Pakistan, for the period of 1960–2020, based on the non-parametric Mann–Kendall test. The obtained results revealed the existence of a positive trend in annual precipitation. However, a negative trend was observed in monthly precipitation, and the trend was statistically significant for two stations. Kumar et al. [13] analyzed 120-year precipitation time series for three Italian meteorological stations (Florence, Pisa and Palermo). Their results showed a decreasing trend in annual precipitation at Pisa and Palermo and a highly significant downward trend in rainy days at Florence and Pisa. For most Mediterranean regions, a significant decrease in precipitation associated with an increase in exceptional phenomena, such as severe droughts, has been detected [14]. Chaouche et al. [7] examined the changes in annual and monthly temperatures, rainfall and potential evapotranspiration time series in a part of southwestern France. Their findings highlighted that the annual mean temperature and potential evapotranspiration significantly increased; however, no significant trend was detected in annual precipitation. In addition, an important seasonal variability in the trends of all the studied climatic variables was recorded.

Tabari et al. [15] analyzed the trends in annual air temperatures and precipitation time series in different regions in Iran during the 1966–2005 period. They observed an upward trend in air temperature at most of the stations, while a mixed tendency of the precipitation data was recorded. According to Aschale et al. [16], the specific humidity, solar radiation and minimum temperature all had a negative impact on the trend of reference evapotranspiration in Sicily, but the maximum temperature, mean temperature and wind speed all had a favorable impact. They also discovered that there was no annual trend for any of the factors studied. Trends are only observed on a monthly or seasonal scale. Except for the minimum temperature and reference evapotranspiration (ET0), all variables showed significant trends at the monthly and seasonal time scales. Monthly ET0 showed a significant trend for November, and the monthly minimum temperature showed a significant trend for August and September.

Numerous statistical tests classified as parametric and non-parametric tests are widely employed to check whether either the data of a given set present an abrupt change in the distribution or have a significant trend [9,17]. It is worth mentioning that parametric tests are better than non-parametric ones and that they require the data to be independent and normally distributed [18]. Meanwhile, non-parametric tests are less sensitive for non-normal distribution, extreme values and missing data, and they require the data to be independent [11].

A long-term time-series trend analysis involves the trend slope and its statistical significance [19]. Non-parametric tests, such as the Mann–Kendall test, Sen’s slope and Spearman’s Rho, are largely used by scientists to detect the trends in historical series of climatic variables [10,20,21,22,23,24]. Nevertheless, non-climatic factors can bias climate patterns and signals, which mislead the conclusions of climatic-related studies. Thus, testing the homogeneity of the long-term time series of climate variables before their use is highly recommended. Different homogeneity tests have been proposed in the literature, including the Pettitt test [25], the Standard Normal Homogeneity Test (SNHT) [26], the Buishand range test [27] and the Von Neumann ratio test. The SNH test detects abrupt shifts at the extremes of the series. However, the Pettitt and Buishand tests identify breakpoints in the middle of the series [28]. The sources of abrupt changes in climate data might be related to the relocations of meteorological stations, the changes in observation time and methods of variable calculation [29]. Another factor likely to lead to climate change is the modification of large-scale atmospheric circulation [30].

Detailed analysis of the hydro-climate trend and variability is essential for better national and local water resources management and, therefore, for better agricultural planning activities, hydrological modeling and climate change studies. Comprehension of water balance components and their pattern in the future is essential for water resources management and sustainable development. Previous research indicates that most climate variables, except temperature, are expected to present varying trends in time and space [16].

Tunisia is facing a moving up of aridity from the south to the north of the country. This has led to a decrease in cultivable areas and a risk of a considerable drop in yields and national agricultural production. In addition, a clear downward trend in water resources availability and a significant increase in water demand have occurred as a consequence of the increase in ET0, resulting in a water balance deficit being observed [31,32]. However, limited information is available on Tunisia, in particular on the Cap-Bon region, which is considered an important agricultural hub contributing 15% of the national agricultural production value. In this study, different methods of trend detection and breakpoint analysis have been employed for long-time series of climatic variables of minimum and maximum air temperature and precipitation of the Nabeul region. The present study could help different actors (scientists, hydrologists, environmentalists, crop producers, irrigation managers, water resources managers, etc.) to adopt strategies for sustainable water resources management and therefore to ensure agricultural and food sustainability.

