Complexity of Forces Driving Trend of Reference Evapotranspiration and Signals of Climate Change

Abstract

Understanding the trends of reference evapotranspiration (ET_o) and its influential meteorological variables due to climate change is required for studying the hydrological cycle, vegetation restoration, and regional agricultural production. Although several studies have evaluated these trends, they suffer from a number of drawbacks: (1) they used data series of less than 50 years; (2) they evaluated the individual impact of a few climatic variables on ET_o, and thus could not represent the interactive effects of all forces driving trends of ET_o; (3) they mostly studied trends of ET_o and meteorological variables in similar climate regions; (4) they often did not eliminate the impact of serial correlations on the trends of ET_o and meteorological variables; and finally (5) they did not study the extremum values of meteorological variables and ET_o. This study overcame the abovementioned shortcomings by (1) analyzing the 50-year (1961–2010) annual trends of ET_o and 12 meteorological variables from 18 study sites in contrasting climate types in Iran, (2) removing the effect of serial correlations on the trends analysis via the trend-free pre-whitening approach, (3) determining the most important meteorological variables that control the variations of ET_o, and (4) evaluating the coincidence of annual extremum values of meteorological variables and ET_o. The results showed that ET_o and several meteorological variables (namely wind speed, vapor pressure deficit, cloudy days, minimum relative humidity, and mean, maximum and minimum air temperature) had significant trends at the confidence level of 95% in more than 50% of the study sites. These significant trends were indicative of climate change in many regions of Iran. It was also found that the wind speed (WS) had the most significant influence on the trend of ET_o in most of the study sites, especially in the years with extremum values of ET_o. In 83.3% of the study sites (i.e., all arid, Mediterranean and humid regions and 66.7% of semiarid regions), both ET_o and WS reached their extremum values in the same year. The significant changes in ET_o due to WS and other meteorological variables have made it necessary to optimize cropping patterns in Iran.

1. Introduction

Assessment of changes in reference evapotranspiration (ET_o) is required in climate change studies, agricultural and forest meteorology, irrigation scheduling, surface water balance, drought analysis, long-term decision-making in food and water security policies, and optimum allocation of water resources [1,2,3,4,5,6,7]. Evaluating the trends of meteorological variables may help determine the effects of major factors on ET_o and climate change [8,9].

Dadaser-Celik et al. [10] evaluated the 32-year trend of ET_o in Turkey. Analysis of climatic data showed an upward trend in air temperature, and downward trends in wind speed and relative humidity in Turkey. Changes in these three variables explained the majority of variations in ET_o. Song et al. [11] assessed the 46-year trend of ET_o in the North China Plain. Their results indicated that the downward trends of net radiation and wind speed had a bigger impact on ET_o compared to the upward trends of maximum and minimum air temperature. Wang et al. [12] characterized the 31-year trend of ET_o in the western Heihe River Basin in China. They found that wind speed and sunshine duration were the two key meteorological variables that decreased ET_o.

Darshana et al. [13] investigated the 30-year trend of ET_o in the Tons River Basin in central India. Their outcomes showed that maximum air temperature and net radiation had a stronger impact on ET_o compared to minimum air temperature and relative humidity. Zongxing et al. [14] studied the 49-year trend of ET_o in the south-west of China. The decrease in wind speed was the main driving force for the reduction of ET_o. This happened because the lower wind speed raised the vapor pressure, which ultimately reduced the evaporative demand of atmosphere. Li et al. [15] examined the 46-year trend of ET_o in the Upper Mekong River Basin. They showed that sunshine duration had a more significant (at the confidence level 95%) effect on ET_o compared to air temperature, relative humidity, and wind speed. Gao et al. [16] evaluated the 45-year trend of ET_o in China, and found that sunshine duration, wind speed, and relative humidity had a more important influence on ET_o than air temperature. Zhang et al. [17] assessed the changes in ET_o and its controlling factors in China. They indicated that maximum air temperature, relative humidity, and wind speed affected ET_o. A number of studies (e.g., [8,9,18,19,20]) showed that the increasing and decreasing trends of ET_o were mainly due to the increase in air temperature and decrease in wind speed, respectively.

Tabari et al. [21,22] evaluated the 40-year (1966–2005) trend of ET_o in the west and southwest of Iran with arid and semiarid climates. They studied the effect of air temperature, relative humidity, vapor pressure, wind speed, and rainfall on ET_o, and found that wind speed had the most influence on ET_o. Unfortunately, they did not consider the impact of serial correlations (autocorrelation) on the trends of ET_o and meteorological variables. Kousari and Ahani [23] assessed trends of ET_o and six meteorological variables (mean, minimum and maximum air temperature, relative humidity, wind speed and sunshine hours) in three climate regions (arid, semiarid and humid) of Iran during 1975–2005. They did not take into account the influence of serial correlations on the trends. Shadmani et al. [24] evaluated the 41-year (1965–2005) trend of ET_o from 11 weather stations in arid regions of Iran but did not specify which meteorological variables controlled trend of ET_o.

The existing studies (1) evaluated the impact of various climatic variables on ET_o and (2) showed the signals of climate change in Iran based on upward/downward trends of ET_o and meteorological variables [8,9,20,21,22,23]. However, they suffer from the following shortcomings: (1) they used data series of less than 50 years which are suitable only for evaluating climate variability and not climatic change. At least a 50-year period is necessary to study climate change [25,26,27,28,29,30,31,32]. Borman [26] used a 50-year period to analyze the sensitivity of ET_o to climate change in Germany. Kingston et al. [27] found a considerable uncertainty in evaluating the changes of ET_o due to climate change using 30-year data. Several studies assessed the response of ET_o to climate change in China using 50-year data [28,29,30,31]. Hence, there is a consensus in the literature to use at least a 50-year period to evaluate the trend of ET_o due to climate change. (2) They evaluated the individual impact of a few climatic variables on ET_o, and thus could not represent the interactive effects of all forces driving trends of ET_o. (3) They mostly studied trends of ET_o and meteorological variables in similar climate regions. (4) They often did not eliminate the impact of serial correlations (autocorrelation) on the trends of ET_o and meteorological variables. Finally (5) they did not assess if the extremum values of meteorological variables and ET_o occur simultaneously. This study overcame the abovementioned drawbacks by analyzing the 50-year (1961–2010) trends of ET_o estimates and 12 meteorological variables in Iran using the trend-free pre-whitening Mann–Kendall (TFPW-MK) and Spearman’s Rho tests. These sites were chosen to cover four different climates, namely arid, semiarid, Mediterranean and humid. Moreover, variations of 12 meteorological variables (mean, maximum, and minimum air temperature, difference between the maximum and minimum air temperature, rainfall, wind speed and direction, mean and minimum relative humidity, vapor pressure deficit, number of cloudy days, and sunshine hours) were investigated to characterize forces that drive trends of ET_o. Furthermore, the effect of serial correlations on the trends was eliminated by the TFPW approach [24]. Finally, during the period 1961–2010, the coincidence of annual extremum values of ET_o and meteorological variables was studied to identify the main meteorological variables that affect variations of ET_o.

