Hydroclimate Drivers and Spatiotemporal Dynamics of Reference Evapotranspiration in a Changing Climate

Abstract

Evapotranspiration (ET) variation is typically influenced by climatic factors, which are considered the primary drivers of agricultural water requirements. Any changes in ET rates directly affect crop water demands. In this study, temporal trends and magnitudes of key climatic variables, and their impacts on reference evapotranspiration (ETo) during 1981–2020, were evaluated across 36 districts of Punjab, Pakistan. Positive serial correlations, ranging from 0.29 to 0.48, were identified and removed using the pre-whitening technique. Increasing trends in maximum temperature (Tmax) and wind speed (WS) across Punjab and its subregions were observed, while relative humidity (RH) exhibited both increasing and decreasing trends. No significant trends were detected for the minimum temperature (Tmin). On a monthly scale, in the Southern Punjab (SP) region, Sen’s slope estimated an increase in ETo, ranging from 0.239 mm/year in November to 0.636 mm/year in May, at a significance level of α = 0.05 (5%). At the provincial scale, significant upward trends in ETo were observed for the annual, Kharif, and autumn seasons, with Z-values of 2.04, 2.16, and 3.13, respectively, at α = 0.05 and 0.01. It was determined that, on an annual scale in Punjab, ETo sensitivity to climatic parameters followed the following order: Tmax > wind speed (WS) > Tmin > RH. The best-fitted models for Tmax, Tmin, WS, and RH were Gaussian, exponential, and spherical. ETo was found to increase spatially from North to South Punjab, with an approximate rise of 70–80 mm/decade. The results provide a scientific basis for understanding hydroclimatic drivers of ETo in semi-arid regions and contribute to improving climate impact assessments on agricultural water use. The observed ETo increases, particularly in South Punjab and lower Central Punjab, highlight the need for region-specific irrigation scheduling and water allocation. These findings can guide cropping calendars, improve irrigation efficiency, and increase canal water supplies to high-ETo areas, supporting adaptive strategies against climate variability in Punjab.

1. Introduction

Water scarcity is a globally challenging issue, especially for countries having arid to semi-arid climatic features. By adopting innovative strategies of water management, each country utilizes water in an efficient way, as the population is increasing significantly. The enhancement of water management systems and quantitative irrigation scheduling are very important aspects of solving the problem of scarcity. Climate change is also playing a vital role in issues of water, agriculture, and food production [1,2]. Therefore, it is very necessary to study climate change’s impact on agriculture, which is the major consumer of water. Reference evapotranspiration (ETo) is the key element of the hydrologic cycle to evaluate the degree of climates’ wetness or dryness and for the estimation of the potential of both crop production and crop water needs [3]. For monitoring crop growth and managing agriculture, ETo is the key index [4]. The physiological characteristics of crops are impacted by climate change due to the impact of Eto, which ultimately leads to an impact on agricultural production [5,6]. The concept of the spatial and temporal evolution of ETo is the preliminary step to calculate reference evapotranspiration and irrigation scheduling, and it has necessary implications for irrigation water use and to assess the crop water stress in agroecosystems [1,7].

The Energy Balance Bowen Ratio, weighing lysimeters, Eddy Covariance, and scintillometry systems are used as conventional techniques by employing point measurements, having the operational potential to provide reasonable values of ETo over homogenous surfaces [8]. Owing to intense atmospheric transport processes, dynamic land surface heterogeneity, and sparseness of data points, these methods have scientific limitations in providing spatial distributions over landscape or region points [9,10]. In order to address this problem, ETo measurement using spatial techniques has gained popularity by continuously retrieving land surface parameters. From a global perspective, it is observed that the whole change trend of ETo has decreased in recent decades, while an increasing ETo has been reported in some areas since the 1980s [3,11,12], i.e., Iran and some Mediterranean countries [13,14]. The changes in ETo have attracted a large number of scholars to explore their causes. However, different researchers have different explanations, as there are differences in climatic conditions and geographical locations.