2. Materials and Methods

2.1. Data Collection

The study area is the Cap-Bon region, located in North East Tunisia (36°27′ N, 10°44′ E) (Figure 1). The climate of the study area is Mediterranean semi-arid with an average annual precipitation of 600 mm and annual reference evapotranspiration estimated using FAO-56 Penman–Monteith exceeding 1100 mm. This semi-arid region of Tunisia covers the administrative region called the Governorate of Nabeul. It occupies 2.825 km², and it is characterized by a great variety of landscapes.

In the current study, a valid and adequate daily data series of maximum (Tmax) and minimum (Tmin) air temperature and precipitation (P) were collected from the National Meteorological Institute (INM) for the period of 1982–2020. The consistency of the databases was checked, and a graphical analysis of the daily time series was performed. Despite the overall quality of the used data being fairly good, there were some missing data. These gaps were filled with the corresponding average value of the other years. Then, the datasets were aggregated into monthly seasonal and annual time scales and investigated for homogeneity and trend detection analysis.

2.2. Homogeneity Tests

Numerous tests can be used for the homogeneity evaluation of hydro-climatological series data. In the present study, both parametric (Standard Normal Homogeneity Test: SNHT) and non-parametric (Pettitt) tests were used to analyze the homogeneity in monthly, seasonal and annual series of minimum (Tmin) and maximum (Tmax) air temperature and precipitation (P).

2.2.1. Pettit Test

The Pettitt test is a non-parametric method based on ascending order (r_i) of rank of the n elements in the series [25]. The test detects the breaks which occur near the middle of the time series [33,34]. The ranks r₁, r₂, …, r_k of the Y₁, Y₂, …, Y_i, …, Y_k data were used to calculate the X_k Pettitt test’s statistic:

$${\mathrm{X}}_{\mathrm{k}}={{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{k}}{\mathrm{r}}_{\mathrm{i}}-\mathrm{k}\left(\mathrm{n}+1\right)(\mathrm{K}=1,2\dots \dots \mathrm{n})$$

(1)

where k is the years of record and n is the length of the dataset.

A break occurs in a given year when the absolute value of X_k reaches its maximum value.

$${\mathrm{X}}_{\mathrm{k}}=\max \left|{\mathrm{X}}_{\mathrm{k}}\right|\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}(1\le \mathrm{k}\le \mathrm{n})$$

(2)

2.2.2. Standard Normal Homogeneity Test (SNHT)

The SNHT is a parametric test that is usually more sensitive to breaks near the boundaries of the time series [35]. Alexandersson et al. [36] defined the T(k) statistic for the SNHT as:

$$\mathrm{T}\left(\mathrm{k}\right)={\mathrm{k}\overline{\mathrm{Z}}}_1^2+\left(\mathrm{n}-\mathrm{k}\right){\overline{\mathrm{Z}}}_2^2\to (\mathrm{k}=1,2,\dots .,\mathrm{n})$$

(3)

$$\overline{{\mathrm{Z}}_1}=\frac{1}{\mathrm{k}}{{\displaystyle \sum }}_{\mathrm{i}=1}^{\mathrm{k}}({\mathrm{Y}}_{\mathrm{i}}-\overline{\mathrm{Y}})/\mathsf{\sigma }\mathrm{and}\overline{{\mathrm{Z}}_2}=\frac{1}{\mathrm{k}-1}{{\displaystyle \sum }}_{\mathrm{i}=\mathrm{k}+1}^{\mathrm{k}}({\mathrm{Y}}_{\mathrm{i}}-\overline{\mathrm{Y}})/\mathsf{\sigma }$$

(4)

where

$\overline{Z_1}$ and $\overline{Z_2}$ are the parameters of T(k) statistic;

$\overline{Y}$ is the average of time series;

σ is the standard deviation.