2. Data

Daily meteorological data in the 18 study sites were downloaded from the Iran Meteorological Organization (IMO) archive. These data cover a period of 50 years (1961–2010), and include mean, minimum, and maximum daily air temperature (°C), vapor pressure deficit (kPa), mean and minimum relative humidity (%), wind speed (m/s) at the screen-height of 2 m, rainfall (mm/day), number of cloudy days, and number of sunshine hours per day (hr/day).

Characteristics of the study sites are indicated in

Table 1. Location, altitude, and climate of the 18 study sites.

Synoptic Station	ICAO Code	Latitude (°N)	Longitude (°E)	Altitude (masl)	Climate Type
Ahvaz (AH)	40811	31°20′	48°40′	22.5	Arid
Arak (AR)	40769	34°6′	49°46′	1708.0	Semiarid
Bushehr (BU)	40858	28°58′	50°49′	9.0	Arid
Esfahan (ES)	40800	32°37′	51°40′	1550.4	Arid
Hamedan (HA)	40768	34°52′	48°32′	1741.5	Semiarid
Jiroft (JI)	40866	28°35′	57°48′	601.0	Arid
Kerman (KE)	40841	30°15′	56°58′	1753.8	Arid
Mashhad (MA)	40745	36°16′	59°38′	999.2	Semiarid
Moghan (MO)	40700	39°39′	47°55′	31.9	Semiarid
Qazvin (QA)	40731	36°15′	50°3′	1279.2	Semiarid
Rasht (RA)	40719	37°19′	49°37′	−8.6	Humid
Sanandaj (SA)	40747	35°20′	47°0′	1373.4	Mediterranean
Shahrekord (SK)	40798	32°17′	50°51′	2048.9	Semiarid
Shiraz (SH)	40848	29°32′	52°36′	1484.0	Semiarid
Tabriz (TA)	40706	38°5′	46°17′	1361.0	Semiarid
Urmia (UR)	40712	37°40′	45°3′	1328.0	Semiarid
Yazd (YA)	40821	31°54′	54°17′	1237.2	Arid
Zabol (ZA)	40829	31°2′	61°29′	489.2	Arid

ICAO: International Civil Aviation Organization; masl: meter above sea level.

. Figure 1 shows the location of these sites in Iran. They are chosen to cover various climate regions (arid, semiarid, humid, and Mediterranean) across Iran.

3. Models and Methods

3.1. FPM Equation

ET_o is the rate of evapotranspiration from a uniform height, actively growing, well-watered, and completely shaded hypothetical crop [33]. The hypothetical crop has a height 0.12 (m), a fixed surface resistance of 70 (s/m), and an albedo of 0.23, closely resembling the evapotranspiration from an extensive surface of green grass [26]. In this study, daily meteorological variables from the 18 study sites were used in the Agricultural Organization of the United Nation (FAO)-Penman–Monteith (FPM) equation to estimate ET_o on a daily basis. ET_o estimates from the FPM equation were validated against the lysimeter measurements in the 18 study sites [34,35,36,37].

The FPM equation is given by [33]:

$$E{T}_o=\frac{0.408\left(R_n-G\right)+\gamma \frac{900}{T+273}u\left(es-ea\right)}{\mathsf{\Delta }+\gamma \left(1+0.34u\right)}$$

(1)

where ET_o is the daily reference evapotranspiration (mm/day), γ is the psychrometric constant (kPa/°C), Δ is the slope of the saturation vapor pressure–temperature curve (kPa/°C), T is the mean daily air temperature (°C), and u is the mean daily wind speed at the screen-height of 2 m (m/s). G is the ground heat flux (MJ/m²/day) and is negligible on daily timescales [21,33].

es and ea are, respectively, the saturation and actual vapor pressures (kPa), which are given by [33]:

$$es=\frac{e_{\left(Tmax\right)}+e_{\left(Tmin\right)}}{2}$$

(2)

$$ea=\frac{e_{\left(Tmin\right)}\frac{RHmax}{100}+e_{\left(Tmax\right)}\frac{RHmin}{100}}{2}$$

(3)

where $e_{\left(Tmax\right)}$ and $e_{\left(Tmin\right)}$ are the saturation vapor pressures (kPa) at daily maximum and minimum air temperatures, respectively. RHmax and RHmin are the daily maximum and minimum relative humidity (%), respectively.

R_n is the net radiation (MJ/m²/day), and is estimated by [21,33]:

$$R_n=\left(a_s+b_s\frac{n}{n_{max}}\right)R_a$$

(4)

where n is the number of sunshine hours per day (hr/day), n_max is the maximum possible duration of sunshine per day (hr/day), and R_a is the extraterrestrial radiation (MJ/m²/day). R_a and n_max depend on the latitude and julian day [33]. a_s and b_s are empirical coefficients, which are obtained from [21] for each study site.

In this study, the annual averages of daily FPM ET_o estimates and meteorological variables were used to analyze the annual trends.

3.2. Mann–Kendall (MK) Test

One of the most well-known methods for detecting trends in a time series is the non-parametric Mann–Kendall (MK) test [38,39,40]. Unlike the parametric statistical approaches, the non-parametric statistical tests are more suitable for non-Gaussian distributed data, which are frequently observed in hydrologic time series [34]. The MK test is defined as follows [38,39,40]:

$$S={\displaystyle \sum }_{i=1}^{N-1}{\displaystyle \sum }_{j=i+1}^{N}sign\left(x_j-x_i\right)$$

(5)

$$sign\left(x_j-x_i\right)=\{\begin{array}{c}1 if \left(x_j-x_i\right)>0\\ 0 if \left(x_j-x_i\right)=0\\ -1 if \left(x_j-x_i\right)<0\end{array}$$

(6)

$$VAR\left(S\right)=\frac{1}{18}\left[N\left(N-1\right)\left(2N+5\right)-{\displaystyle \sum }_{p-1}^g{t}_p\left(t_p-1\right)\left(2t_p+5\right)\right]$$

(7)

If S > 0 then Z = (S − 1)/VAR (S) ^0.5

(8a)

If S = 0 then Z = 0

(8b)

If S < 0 then Z = (S + 1)/VAR(S) ^0.5

(8c)

where N is the number of data, S is the summation of signs, VAR is the variance, and x_j and x_i are the data values in years j and i, respectively (with j > i). t_p and g indicate ties and the number of ties, respectively. The MK test determines whether to reject the null hypothesis (H0) or accept the alternative hypothesis (Ha), where H0: no monotonic trend is present and Ha: a monotonic trend is present. If 1.65 < Z ≤ 1.96, 1.96 < Z ≤ 2.58, Z > 2.58, the trend is significant at the confidence levels of 90%, 95%, and 99%, respectively [38,39,40].