Much research has been carried out in the past to investigate the global trends in ETo and climatic variables, as well as Eto’s sensitivity to climatic variables. For instance; refs. [7,15] reported that the relative humidity (RH) is the most sensitive climatic variable to reference evapotranspiration, followed by sunshine’s duration (SD), maximum air temperature (Tmax), minimum air temperature (Tmin), and wind speed at 2 m height (WS). Refs. [7,16] found that WS and SD are the important factors affecting Eto, and they are recognized as such by most scholars. Refs. [17,18] revealed that Tmax has played a crucial role in determining the ETo. Furthermore, the pattern of evapotranspiration is not clear yet, but climate change is almost unanimously acknowledged. ETo trends might be rising or dropping, depending on the area and environmental circumstances [18,19]. Assessing the spatial variability in climatic parameters and reference evapotranspiration (ETo) is critically important for the irrigated agriculture of Pakistan. However, research on this subject remains scarce, highlighting the urgent need to quantify these variations and evaluate their local influence on ETo. This study presents the first multi-decadal spatiotemporal analysis of ETo and its climatic drivers across Punjab, integrating advanced trend detection, sensitivity assessment, and optimized semi-variogram-based spatial interpolation. Therefore, the objectives of this study were to statistically analyze the temporal trends of climatic factors and ETo in Punjab, Pakistan, using the Mann–Kendall (M-K) test and Sen’s slope estimator; to identify the most sensitive climatic driver influencing ETo; and to investigate the spatiotemporal patterns of climatic variables and ETo across the region.

2. Materials and Methods

The current research was conducted in Punjab, Pakistan, which has 36 districts and is located at 31.1704° N and 72.7097° E with a land slope from NE to SW, having an area of around 205,344 km² (Figure 1). A lush plain with rivers and irrigation canals makes up the majority of Punjab province, and it also covers the province’s southernmost deserts, such as Thal and Cholistan. The Indus River and its tributaries flow through Punjab from north to south. A large portion of the province is covered in canals, and the terrain is noticeably irrigated. Punjab comprises 36 districts, from which ten random locations were selected in each district to ensure a comprehensive spatial representation. These districts were subsequently grouped into three main regions such as North Punjab (NP), Central Punjab (CP), and South Punjab (SP) as illustrated in Figure 2. The mean monthly data of climatic parameters such as temperature (Tmax, Tmin), relative humidity, wind speed, and sunshine hours (SSHs) were downloaded from NASA Langley Research Center’s Prediction of Worldwide Energy Resources (POWER) from the year 1981 to 2020 (40 years) for research. In the spatial maps, the year 1981 was considered as the baseline, and decadal average values (10-year means) of each climatic parameter and reference evapotranspiration (ETo) were calculated and presented to assess spatiotemporal variations over the study period.

The research methodology focuses on evaluating ETo (evapotranspiration) trends through temporal and spatial analyses. Temporal analysis involves detecting serial correlation, applying pre-whitening techniques, and using the Mann–Kendall (M-K) test alongside Sen’s slope estimator to identify trends. Spatial analysis utilizes the semi-variogram model (GS+), followed by interpolation through Arc-GIS to produce spatial images. Both analyses contribute to the sensitivity analysis, with results discussed and summarized in the conclusions (Figure 3).

2.1. Estimation of Reference Evapotranspiration (ETo)

In this work, reference evapotranspiration (ETo) was calculated using CROPWAT 8.0 software, which requires climatic inputs including maximum temperature (Tmax), minimum temperature (Tmin), relative humidity (RH), wind speed (WS), and sunshine hours (SSHs) for application of the FAO-56 Penman–Monteith method. The Penman–Monteith approach is widely recognized as the standard for ETo estimation, because it combines the principles of energy balance and aerodynamic transport to account for both radiative and turbulent heat fluxes [20]. Meteorological data were utilized to estimate the monthly ETo, which was subsequently employed for spatiotemporal analyses of ETo and its key climatic drivers on a monthly, seasonal, crop-seasonal (Rabi and Kharif), and annual basis for the entire Punjab province. The four meteorological seasons are winter (December–February), spring (March–May), summer (June–September), and autumn (October–November). For agricultural assessment, the two principal crop seasons are Rabi, extending from October to March (sowing in October–November, harvesting in March–April), dominated by wheat, barley, and oilseeds, and Kharif, spanning April to September (sowing in April–June, harvesting in October–November), with major crops including rice, maize, sugarcane, and cotton.