The statistical test T(k) is used to compare the mean of the first n observations with the mean of the last (n − k) observations with n data points. The year k can be considered as a change point and consists of a break where the value of Tk reaches its maximum value. The T0 statistic in the standard normal homogeneity test is defined as:

$${\mathrm{T}}_0={\mathrm{maxT}}_{\mathrm{k}}\hspace{1em}\hspace{1em}\hspace{1em}(1\le \mathrm{k}\le \mathrm{n})$$

(5)

When the statistical test surpasses the critical value tabulated in [37], which depends on the sample size (n), the null hypothesis is rejected.

2.3. Temporal Trend Analysis

Trend detection in long-term series of monthly, seasonal and annual series of minimum and maximum air temperature and precipitation were carried out using Mann–Kendall, Modified Mann–Kendall and Trend-Free Pre-whitened MK tests and Sen’s estimator.

2.3.1. Mann–Kendall Test (MK)

The Mann–Kendall is a rank-based non-parametric test, widely employed to detect monotonic trends (increasing or decreasing) in time series of hydro-climatic variables [38]. Mann–Kendall test does not require the data to be normally distributed, and has low sensitivity to abrupt changes in time series [39]. For the null hypothesis H0 for MK test, the deseasonalized data (x₁, …, x_n) comprise a sample of N independent and identically distributed random variables. For hypothesis H1, the distributions of x_k and x_j are not identical for all k, j ≤ N with k ≠ j. The Mann–Kendall test is based on the statistic S:

$$\begin{gathered} \mathrm{S}={\displaystyle \sum }_{\mathrm{Z}=1}^{\mathrm{N}-1}{\displaystyle \sum }_{\mathrm{j}=\mathrm{Z}+1}^{\mathrm{N}}\mathrm{sign}(\mathrm{y}) \\ \mathrm{where}\mathrm{sign}(\mathrm{y})=\mathrm{sign}({\mathrm{x}}_{\mathrm{j}}-{\mathrm{x}}_{\mathrm{z}})=\left\{\begin{array}{c}+1 \mathrm{if} y>0\\ 0 \mathrm{if} y=0 \\ -1 \mathrm{if} y<0\end{array}\right. \end{gathered}$$

(6)

where x_j and x_z are the annual values (j > z).

A very high positive value of S indicates an upward trend, and vice versa. For N > 10, the test Z approximately follows a normal distribution. It is estimated as follows:

$$\mathrm{Z}=\left\{\begin{array}{c}\frac{\mathrm{S}-1}{\sqrt{\mathrm{Var}\left(\mathrm{S}\right)}} \mathrm{if} S>0\\ 0 \mathrm{if} S=0\\ \frac{\mathrm{S}+1}{\sqrt{\mathrm{Var}\left(\mathrm{S}\right)}}\mathrm{if} S<0\end{array}\right.$$

(7)

In this test, the null hypothesis H0 is that there is no monotonic trend in the series. If Z > 0, it indicates an increasing trend, while if Z < 0, it indicates a downward trend in the series. When testing either positive or negative monotonic trends at the α significance level, the null hypothesis is rejected for an absolute value of Z greater than Z1-α/2 [5].

2.3.2. Modified Mann–Kendall Test (mMK)

Similarly to the MK test, the mMK follows the same approach to give the monotonic trend result. However, the difference is the addition of a variance correction approach in the mMK. Based on the approach of variance correction, it is considered that existence of the autocorrelation between some N data consists of the same information as with uncorrelated data [40]. Furthermore, based on sets of Monte Carlo simulations, no change in the convergent normality and mean of Mk test statistic (S) have been proven in the presence of serial correlation, while it changes the variance of the distribution of S.