Local (at-site) significance levels for each trend test can be obtained from

$$p=2\left[1-\Phi \left(\left|Z\right|\right)\right]$$

(9)

where | | denotes the absolute value, and $\Phi ( )$ is defined as,

$$\Phi \left(\left|Z\right|\right)=\frac{1}{\sqrt{2\pi }}{{\displaystyle \int }}_0^{\left|Z\right|}\exp \left(\frac{t^2\theta }{2}\right)dt$$

(10)

where θ is the random variable, and t is the variable for which the cumulative distribution function should be calculated. If p ≤ α, the existing trend is statistically significant at the significance level of α [41,42].

Sometimes the MK test detects trends because of the serial correlations of time series data, which leads to an increased rejection rate of the null hypothesis [41,43].

Hydrologic time series often have significant serial correlations that can undermine the ability of the MK test to correctly assess the significance of trend [41,43].

In this study, trend-free pre-whitening (TFPW) was used to effectively eliminate the impact of serial correlations on the MK test [41,43,44,45].

This method is defined by the following steps:

1. Calculate the slope of trend using the Sen’s slope estimator [43,44]:

$$Q_i=\frac{x_j-x_k}{j-k}$$

(11)

$$Q=\{\begin{array}{c}{Q}_{\frac{N+1}{2}} N is odd\\ \frac{1}{2}\left(Q_{\frac{N}{2}}+Q_{\frac{N+2}{2}}\right) N is even\end{array}$$

(12)

where Q stands for the slope of trend. Q_i is the Sen’s slope estimator for each value of i, x_j and x_k are the numerical values at times j and k (j > k), respectively.

2. There is no trend if Q is equal to zero. Otherwise, it is assumed that the existing trend is monotonic, and the time series data are de-trended as follows:

X_t^′=X_t − Qt

(13)

where X_t^′ is the lag-1 autocorrelation of the de-trended time series and X_t is the autocorrelation of time series.

3. Using the rank correlation coefficient estimator, the lag-1 autocorrelation of the de-trended time series is estimated by replacing the sample data by their ranks as follows [45]:

$$r_j=\frac{\frac{1}{n-j}{\sum }_{i=1}^N\left(X_i-\overline{X}\right)\left(X_{i+j}-\overline{X}\right)}{\frac{1}{n}{\sum }_{i=1}^N{\left(X_{i+j}-\overline{X}\right)}^2}$$

(14)

where r_j is the lag-j autocorrelation coefficient, and $\overline{X}$ is the average of the data. Then, the estimated lag-1 autocorrelation is removed from the time series as follows:

$$X_t^{″}=X_t^{\prime }-r_j{X}_{t-1}^{\prime }$$

(15)

4. Adding the removed trend in step 2:

$$X_t^{‴}=X_t^{″}+Qt$$

(16)

3.3. Spearman’s Rho Test

The Spearman’s Rho test was applied to investigate correlation among all the meteorological variables and ET_o estimates from the FPM equation.

The Spearman’s Rho test is defined as [18]:

$$\rho =1-\frac{6{\sum }_{i=1}^N{\left(x_i-i\right)}^2}{N\left(N^2-1\right)}$$

(17)

$$Z=\rho \sqrt{N-1}$$

(18)

where ρ is the correlation coefficient of linear regression between series i (the order of the data in the original series) and x (data values) [18]. If |Z| > Z_α at a significance level of α, then the null hypothesis of no trend is rejected [18].

Following the literature [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24], the confidence levels of 90%, 95%, and 99% were used in this study to evaluate the trends of ET_o and meteorological variables.

If there is no mention of confidence level, it meant a confidence level of 95% was adopted. Otherwise, we explicitly mention the confidence levels of 90% and 99% wherever they were used.

4. Results and Discussions

4.1. Comparison of p-Values from the MK and TFPW-MK Tests

Table 2. Estimated p-values from the Mann–Kendall (MK) and trend-free pre-whitening Mann–Kendall (TFPW-MK) tests in the 18 study sites.