2.2. Statistical Approaches for Trend Assessment

The Mann–Kendall and Sen’s slope estimator tests were employed to identify any significant trends in climatic variables and ETo and their rate of change in the time series data under consideration.

2.2.1. Serial Correlation Effect

Serial reliance is one of the challenges in identifying and understanding patterns in climate data. If the data exhibit a positive serial correlation (persistence), the test results will frequently reveal a noteworthy tendency that is random. In this study, the pre-whitening strategy was used to remove the element of the serial correlation structure from the time series data before applying any analysis. Many scientists have successfully used this concept [16,17,19,21,22,23]. The sample data xi’s lag-1 serial correlation coefficient can be calculated by Equations (1) and (2) [24,25].

$$r_1={\displaystyle \frac{\frac{1}{n-1} \sum _{i=1}^{n-1}\left(x_i-E\left(x_i\right)\right).(x_{i+1}-E\left(x_i\right))}{\frac{1}{n}\sum _{i=1}^n{(x_i-E\left(x_i\right))}^2}}$$

(1)

$$E \left(x_i\right)={\displaystyle \frac{1}{n}} \sum _{i=1}^n{x}_i$$

(2)

where (x_i) is an average of sample data, and “n” is the sample size. The algorithms were used to look for potential statistically significant trends in sample data (x₁, x₂, x₃,…, x_n). The determination of serial correlation (r₁) requires Equations (1) and (2), which have been proposed by several researchers, including [13,21,22]. If the estimated r₁ is not significant at the 5% significance level, the Mann–Kendall test and Sen’s slope estimator may be used to analyze the time series initial values. If found significant, the “pre-whitened” time series may be applied before trend analysis.

Whether the test is one-tailed or two-tailed affects the crucial value of r₁ for a specific significance level. The value of r₁ > 0 for an alternative hypothesis of the one-tailed test, whereas the alternative hypothesis for the two-tailed test is that r₁ is not 0, without any indication of whether this difference is +ve or −ve. One can calculate the probability bounds of an independent series for r₁ by Equation (3).

$$r_1=\left\{\begin{array}{c}{\displaystyle \frac{-1+1.645\sqrt{n-2}}{n-1}}, one-tailedtest\\ {\displaystyle \frac{-1\pm 1.645\sqrt{n-2}}{n-1}}, two-tailedtest\end{array}\right.$$

(3)

2.2.2. Mann–Kendall (MK) Test

The World Meteorological Organization (WMO) suggested the non-parametric Mann–Kendall (MK) test to determine the consequence of the change trend in hydro-meteorological time series. The MK test is distribution-free, so sample points need not follow a specific distribution, and it has been widely used for trend analysis [19,26,27,28]. Compared to many common methods, it is less sensitive to outliers, more effective at detecting trends, and unaffected by the actual distribution of time series data [22]. According to the null hypothesis Ho, a sample of data {x_i, i = 1, 2,…, n}, x_i is independent and identically distributed. The other hypothesis H1 states that X exhibits a monotonic trend. Equations (4)–(7) were used to determine test statistics, test statistic variance, and standardized test statistics, respectively.

$$S={\sum }_{i=1}^{n-1}{\sum }_{j=i+1}^{n}sgn (x_j-x_i)$$

(4)

where (x₁, x₂, … x_n) showcase “n” data points, where x_j shows the data point at time j. An inflated +ve value of S means a climbing trend, and a stunted −ve value indicates a downward trend.

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

(5)

This means that an upward value of S denotes an increasing trend in the data series, whereas a downward value shows a decreasing trend. Assuming an independent and identical data distribution, the S statistic’s variance is as follows (Equation (6)).

$$VAR (S)={\displaystyle \frac{n(n-1) (2n+5)-{\sum }_{i=1}^m{t}_i(t_i-1\left)\right(2t_i+5)}{18}}$$

(6)

Here, “n” denotes the total number of observations, “m” the total number of ties among groups, and “t_i” the number of ties of extent i. A tied group consists of data points with identical values. For samples where n > 10, Equation (7) is used to calculate the standard normal test statistic (Z_s).

$$Z_s=\left\{\begin{array}{c}{\displaystyle \frac{s-1}{\sqrt{VAR \left(S\right)}}}, if S>0\\ 0, if S=0 \\ {\displaystyle \frac{s+1}{\sqrt{VAR \left(S\right)}}},if S<0 \end{array}\right.$$

(7)

where VAR is the variance of S, Z the standardized test statistic, and n the sample size. A positive Z_s indicates an upward trend, while a negative Z_s indicates a downward trend. At significance levels α = 0.05 and α = 0.01 (critical values 1.96 and 2.56), a significant trend exists if |Z_s| > Z_1−α/2, rejecting the null hypothesis of no trend.