However, time series that are not random and influenced by autocorrelation makes trend test too liberal when none exists [40]. Hamed and Rao [41] have proposed a modified version of the Mann–Kendall test, which is appropriate in the presence of autocorrelation. This version is based on the modified variance of S (Var*(S)) (Hamed and Rao, [41]). First, data are detrended. Then, the ranks of significant serial correlation coefficients are used to calculate the effective sample size of the dataset. These coefficients are then used to adjust the inflated/deflated variance of the statistical test.

$$Va{r}^{*}\left(S\right)=Var \left(S\right)\frac{n}{n_s^{*}}$$

(8)

The correction factor $\frac{n}{n_s^{*}}$ is estimated as follows [40]:

$$\frac{n}{n_s^{*}}=1+\frac{2}{n*\left(n-1\right)*\left(n-2\right)}{\displaystyle \sum }_{i=1}^{n-1}\left(n-i\right)\left(n-i-1\right)\left(n-i-2\right){\rho }_s\left(i\right)$$

(9)

where ${\rho }_s\left(i\right)$ is the autocorrelation between the observation ranks.

$n_s^{*}$ is the effective number of observations.

The autocorrelation is computed after subtraction of the non-parametric trend estimator (e.g., Sen’s slope estimator). Only significant values of the ranks ${\rho }_s\left(i\right)$ are used to calculate the variance correction factor.

2.3.3. Trend-Free Pre-Whitened-MK Test (TFPW-MK)

The TFPW-MK test proposed by Yue et al. [1] is applied to detect a significant trend in a serially correlated time series. According to [42,43], the different steps are carried out as follows:

-

The time series trend slope (Q) is evaluated based on the assumption of Theil (1950) and Sen (1968) detailed below. Before trend analysis, the original sample data X_t are unitized by dividing all values with the sample mean [1]. This procedure preserves the properties of the original data, and the mean of each dataset is set to 1. If Q = 0, it is not recommended to carry out trend analysis. If Q ≠ 0, then it is expected to be linear, and the sample data are de-trended as follows:

$${\stackrel{`}{\mathrm{X}}}_{\mathrm{t}}={\mathrm{X}}_{\mathrm{t}}-{\mathrm{T}}_{\mathrm{t}}={\mathrm{X}}_{\mathrm{t}}-\mathrm{Q}*\mathrm{t}$$

(10)

-

For the detrended series X_t, the lag-1 serial correlation coefficient (r₁) is computed. If r₁ tends to zero, the sample data are considered serially independent, and the MK test is thus applied directly to the original sample data. Otherwise, the data are pre-whitened using the lag-1 serial correlation coefficient using the following equation:

$${\stackrel{`}{\mathrm{Y}}}_{\mathrm{t}}={\stackrel{`}{\mathrm{X}}}_{\mathrm{t}}-{\mathrm{r}}_1{\stackrel{`}{\mathrm{X}}}_{\mathrm{t}-1}$$

(11)

-

The obtained trend (T_t) and the residual ${\stackrel{`}{\mathrm{Y}}}_{\mathrm{t}}$ are expressed as:

$${\mathrm{Y}}_{\mathrm{t}}={\stackrel{`}{\mathrm{Y}}}_{\mathrm{t}}-{\mathrm{T}}_{\mathrm{t}}$$

(12)

-

Finally, the data series (Y_t) is tested using the Mann–Kendall trend test [19].

2.3.4. Sen’s Estimator

Sen’s test [44] is implemented to quantify the magnitude of the linear trend slope in the climatic time series. The slopes Q_i of all data pairs are computed as [11]:

$${\mathrm{Q}}_{\mathrm{i}}=\frac{\left({\mathrm{x}}_{\mathrm{j}}-{\mathrm{x}}_{\mathrm{k}}\right)}{\mathrm{j}-\mathrm{k}}\hspace{1em}\hspace{1em}\hspace{1em}\mathrm{pour}\mathrm{i}=1\dots \mathrm{N}$$

(13)

where x_j and x_k are data values. The Qi median is the Sen’s slope, which is calculated as

$${\mathrm{Q}}_{\mathrm{med}}={\mathrm{Q}}_{(\left(\mathrm{N}+1\right)/2)}\hspace{1em}\hspace{1em}\hspace{1em}\hspace{1em}\mathrm{if}\mathrm{N}\mathrm{is}\mathrm{odd}$$

(14)

$${\mathrm{Q}}_{\mathrm{med}}=\left[{\mathrm{Q}}_{\left(\mathrm{N}/2\right)}+{\mathrm{Q}}_{((\mathrm{N}+2)/2)}/2\right]\hspace{1em}\hspace{1em}\hspace{1em}\mathrm{if}\mathrm{N}\mathrm{is}\mathrm{even}$$

(15)

A positive value of Q refers to an upward trend, and a negative value indicates a downward trend in the time series.