Sites	Method	ETo	Tmean	Tmin	Tmax	Tmax-Tmin	es-ea	RH	RHmin	P	WD	WS	CD	n
Ahvaz	TFPW-MK	0.146	0.000	0.000	0.001	0.000	0.170	0.725	0.000	0.598	0.430	0.907	0.040	0.041
	MK	0.276	0.000	0.000	0.001	0.000	0.059	0.725	0.000	0.598	0.061	0.928	0.003	0.101
Arak	TFPW-MK	0.003	0.682	0.358	0.980	0.255	0.332	0.645	0.462	0.066	0.101	0.034	0.353	0.719
	MK	0.000	0.682	0.358	0.981	0.255	0.332	0.632	0.462	0.066	0.581	0.034	0.420	0.464
Bushehr	TFPW-MK	0.184	0.000	0.000	0.002	0.006	0.031	0.238	0.270	0.340	0.121	0.003	0.146	0.011
	MK	0.065	0.000	0.000	0.007	0.006	0.031	0.349	0.270	0.301	0.002	0.003	0.041	0.011
Esfahan	TFPW-MK	0.000	0.000	0.034	0.000	0.153	0.001	0.155	0.058	0.063	0.828	0.000	0.001	0.096
	MK	0.000	0.000	0.034	0.000	0.153	0.001	0.155	0.058	0.013	0.944	0.000	0.001	0.005
Hamedan	TFPW-MK	0.987	0.019	0.014	0.358	0.143	0.980	0.024	0.110	0.732	0.920	0.366	0.017	0.137
	MK	0.987	0.019	0.014	0.414	0.143	0.980	0.024	0.048	0.732	0.965	0.366	0.017	0.137
Jiroft	TFPW-MK	0.037	0.291	0.717	0.025	0.003	0.085	0.131	0.027	0.004	0.608	0.349	0.110	0.566
	MK	0.037	0.291	0.717	0.025	0.003	0.085	0.131	0.027	0.002	0.769	0.008	0.110	0.566
Kerman	TFPW-MK	0.032	0.000	0.000	0.000	0.047	0.412	0.622	0.375	0.039	0.119	0.014	0.000	0.034
	MK	0.000	0.000	0.000	0.000	0.047	0.412	0.193	0.475	0.001	0.001	0.014	0.000	0.034
Mashhad	TFPW-MK	0.001	0.000	0.000	0.002	0.000	0.001	0.020	0.040	0.947	0.043	0.128	0.744	0.472
	MK	0.001	0.000	0.000	0.002	0.000	0.005	0.020	0.040	0.947	0.288	0.085	0.744	0.277
Moghan	TFPW-MK	0.359	0.004	0.000	0.113	0.574	0.441	0.091	0.574	0.692	0.278	0.348	0.032	0.441
	MK	0.359	0.004	0.000	0.225	0.574	0.441	0.091	0.539	0.692	0.058	0.348	0.032	0.441
Qazvin	TFPW-MK	0.000	0.362	0.457	0.417	0.394	0.000	0.001	0.035	0.610	0.316	0.000	0.001	0.847
	MK	0.000	0.362	0.457	0.417	0.394	0.000	0.000	0.035	0.610	0.038	0.000	0.001	0.803
Rasht	TFPW-MK	0.375	0.000	0.000	0.558	0.000	0.003	0.001	0.340	0.598	0.472	0.569	0.452	0.328
	MK	0.097	0.000	0.000	0.351	0.000	0.001	0.000	0.445	0.598	0.803	0.569	0.262	0.328
Sanandaj	TFPW-MK	0.358	0.000	0.047	0.000	0.120	0.013	0.124	0.000	0.011	0.178	0.913	0.001	0.046
	MK	0.161	0.000	0.052	0.000	0.198	0.013	0.119	0.000	0.011	0.589	0.727	0.000	0.090
Shahrekord	TFPW-MK	0.000	0.001	0.000	0.195	0.091	0.000	0.106	0.000	0.757	0.389	0.000	0.094	0.320
	MK	0.000	0.001	0.002	0.195	0.091	0.000	0.106	0.000	0.610	0.061	0.000	0.175	0.320
Shiraz	TFPW-MK	0.024	0.000	0.000	0.001	0.001	0.000	0.037	0.012	1.000	0.821	0.000	0.738	0.457
	MK	0.024	0.000	0.000	0.001	0.001	0.000	0.037	0.002	1.000	0.933	0.000	0.738	0.415
Tabriz	TFPW-MK	0.225	0.000	0.000	0.000	0.744	0.000	0.000	0.000	0.000	0.682	0.634	0.000	0.213
	MK	0.193	0.000	0.000	0.000	0.744	0.000	0.000	0.000	0.006	0.864	0.634	0.000	0.091
Urmia	TFPW-MK	0.015	0.069	0.061	0.022	0.375	0.320	0.375	0.000	0.153	0.130	0.021	0.002	0.049
	MK	0.006	0.054	0.061	0.022	0.375	0.076	0.490	0.000	0.107	0.367	0.021	0.002	0.014
Yazd	TFPW-MK	0.763	0.000	0.000	0.000	0.000	0.001	0.076	0.040	0.384	0.206	0.050	0.068	0.019
	MK	0.679	0.000	0.000	0.000	0.000	0.000	0.149	0.185	0.384	0.093	0.013	0.112	0.009
Zabol	TFPW-MK	0.000	0.000	0.000	0.011	0.751	0.201	0.349	0.007	0.403	0.101	0.000	0.462	0.763
	MK	0.000	0.002	0.000	0.000	0.775	0.201	0.460	0.042	0.001	0.101	0.000	0.386	0.763

compares p-values from the MK and TFPW-MK tests for ET_o and the 12 meteorological variables in the 18 study sites.

The p-value is a random variable derived from the distribution of the test statistic to analyze a data set and to test a null hypothesis [41,42]. If p-values from the abovementioned tests are different, but they result in an identical trend for a particular time series (i.e., an insignificant trend or a significant trend at the same confidence level), the cells in Table 2 are highlighted in yellow color. If the two tests yield different trends for a specific time series (i.e., one leads to a significant trend, while the other one results in an insignificant trend, or they both lead to a significant trend but with different confidence levels), the cells are highlighted in red color.

As shown, in each study site, p-values from the MK and TFPW-MK tests are different at least for three meteorological variables (yellow and red cells). Overall, 38% of the obtained p-values from the two tests are different due to the serial correlations of time series data. In each study site, the two tests also lead to different trends for at least one variable (red cells). As shown in Table 2, 15% of the trends from two tests are different. These results indicate that the serial correlation undermine the ability of the MK test to correctly determine the trends and their confidence level [41,43]. Hence, the TFPW method should be applied to eliminate the impact of the serial correlations on the MK test.

As can be seen in Table 2, some meteorological variables are subject to different p-values from the MK and TFPW-MK test more often than others. For example, wind direction (WD), wind speed (WS), and rainfall (P) were flagged for more study sites compared to mean air temperature (Tmean) and minimum air temperature (Tmin). This happens because WD, WS, and P time series have more serial correlations compared to Tmean and Tmin time series. Similar findings were reported in other studies [46,47,48,49].

4.2. Trends of ET_o and Meteorological Variables in the Study Sites

Variations of the meteorological variables were assessed in all of the study sites to identify the ones that control the trend of annual mean of ET_o. However, herein, the trends are presented only in two sites, Kerman and Qazvin (Figure 2 and Figure 3).

4.2.1. Kerman

Figure 2 shows the 50-year trends in the annual averages of daily FPM ET_o estimates, and meteorological variables including mean (Tmean), minimum (Tmin), and maximum (Tmax) air temperature, the difference between maximum and minimum air temperature (Tmax-Tmin), vapor pressure deficit (es-ea), relative humidity (RH), minimum relative humidity (RHmin), rainfall (P), wind speed and direction (WS and WD), number of cloudy days (CD), and sunshine hours (n) in Kerman.

As can be seen, the significant upward trends in Tmean, Tmin, Tmax, and n (at confidence level 95%) as well as the substantial downward trends in CD, Tmax-Tmin, P, and WS (at confidence level 95%) led to a negative trend (at confidence level 95%) in the FPM ET_o estimates.

Although the upward trends in Tmax, Tmin, Tmean, and n could potentially increase ET_o, a downward trend in WS and (es-ea) led to a significant decreasing trend in the FPM ET_o values. It is worth mentioning that the minimum of both WS and ET_o occurred in the same years (i.e., 1982 and 1984) (see the arrows in Figure 2). Similarly, the course of (es-ea) shows lower values in 1982–1986. Thus, WS and to a lesser extent (es-ea) may be the driving force for the variations of ET_o in Kerman.

Although CD showed the strongest downward trend among all the meteorological variables in Kerman, it could not capture the variations of ET_o as good as WS. Both WS and ET_o showed significant trends at a confidence level of 95%, while CD showed a significant trend at a confidence level of 99%.