2.4. Data Analysis

All analyses were performed using the software R version 4.2.2 and XLSTAT v2021.2. Both packages ‘modifiedmk’ version 1.6 and package ‘trend’ version 1.1.4 were used to perform homogeneity and trends detection analysis, respectively. XLSTAT software was used for better presentation of the results.

3. Results and Discussion

3.1. Homogeneity Analysis of Air Temperature and Precipitation Data

The results of the Pettitt test and the SNHT applied separately to detect inhomogeneities in annual, seasonal and monthly minimum and maximum air temperature and precipitation time series for the whole 1982–2020 period are summarized in

Table 1. Results of Pettitt and SNHT homogeneity tests.

	Tmax				Tmin				P
	Pettitt		SNHT		Pettitt		SNHT		Pettitt		SNHT
	Break Year	p-Value	Break Year	p-Value	Break Year	p-Value	Break Year	p-Value	Break Year	p-Value	Break Year	p-Value
Annual	2000	<0.0001	2000	<0.0001	1992	0.439	1987	0.1184	2011	0.141	2014	0.0696
Winter	2006	0.3587	2006	0.3796	1992	0.6604	2019	0.71	2009	0.1602	2009	0.1408
Spring	1999	<0.0001	1999	<0.0001	1993	0.1157	1987	0.00665	2002	0.6331	2015	0.3339
Summer	2001	<0.0001	2001	<0.0001	1992	0.1814	2019	0.4622	2011	0.6197	2007	0.5125
Autumn	1998	0.0003536	1998	<0.0001	2001	0.746	2003	0.5658	2011	0.6604	2012	0.4998
January	2006	0.101	2006	0.1968	1993	0.3975	1993	0.4011	2007	0.4179	1984	0.2497
February	1994	1	2019	0.2884	2002	0.7169	2002	0.8094	2002	0.7313	1984	0.8367
March	1999	0.0019	1999	0.0011	1987	0.1503	1987	0.0357	2015	0.6467	2015	0.6572
April	1999	<0.0001	2005	<0.0001	1992	0.0453	1992	0.0043	2013	0.3975	2019	0.5739
May	1998	<0.0001	2004	<0.0001	1992	0.2894	1992	0.2387	2011	0.8523	2011	0.8789
June	2001	<0.0001	2001	<0.0001	1992	0.2444	1992	0.2972	2007	0.4608	2007	0.7036
July	2002	<0.0001	2002	<0.0001	1993	0.1929	2014	0.2985	1985	0.6883	2006	0.8874
August	2002	<0.0001	2002	<0.0001	2003	0.746	2016	0.7038	1994	0.746	1988	0.6454
September	2006	0.0003	2006	<0.0001	2001	0.101	2001	0.215	1994	0.2174	1994	0.7795
October	2000	0.0062	1998	0.0067	2006	0.9644	1985	0.8129	1984	1	1982	0.6582
November	1999	0.0281	2007	0.0194	2003	0.6883	2003	0.9026	2005	0.1321	2005	0.3999
December	1992	0.5305	1984	0.4622	1991	0.6883	1991	0.6854	2007	0.0184	2007	0.0477

. The inhomogeneity was assessed for a significance level of 95% and the datum of the breakpoints was determined.

The Pettit test showed that maximum air temperature time series were inhomogeneous at annual, seasonal and monthly time scales. However, for January, February and December, which refer to the winter season, the change points were not statistically significant. The significant breaks were upward shifts and generally observed between 1998 and 2002 and in 2006.

Relatively similar results were registered by the SNHT with inhomogeneities structure recorded between 1998 and 2007. Indeed, during the 1994–2011 period, the Nabeul region, as one of the competitive triangle areas surrounding the capital (Tunis), has seen the emergence of a metropolization, mainly of an industrial origin [45]. This industrial development has resulted in an increase in urban areas in the region, which would contribute to global warming.