During 1960–1980s, the north-western wind was prevailing, which blows from the Zangi Abad region with an arid climate into Kerman [50,51]. However, from 1980 until 2010, the northern wind was dominant, which comes from Chatroud region with dry temperate climate [50,51]. Hence, the change in wind direction (WD) reduced the impact of the arid climate in Kerman, and consequently decreased ET_o.

The decreasing rate of Tmax-Tmin (−0.3%/decade) was due to a more rapid increase in Tmin (+3.8%/decade) compared to Tmax (+0.8%/decade). Tmin, P, WS, and CD had the highest changes (larger than ±2% per decade), implying climate change in Kerman [50,51]. The lowest P and CD, and the highest Tmax and n values (during the 50-year period) were observed in 2010, which may an indication of drought in this region [50,52,53,54,55,56,57]. These results show the complexity of forces driving the trend of ET_o and may highlight importance of considering different ET_o models and variables in various climates for local and global studies.

4.2.2. Qazvin

Figure 3 shows the 50-year trends of annually averaged ET_o and meteorological variables in Qazvin. As can be seen in Figure 3, the significant upward trend in RH as well as substantial downward trends in es-ea, WS, CD, and RHmin (at the confidence level 95%) led to a negative trend in the FPM ET_o estimates. It is worth mentioning that these significant trends reduced ET_o and may cause alarming climate change in Qazvin [57,58,59]. During the 50-year study period (1961–2010), Tmax reached its highest value in 2010. It should be noted that the minimum es-ea and ET_o occurred in 1991 (see the arrows in Figure 3). The highest decreasing rate was for WS (−6.8%/decade). RH showed an upward trend as a result of decreasing WS and es-ea (Figure 3). The warm dry south-eastern winds (called Raz or Shareh) originated from the arid areas of central Iran (Yazd). They were prevailing from 1961 to 2010 and decreased RHmin in Qazvin [60,61,62,63].

4.3. Trends of ET_o and Meteorological Variables in Iran

Table 3. Trends of Agricultural Organization of the United Nation (FAO)-Penman–Monteith (FPM) evapotranspiration (ET_o) estimates and 12 meteorological variables in the 18 study sites. U and D represent the significant upward and downward trends, respectively. NS denotes no significant trend.

Variations	ETo	Tmean	Tmin	Tmax	Tmax-Tmin	es-ea	RH	RHmin	P	WD	WS	CD	n
Ahvaz	NS	U ***	U ***	U ***	D ***	NS	NS	D ***	NS	NS	NS	U **	U **
Arak	U ***	NS	NS	NS	NS	NS	NS	NS	D *	NS	U **	NS	NS
Bushehr	NS	U ***	U ***	U ***	D ***	U **	NS	NS	NS	NS	D ***	NS	D **
Esfahan	D ***	U ***	U **	U ***	NS	U ***	NS	D **	U *	NS	D ***	D ***	U *
Hamedan	NS	U **	U **	NS	NS	NS	U **	NS	NS	NS	NS	U **	NS
Jiroft	U **	U **	NS	U **	U ***	U *	NS	D **	D ***	NS	NS	NS	NS
Kerman	D **	U ***	U ***	U ***	D **	NS	NS	NS	D **	NS	D **	D ***	U **
Mashhad	U ***	U ***	U **	U ***	D ***	U ***	D **	D **	NS	D **	NS	NS	NS
Moghan	NS	U ***	U ***	NS	NS	NS	U *	NS	NS	NS	NS	D **	NS
Qazvin	D ***	NS	NS	NS	NS	D ***	U ***	D **	NS	NS	D ***	D ***	NS
Rasht	NS	U ***	U ***	NS	D ***	D ***	U ***	NS	NS	NS	NS	NS	NS
Sanandaj	NS	U ***	U **	U ***	NS	D **	NS	D ***	D **	NS	NS	D ***	U **
Shahrekord	U ***	D ***	D ***	NS	U *	D ***	NS	D ***	NS	NS	U ***	D *	NS
Shiraz	D **	U ***	U ***	U ***	D ***	U ***	D **	D **	NS	NS	D ***	NS	NS
Tabriz	NS	U ***	U ***	U ***	NS	U ***	D ***	D ***	D ***	NS	NS	D ***	NS
Urmia	U **	U **	U *	U **	NS	NS	NS	D ***	NS	NS	U *	D ***	U *
Yazd	NS	U ***	U ***	U ***	D ***	U ***	D *	D **	NS	NS	D **	U *	U **
Zabol	U ***	U ***	U ***	U **	NS	NS	NS	D ***	NS	NS	U ***	NS	NS
U (%)	33.3	77.8	77.8	66.7	11.1	38.9	22.2	0.0	5.6	0.0	22.2	16.7	33.3
D (%)	22.2	5.6	5.6	0.0	38.9	22.2	22.2	66.7	27.8	5.6	33.3	44.4	5.6
U+D (%)	55.5	83.4	83.4	66.7	50.0	61.1	44.4	66.7	33.4	5.6	55.5	61.1	38.9

*, **, *** are significant trends at a confidence level of 90%, 95%, and 99%, respectively.

and Figure 4 summarize trends of FPM ET_o estimates and meteorological variables in the 18 study sites. The FPM ET_o estimates showed significant upward and downward trends, respectively, in 33.3% (28.6% of arid regions vs. 44.4% of semiarid regions) and 22.2% (28.6% of arid regions vs. 22.2% of semiarid regions) of the study sites (Table 3 and Figure 4). Hence, the FPM ET_o retrievals had significant variations in most parts of Iran (57.2% of arid regions vs. 66.6% of semiarid regions), which may indicate climatic change signals.

Results showed that ET_o had a significant positive (negative) trend in the west and east (center) of Iran. WS was the only meteorological variable with the same rising/falling trend.

As shown in Table 3, in eight of the study sites (i.e., 44.4% of sites), there were significant trends in both ET_o and Tmean/RHmin/WS. Moreover, both ET_o and Tmin/Tmax/es-ea showed significant trends in 33.3% of the study sites. Hence, these meteorological variables could be considered as driving forces to control variations of ET_o in Iran.

WS had significant upward and downward trends in 22.2% (14.3% of arid regions vs. 33.3% of semiarid regions) and 33.3% (57.1% of arid regions vs. 22.2% of semiarid regions) of the study sites, respectively. Therefore, significant variations in WS were observed in 55.5% (71.4% of arid regions vs. 55.5% of semiarid regions) of the sites. Table 3 and Figure 2, Figure 3 and Figure 4 indicate that WS is the driving force for the trends of ET_o in 55.5% of the study sites. Thus, WS may be the most important variable that controls the variations of ET_o, particularly in the arid regions. These findings are further validated by our outcomes in Table 5 (Section 4.6) that lists the three meteorological variables with the highest correlation with ET_o. As shown, WS has the highest significant correlation with ET_o in all the study sites expect Rasht (RA). Studying the trends of ET_o and meteorological variables over longer time periods and more sites across Iran leads to more reliable results.