For the annual and seasonal minimum air temperature datasets, no significant shifts were detected by either the Pettitt or SNHT methods. In the time series of Spring Tmin, a significant breakpoint was detected based on the SNHT. The breakpoint was an upward change and occurred in 1987. Regarding monthly Tmin, a significant break point was found in April using both tests. In addition, starting from 1987, a significant upward change point in March was detected based on the SNHT. The homogeneity results are influenced by several factors, among them, the adopted homogeneity test [35]. As described before, contrarily to the Pettitt method, the SNHT is appropriate for detecting abrupt changes at the beginning and at the end of the records [33].

According to the Pettitt test and the SNHT, the annual and seasonal time series of precipitation are homogenous (Figure 2).

However, for the monthly precipitation series, the Pettitt test and the SNHT revealed the existence of a significant downward breakpoint in December (Figure 3).

Before and after the break point, the monthly mean precipitation values were 68.74 and 21.7 mm, respectively.

3.2. Trend in Climatic Variables

3.2.1. Trend in Annual, Seasonal and Monthly Air Temperature

Sen’s slopes of the annual, seasonal and monthly tendencies in Tmax time series obtained from the MK, mMK and TFPWMK tests are shown in

Table 2. Results of the statistical tests for Tmax.

	Before Break Point If Any					After Break Point If Any
	MK p-Value	mMK p-Value	Sen	TFPWMK p-Value	Sen	MK p-Value	mMK p-Value	Sen	TFPWMK p-Value	Sen
Annual	0.050	0.050	0.036	0.041	0.043	0.025	0.011	0.040	0.108	0.041
Winter	0.105	0.105	0.021	0.066	0.021
Spring	0.058	0.058	0.060	0.108	0.064	0.057	0.057	0.042	0.112	0.040
Summer	0.112	0.112	0.030	0.025	0.041	0.327	0.327	0.049	0.762	0.022
Autumn	0.266	0.266	−0.022	0.300	−0.022	0.338	0.338	0.019	0.319	0.024
January	0.095	0.095	0.033	0.056	0.026
February	0.364	0.438	0.022	0.365	0.022
March	0.240	0.240	0.058	0.343	0.054	0.976	0.976	−0.005	0.770	−0.010
April	0.649	0.649	0.018	0.711	0.022	0.174	0.174	0.071	0.144	0.092
May	0.058	0.058	0.063	0.079	0.069	0.071	0.071	0.096	0.124	0.097
June	0.020	0.020	0.065	0.003	0.084	0.940	0.940	−0.002	0.711	0.026
July	0.739	0.583	0.008	0.347	0.046	0.544	0.544	0.036	0.711	0.035
August	0.020	0.020	0.045	0.041	0.049	0.940	0.940	−0.010	0.650	0.050
September	0.224	0.224	0.022	0.097	0.031	1.000	1.000	−0.003	0.583	−0.041
October	0.944	0.944	−0.004	1.000	0.000	0.496	0.496	−0.057	0.675	−0.025
November	0.484	0.484	0.019	0.940	0.006	0.871	0.871	0.004	0.834	0.011
December	0.231	0.087	0.023	0.407	0.018

. Sen’s slopes of significant trends (95% levels) are presented in bold characters in the table.

Despite dividing the time series of annual Tmax at the significant change point (2000), a significant increasing trend was recorded for both divided parts. Thus, the warming trend in the annual Tmax must be accepted as evidence, which is probably due to the increase in urbanization and industrialization in the study region. The magnitude of the significant increasing trends estimated using Sen’s slope method were 0.036 and 0.04 °C/year before and after the breakpoint, respectively. Our results are in the same order of magnitude as those reported by Espadafor et al. [38]. The authors reported an increasing trend in Tmax, in different locations in Southern Spain, with Sen’s slope values varying from 0.16 to 0.41 °C/decade.