Table 3 and Figure 4 also showed that there were significant upward trends for Tmean and Tmax in 77.8% (85.7% of arid regions vs. 55.6% of semiarid regions) and 66.7% (100.0% of arid regions vs. 33.3% of semiarid regions) of the study sites, respectively. In Jiroft, Mashhad, and Urmia (3 out of 18 study sites, i.e., 16.7% study sites), there were upward significant trends between ET_o and Tmean/Tmax (Table 3). There was a significant downward trend for RHmin in 66.7% (71.4% of arid regions vs. 66.7% of semiarid regions) of the study sites (Figure 4).

Tmax, Tmean, Tmin, es-ea, and n had significant increasing trends, respectively, in 100.0%, 85.7%, 85.7%, 57.1%, and 57.1% of the arid sites. On the other hand, RHmin, Tmax-Tmin, and WS had significant decreasing trends in 71.4%, 57.1%, and 57.1% of the arid sites, respectively. These results indicated that the arid sites were affected more than the semiarid ones by climate change. In most of the semiarid sites, Tmin and Tmean showed increasing, and RHmin and CD indicated decreasing trends.

According to the results, P, WS, and CD showed significant trends in 42.9%, 28.6%, and 28.6% of the arid sites, respectively. WS, Tmin, RHmin, P, and CD had substantial trends in 44.4%, 22.2%, 11.1%, 11.1%, and 11.1% of the semiarid sites, respectively. Furthermore, WS, Tmin, and CD had a rate of more than ±2%/decade in 66.7% (85.7% of arid regions vs. 66.7% of semiarid regions), 66.7% (42.9% of arid regions vs. 77.8% of semiarid regions), and 55.6% (57.1% of arid regions vs. 55.6% of semiarid regions) of the study sites, respectively.

As mentioned above, variations of ET_o in each study site are controlled by meteorological variables such as air temperature and wind speed. The Arak and Esfahan sites show contrasting trends in ET_o although they are relatively close to each other. This happens because wind speed (which is the dominant controlling variable in both sites) has an increasing (decreasing) trend in Arak (Esfahan). Furthermore, Arak and Esfahan have different climate types. Arak is located in a cold mountainous area, while Esfahan is placed in a hot flat desert.

The focus of this study was to evaluate the annual trends of ET_o and meteorological variables over a 50-year period (1961–2010). However, according to Table 3 and Figure 4, we can conclude that their seasonality is important. For example, the annual minimum air temperature, Tmin (that occurs in cold seasons, i.e., fall–winter) had significant increasing trends in 83.4% of sites. While, the annual maximum air temperature, Tmax (which happens in warm seasons, i.e., spring–summer) showed significant rising trends in 66.7% of sites. In addition, the significant downward trends of Tmax-Tmin were observed in 41.5% of sites, whereas its upward trends were seen only in 9.3% of sites. Its significant downward trends were due to the higher increasing rate of Tmin in cold seasons than that of Tmax in warm seasons. These results imply that the cold seasons have more influence on the significant trends than warm seasons. It should be noted that these findings are primitive, and future studies should be directed towards evaluating the seasonal variations of ET_o and other climatic variables over long periods.

4.4. Range of Variations of ET_o and Meteorological Variables

Figure 5 shows the box plots for the 50-year variations of ET_o and meteorological variables in each study site.

As can be seen, the largest 50-year variations in both ET_o and WS happened in Zabol (ZA), which highlighted the significant influence of WS on ET_o.

The second highest variations of ET_o was seen in Ahvaz (AH). AH had also the highest variations of Tmean compared to the other study sites. Hence, the variations of ET_o in Ahvaz can be related to that of Tmean.

The lowest variations in Tmax-Tmin, es-ea, and n values were observed in Rasht (RA), which led to the smallest 50-year change in ET_o [9,10]. Most of the outliers occurred in 2010, which signaled a drought condition in different regions of Iran. The FPM ET_o values ranged from 2 to 11 mm day⁻¹ in various climates of Iran.

As shown in Equation (1), ET_o is affected by the atmospheric vapor pressure deficit (es-ea). As expected, a high positive correlation exists between ET_o and es-ea in Figure 5. For example, high values of ET_o and es-ea are seen in Zabol (ZA). Moreover, the lowest ET_o and es-ea values are observed in Rasht (RA). This happens because the atmospheric demand to water vapor increases as the vapor pressure deficit (es-ea) rises [33]. Positive correlations are also observed between ET_o and Tmax, and ET_o and n. A larger Tmax leads to more potential for converting soil moisture to water vapor and causes plants to open up their stomata and release more water vapor [33]. Furthermore, larger values of n yield higher R_n (Equation (4)) and ET_o (Equation (1)). On the other hand, a negative correlation is found between ET_o and RH. The highest ET_o and lowest RH values are seen in RA. This is due to the fact that the atmospheric demand for water vapor decreases as RH increases.

4.5. Annual Extremum Values of ET_o and Meteorological Variables during the Study Period (1961–2010)

Table 4. The years of occurrence of maximum and minimum values of ET_o and meteorological variables in each study site during the 50-year study period. If the annual extremum values of ET_o and a particular meteorological variable coincide, they are highlighted in green color.