In the spring, summer and autumn seasons, where significant upward trends were observed in the whole data series, no significant trend was detected if the divided parts were analyzed separately. Furthermore, similar behavior was observed with the monthly time series (except for June and August). It is therefore evident that for seasons and months where a significant upward trend was observed for the entire dataset, it might have resulted from the abrupt point in the mean. In the first period (before the break point), the MK test detected a significant increasing trend in Tmax for June and August. Their corresponding rates of change quantified using the method of Sen were 0.065 and 0.045 °C/year, respectively. It is worth mentioning that the second part of the monthly Tmax time series (after the dates of significant break) shows a mix of non-statistically significant positive and negative trends. Despite the existence of significant positive trends in June and August, no significant upward trend in the summer season was recognized.

The Mann–Kendall and modified Mann–Kendall p-values and the trend slopes estimated by Sen’s approach for annual, seasonal and monthly minimum air temperature are shown in

Table 3. Results of the statistical tests for Tmin.

	Before Break Point If Any					After Break Point If Any
	MK p-Value	mMK p-Value	Sen	TFPWMK p-Value	Sen	MK p-Value	mMK p-Value	Sen	TFPWMK p-Value	Sen
Annual	0.453	0.546	0.007	0.280	0.010
Winter	0.790	0.790	0.007	0.513	0.008
Spring	0.183	0.142	0.014	0.209	0.013
Summer	0.116	0.021	0.018	0.152	0.020
Autumn	0.790	0.790	−0.002	0.880	0.002
January	0.333	0.333	0.018	0.152	0.028
February	0.904	0.887	−0.002	0.563	0.007
March	0.397	0.473	0.013	0.841	0.008
April	0.755	0.755	0.042	1.000	−0.029	0.984	0.984	−0.0005	0.867	0.0028
May	0.628	0.628	0.006	0.744	0.006
June	0.310	0.423	0.019	0.131	0.026
July	0.024	0.024	0.033	0.006	0.037
August	0.753	0.621	0.007	0.940	0.003
September	0.358	0.469	−0.013	0.513	−0.009
October	0.628	0.628	−0.006	0.706	−0.009
November	0.904	0.904	0.004	0.782	0.005
December	0.781	0.781	0.005	0.782	0.005

. Sen’s slopes of significant trends (95% levels) are shown in bold characters.

For the annual Tmin dataset, a non-significant increasing trend was recorded. Furthermore, no significant upward or downward tendencies were registered in the monthly time series. These results matched those obtained by Tabari et al. [5] who reported that there were no significant upward or downward trends in the monthly Tmin series at Zanjan station in Iran. However, several studies have reported the existence of significant increasing trends in Tmin and that the increment in Tmin is more pronounced than in Tmax [20,22,46,47]. In our case study, the non-significant decreasing trends were captured in February, September and October with Sen’s slope values equal to −0.002, −0.013 and −0.006, respectively. Particularly for July, the Tmin dataset exhibited a significant upward trend. The corresponding Sen’s slope value was equal to 0.033 °C/year. In the case of April, when a significant break-point was identified, both parts did not exhibit any significant trend. Therefore, it can be concluded that the observed significant increasing trend for the whole dataset of April may be related to the positive shifts in the mean.

3.2.2. Trend in Annual, Seasonal and Monthly Precipitation

The results of the statistical tests applied for annual and monthly precipitation for the period (1982–2020) are summarized in

Table 4. Results of the statistical tests for the precipitation.

	Before Break Point If Any					After Break Point If Any
	MK p-Value	mMK p-Value	Sen	TFPWMK p-Value	Sen	MK p-Value	mMK p-Value	Sen	TFPWMK p-Value	Sen
Annual	0.161	0.161	−3.754	0.327	−2.505
Winter	0.093	0.093	−1.674	0.191	−1.516
Spring	0.236	0.236	0.440	0.327	0.733
Summer	0.681	0.787	0.044	0.960	0.025
Autumn	0.371	0.413	−1.238	0.633	−0.571
January	0.157	0.102	−0.636	0.280	−0.418
February	0.468	0.258	0.217	0.393	0.315
March	0.865	0.865	−0.073	0.940	0.070
April	0.961	0.961	−0.021	1.000	−0.006
May	0.553	0.616	−0.116	0.687	−0.066
June	0.809	0.809	−0.006	0.513	−0.026
July	0.815	0.815	0.000	0.279	−0.034
August	0.707	0.772	0.004	0.597	0.029
September	0.327	0.255	0.427	0.269	0.546
October	0.654	0.654	−0.173	0.980	0.002
November	0.061	0.061	−0.850	0.078	−0.614
December	0.290	0.290	1.35	0.154	1.39	1.000	1.000	0.000	0.940	−0.034