Sites		ETo	Tmean	Tmin	Tmax	Tmax-Tmin	es-ea	RH	RHmin	P	WS	CD	n
Ahvaz (arid)	Max	1964	2010	2010	2010	1973	1973	1982	1961	1997	1964	1982	1998
	Min	1962	1969	1968	1992	1992	1984	1964	1990	1973	1979	1964	1992
Arak (semiarid)	Max	1973	1966	1966	2010	1964	1973	1992	1993	1969	1982	1974	1973
	Min	1966	1992	1983	1992	1996	1992	1973	2010	1973	1966	2010	1984
Bushehr (arid)	Max	1967	2010	2010	1962	1962	1981	1972	1963	1997	1967	1974	1970
	Min	1979	1964	1964	1972	2009	1972	1981	1981	2010	1998	2008	1992
Esfahan (arid)	Max	1963	2010	1998	2010	2000	2002	1972	1972	2006	1967	1968	1995
	Min	1996	1972	1972	1972	1993	1972	2002	2002	2008	1996	2010	1982
Hamedan (semiarid)	Max	1967	2010	1961	2010	1964	1964	1986	2008	1968	1967	1982	2005
	Min	1986	1972	1964	1972	1961	1986	1967	1978	1964	1961	1964	1972
Jiroft (arid)	Max	2007	2001	1999	2001	2001	1998	1991	1991	1992	2007	1992	2001
	Min	1996	1992	1992	1992	1997	1995	1998	1998	2001	1996	2001	2009
Kerman (arid)	Max	1966	2010	2009	2010	1973	1970	1983	1974	1974	1969	1963	2010
	Min	1984	1972	1973	1972	1976	1992	1998	1962	2010	1984	2010	1983
Mashhad (semiarid)	Max	2010	2010	2006	2010	1980	2010	1982	1982	1976	1999	1992	2000
	Min	1981	1972	1972	1972	1991	1982	2010	2010	2006	1977	1983	1993
Moghan (semiarid)	Max	1989	2010	2010	2010	1989	1989	2003	2003	2003	1989	1988	1999
	Min	1994	1993	1993	1993	2003	2003	1989	1989	1996	1994	2004	1987
Qazvin (semiarid)	Max	1962	1966	1966	1966	1961	1966	1992	1972	1972	1976	1969	1963
	Min	1992	1974	1976	1974	1972	1992	1966	1994	2007	1998	1995	1992
Rasht (humid)	Max	1975	2010	2010	2010	1971	1971	1988	1988	1972	1975	1977	1995
	Min	1993	1972	1972	1969	1984	1988	1967	1971	2010	1988	1989	1974
Sanandaj (Mediterranean)	Max	1971	2010	2010	2010	1983	1970	1982	1982	1969	1971	1969	2001
	Min	1986	1992	1983	1992	1967	1992	1997	2010	1973	1986	2010	1986
Shahrekord (semiarid)	Max	2008	1978	1978	1978	2010	1964	1982	1980	2006	2008	1986	1973
	Min	1968	1992	2005	1992	1986	1992	1973	2010	2008	1968	1977	1984
Shiraz (semiarid)	Max	1981	1999	1999	2010	1973	2001	1969	1969	2004	1981	1992	2001
	Min	1992	1972	1968	1992	1996	1972	1966	2010	1966	2010	2008	1992
Tabriz (semiarid)	Max	1961	2010	2010	2010	2010	2010	1982	1982	1963	1961	1969	1999
	Min	1991	1972	1972	1972	1982	1982	2010	2010	1990	1992	1999	1969
Urmia (semiarid)	Max	2010	2010	2001	2010	2010	2001	1982	1982	1994	2007	1969	1962
	Min	1992	1982	1982	1982	1969	1982	2001	2001	2005	1984	2009	1982
Yazd (arid)	Max	1971	2010	2010	2010	1979	1970	1982	1972	1986	1971	1986	2010
	Min	1995	1972	1964	1972	1996	1986	2010	2010	2010	1995	1966	1971
Zabol (arid)	Max	1984	2006	2006	2010	1971	1971	1991	1982	2005	1984	1991	1983
	Min	1968	1972	1972	1972	1982	1991	1971	2010	2010	1968	1963	2003

shows the years of occurrence of maximum and minimum values of ET_o and meteorological variables in each study site during the 50-year study period (i.e., 1961–2010).

If the annual extremum values of ET_o and a particular meteorological variable coincide, they are highlighted by green color (Table 4).

As can be seen, the maximum of both ET_o and WS in Ahvaz, Bushehr, Hamedan, Jiroft, Moghan, Rasht, Sanandaj, Shahrekord, Shiraz, Tabriz, Yazd, and Zabol happened in 1964, 1967, 1967, 2007, 1989, 1975, 1971, 2008, 1981, 1961, 1971, and 1984, respectively.

Similarly, the minimum of ET_o and WS in Arak, Esfahan, Jiroft, Kerman, Moghan, Sanandaj, Shahrekord, Yazd, and Zabol occurred in 1966, 1996, 1996, 1984, 1994, 1986, 1968, 1995, and 1968, respectively.

Overall, in 83.3% (all arid, Mediterranean and humid regions and 66.7% of the semiarid regions) of the study sites, annual extremum values of ET_o and WS occurred in the same year. These results imply that WS is the primary driver of the variations in ET_o in Iran. In addition, the annual extremum values of ET_o and RH coincided in 33.3% of the study sites, namely Ahvaz, Arak, Hamedan, Mashhad, Moghan, and Qazvin. Hence, RH is one of the main driving forces for variations of ET_o in these study sites.

Figure 6 illustrates the frequency of maximum and minimum values of ET_o and meteorological variables (namely Tmean, Tmax, Tmin, Tmax-Tmin, es-ea, P, RH, RHmin, CD, WS, WD, and n) in all the study sites for each year. Higher frequencies are observed in 1972, 1982, and 1992, indicating a large number of study sites reached their maximum and minimum values of ET_o and meteorological variables in these years. This happened because 1972, 1982, and 1992 are El Niño-Southern Oscillation (ENSO) years [64,65,66]. The highest frequency during the 50-year period is observed in 2010. This year was considered as a dry year in Iran and most regions of Iran experienced drought alarms [50,52,53,54,55,56,57].

4.6. Correlation between ET_o and Meteorological Variables

Figure 7 shows the correlation coefficient maps among the meteorological variables and reference evapotranspiration for the study sites in arid, Mediterranean, and humid regions. Similarly, Figure 8 indicates the correlation coefficient maps for the sites in semiarid region.

As indicated, WS was the only meteorological variable that had high correlation (ρ ≥ 0.8) with ET_o in Ahvaz, Esfahan, Jiroft, Yazd, Zabol, Sanandaj, Arak, Mashhad, Moghan, Qazvin, Shahrekord, Tabriz, and Urmia. These results imply that WS was the driving force for the variations of ET_o in most (72.2%) of the study sites in Iran, including 87.5% of arid regions and 77.8% of semiarid regions. This is in agreement with the results obtained by the TFPW-MK test (Table 3 and Figure 4).

WS and WD had the lowest correlation with the other meteorological variables. Hence, these variables may be controlled mainly by human activities such as desertification, land use change, and urban development [67,68,69,70,71,72,73,74,75,76,77,78,79]. The highest correlations were observed among the air temperature-related variables (i.e., Tmean, Tmax, Tmin, Tmax-Tmin) and humidity. In Mashhad and Urmia, the FPM ET_o estimates had a good correlation with all of the meteorological variables. The upward trends of Tmin in all the study sites in Iran (except Shahrekord due to its high elevation) indicated a signal of climate change and signaled the need for the optimization of cropping pattern [80,81].

Table 5. The total number of meteorological variables in each site with significant correlations with ET_o, and the three meteorological variables with the highest significant correlation with ET_o.