. A non-significant decreasing trend in annual precipitation was detected, with an amount equal to −37.5 mm/decade. This result is in agreement with those obtained by Scorzini and Leopardi [48], who found a non-significant decreasing trend in annual precipitation for most of the studied stations in central Italy during the period 1951–2012. The estimated average of magnitude of the decreasing trend was equal to −15.2 mm/decade, which is lower than that obtained in our case.

In addition, Benabdelkrim et al. [49], in their assessment of annual and seasonal precipitation trends over the Macta basin (northwest of Algeria), showed that 17 stations among 48 presented insignificant downward trends during the study period with Sen’s slope values ranging from −0.08 to −2.38 mm/year. As it can be inferred from the results, a mix of positive and negative tendencies, which were overall insignificant, characterizes monthly precipitation. Similar findings were reported by Soltani and Soltani [50], who indicated that there is no significant trend in annual and monthly precipitation in Northeast Iran. Regarding the case of December, where a significant downward shift was recorded around the year 2007, the divided parts did not exhibit any significant trend. Thus, the declining trend observed for the whole data is mainly attributed to the shift break in the mean. It is worthwhile to emphasize that hundreds of years are highly required for better precipitation trend detection. However, our collected precipitation data cover a short time interval, which could impede the exact trend detection.

4. Conclusions

The assessment of temporal variability in different climatic variables due to possible climate change or human intervention is important for water resources planning and management at the regional and local scales. In this study, change point detection and trend analysis in the monthly, seasonal and annual maximum and minimum air temperature and precipitation time series in the Nabeul region have been performed. The Pettitt test and the SNHT were used for change point detection. Mann–Kendall and autocorrelated Mann–Kendall tests were adopted for trend analysis. The change point results indicate that the maximum air temperature showed significant breaks in the annual and most of the seasonal and monthly time series. Meanwhile, no significant change point has been recorded in the winter season and in its corresponding months. These break dates have occurred simultaneously with the rapid development period in the region. The development of agricultural, industrial and tourist activities have increased greenhouse gas and air pollution emissions. The application of Pettitt test to minimum air temperature data indicated the existence of significant breaks in both March and April. Regarding the precipitation time series, a significant abrupt change was recorded in December.

Mann–Kendall results show a significant warming trend in annual Tmax with magnitudes equal to 0.065 and 0.045 °C/year, respectively, before and after the breakpoint. Nevertheless, non-significant tendencies were observed in the annual Tmin and P time series. On the monthly scale, Tmax exhibited a significant upward trend in June and August, before the observed breakpoints, with Sen’s slope values equal to 0.065 and 0.045 °C/year, respectively.

Mann–Kendall trend analysis results evidenced a warming trend in the annual Tmax, which is probably due to the fast development and human interventions in the study region. Despite the existence of a significant increasing trend in the whole datasets of most of the seasonal and monthly Tmax series, no significant trend could be recognized when dividing the time series at the significant breakpoint. Except for June and August, significant upward trends were detected, before the observed breakpoints, in the Tmax time series with Sen’s slope values equal to 0.065 and 0.045 °C/year, respectively. In the case of the minimum air temperature, no significant increasing trend was detected in annual and monthly time series, except in July, where a significant positive trend was observed with Sen’s slope equal to 0.033 °C/year. Regarding the precipitation data, an insignificant declining trend was captured for the annual time scale. However, a mix of positive and negative tendencies, which were overall insignificant, characterized the monthly series of precipitation. These results highlight the importance of the application of homogeneity tests before trend analysis since they allow false trend detections to be avoided.