Sites	Number of Significant Meteorological Variables	Three Highest-Ranked (from Left to Right) Significant Meteorological Variables
Ahvaz	5	WS, RHmin, RH
Arak	7	WS, es-ea, RH
Bushehr	2	WS, n
Esfahan	3	WS, P, RH
Hamedan	8	WS, es-ea, RH
Jiroft	9	WS, Tmax-Tmin, P
Kerman	7	WS, es-ea, RH
Mashhad	12	WS, Tmax, Tmin
Moghan	4	WS, RH, es-ea
Qazvin	6	WS, es-ea, RH
Rasht	11	Tmax, RHmin, es-ea
Sanandaj	7	WS, P, n
Shahrekord	4	WS, RHmin, Tmax-Tmin
Shiraz	5	WS, RH, RHmin
Tabriz	10	WS, Tmax, es-ea
Urmia	12	WS, Tmax, RHmin
Yazd	9	WS, RH, es-ea
Zabol	10	WS, Tmax, WD

shows the total number of meteorological variables in each study site with significant correlations with ET_o. It also lists the three meteorological variables that have the highest significant correlation with ET_o. As can be seen, ET_o had the highest significant correlation with WS in all the sites except Rasht. ET_o indicated the largest (the second largest) correlation with Tmax in Rasht (Mahshad, Tabriz, Urmia, and Zabol). According to Table 5, in 66.7% of the study sites, more than five meteorological variables had significant correlations with ET_o. This highlights the complexity of the forces driving variations of ET_o.

In this study, the TFPW-MK and Spearman’s Rho tests were used to identify significant (at the confidence levels 90%, 95% and 99%) trends of ET_o and meteorological variables across Iran. These significant trends are indicative of climate change signals [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24]. For instance, in the Mashhad site, significant upward (downward) trends were seen for ET_o, Tmean, Tmin, Tmax, and es-ea (Tmax – Tmin, WD, RH and RHmin), suggesting climate change alarms. These findings are consistent with those of other studies [82,83,84,85,86,87]. In the Shahrekord site, Tmax – Tmin (Tmean, Tmin, RHmin, and CD) showed significant increasing (decreasing) trends, which can be considered as climate change signals. These findings are also in agreement with those of [8,9,20,21,22,23]. In the Shiraz site, significant upward (downward) trends of ET_o, Tmean, Tmin, Tmax, and es-ea (Tmax – Tmin, WS, RH and RHmin) represented signals of climate change [88]. In the Urmia site, growing (reducing) trends of ET_o, Tmax, WS, n, Tmean and Tmax (RHmin and CD) signified climate change [89,90,91,92]. Finally, in the Zabol site, upward (downward) trends of ET_o, Tmean, Tmin, and WS, and Tmax (RHmin) may be considered as alarms of climate change [84,93].

The FPM ET_o retrievals mainly depend on climate variables such as air temperature and humidity, wind speed, and solar radiation. This equation assumes a stomatal resistance of 70 s/m and an albedo of 0.23 for a standard 12 cm grass, and thus it ignores changes in stomatal resistance response to the elevated CO₂ and the surface albedo, ultimately causing uncertainty in the ET_o estimates. Surface albedo varies with changes in vegetation and soil moisture [94].

Elevated CO₂ emission influences plant physiology by diminishing stomatal conductance [95,96,97]. Ignoring changes in the surface albedo and vegetation response to the elevated CO₂ emission lead to uncertainty in the FPM ET_o estimates and results of this study. A similar uncertainty in the FPM ET_o estimates was reported by other studies. For example, Milly and Dunne [98] showed that ignoring the stomatal resistance response to the increased CO₂ emission in the FPM equation led to discrepancy between the ET_o estimates from climate models and the FPM equation. In a similar effort, Li et al. [99] indicated that the variations in the FPM ET_o estimates are only less than 10% for 40% changes in stomatal resistance. However, there are still high uncertainties in the amount of stomatal response to the raised CO₂ emission. For example, Domec et al. [100] reported that vegetation responses of pine to long term elevated CO₂ were manifested only when soil moisture was at a high level. Future studies should be directed toward evaluating the uncertainties in the FPM ET_o estimates with respect to climate change and drought condition [101,102].

5. Conclusions

This study assessed the trends of meteorological variables (namely, mean, minimum, and maximum air temperature, difference between the maximum and minimum air temperature, mean and minimum relative humidity, wind speed and direction, vapor pressure deficit, rainfall, and sunshine hours) and reference evapotranspiration (ET_o) in the 18 study sites in different climate regions of Iran.

The effect of meteorological variables on ET_o was also evaluated. Using the trend-free pre-whitening Mann–Kendall (TFPW-MK) and Spearman’s Rho tests, wind speed (WS) was found to be the most important variable that controls the trend of ET_o in most regions of Iran. Moreover, the increasing/decreasing trends of other meteorological variables (air temperature and humidity, rainfall, wind speed and direction, and sunshine hours) were indicative of climate change in many regions of Iran. The upward trend of minimum air temperature (Tmin) in all of the study sites in Iran (except Shahrekord due to its high elevation) indicated a signal of climate change and signaled the need for the optimization of cropping system. The significant increasing rate of Tmin may lead to a reduction in the growing degree day (GDD), ultimately decreasing the production of strategic and vital crops in future.

WS had significant upward and downward trends in 22.2% and 33.3% of the study sites, respectively. Therefore, significant variations in WS were observed in 55.5% of the investigated regions. The results also indicated that there were significant upward trends for mean air temperature (Tmean) and maximum air temperature (Tmax) in 77.8% and 66.7% of the study regions, respectively. In most of semiarid climates, Tmin and Tmean showed increasing, and minimum relative humidity (RHmin) and cloudy days (CD) indicated decreasing trends. This indicates that arid regions were affected more than semiarid areas by climate change. Precipitation (P), WS, and CD showed significant trends in 42.9%, 28.6%, and 28.6% of arid regions, respectively. However, WS, Tmin, RHmin, P, and CD had significant trends in 44.4%, 22.2%, 11.1%, 11.1%, and 11.1% of semiarid regions, respectively.

In 83.3% (all arid, Mediterranean and humid regions and 66.7% of semiarid regions) of the study sites, ET_o and WS reached their maximum and/or minimum in the same year(s). Furthermore, ET_o had the highest significant correlation with WS in all the sites except Rasht. This indicates that WS is the primary driver of the variations in ET_o in most weather stations of Iran, especially in the years with extremum values of ET_o. However, in a number of weather stations, the significant correlation between ET_o and other meteorological variables such as Tmax, Tmax-Tmin, es-ea, P, RH, RHmin, and n made it difficult to characterize driving forces for the trend of ET_o and signals of climate change. The results of this study highlighted the complexity of forces that drive variations of ET_o.