2.2.3. Sen’s Slope Estimator

The non-parametric method evaluates the data’s trend’s slope. Sen designed Sen’s slope estimator [29], and it is calculated by Equation (8).

$$\beta =Median \left({\displaystyle \frac{x_j-x_i}{x_j-x_i}}\right), \forall j>i$$

(8)

where x_j and x_i are data values at times j and i; the trend is reportedly escalating if β is positive, and it is reportedly declining if β is negative. Sen’s estimator, which is often used by investigators to estimate the trend line’s gradient in a hydrological data set, delivers a credible calculation of a trend’s degree [16,21,23,27]. The slope gives the magnitude of change per selected time span (i.e., per day, month, or year).

2.3. Sensitivity Analysis

The responsiveness of climatic factors to reference evapotranspiration is statistically examined using this approach, which was developed with the feature of a degree in variability. For the improved Penman–Monteith technique including several explanatory varying indicators of diverse dimensions and ranges, the sensitivity coefficient is influenced by the relative value of ETo and x_j. The sensitivity coefficient has been extensively used to calculate the sensitivity in evapotranspiration studies [15,16,18,30,31]. The corresponding sensitivity coefficient (S) was calculated by Equation (9).

$$S_{xi}=\underset{∆\mathit{xi}\to \mathit{\infty }}{\lim }\left({\displaystyle \frac{\frac{∆{\mathrm{E}\mathrm{T}}_{\mathrm{o}}}{{\mathrm{E}\mathrm{T}}_{\mathrm{o}}}}{\raisebox{1ex}{${∆x}_i$}\!\left/ \!\raisebox{-1ex}{$x_j$}\right.}}\right)={\displaystyle \frac{\partial {\mathrm{E}\mathrm{T}}_{\mathrm{o}}}{\partial x}}·{\displaystyle \frac{\left|x\right|}{{\mathrm{E}\mathrm{T}}_{\mathrm{o}}}}$$

(9)

Skipping the higher order, a first-order Taylor series approach was employed to compute S using Equation (10):

$$S= {\displaystyle \frac{{∆\mathrm{E}\mathrm{T}}_{\mathrm{o}}}{∆x}}\times {\displaystyle \frac{\left|x\right|}{{\mathrm{E}\mathrm{T}}_{\mathrm{o}}}}$$

(10)

where ΔETo is the corresponding variability in reference evapotranspiration induced by Δx, and Δx is the proportionate alteration in the model input value of x. The coefficient of S describes the alterations in ETo urge that arise when a climatic variable’s value varies. The S represents variation in ETo urge by altering the climatic parameter x value. For example, if the sensitivity coefficient is 2, it means a 10% increase in climatic variable (x) would be a reason for a 20% increment in ETo while other climatic factors are constant. The standard table for sensitivity coefficient values, along with sensitivity levels, is shown in

Table 1. Standard values of sensitivity levels.

Sensitivity Coefficient	Sensitivity Level
0.00 ≤ |Svi| < 0.05	Negligible
0.05 ≤ |Svi| < 0.20	Medium
0.20 ≤ |Svi| < 1.00	High
|Svi| ≥ 1.00	Very high

.

The consistency or contrary nature of ETo with input factors can be evaluated with positive and negative values of sensitivity; the value of S is directly proportional to the impact of meteorological variables on ETo. If S > 0, it means reference ETo will rise as the climatic variable rises; if S < 0, ETo will decrease as the variable increases. S is a non–dimensional variable that can be employed in sorting the influence of each variable on ETo. This study kept all the other parameters constant, except the one under investigation, at a time while performing sensitivity analysis for monthly and annual average ETo in each year for the area under investigation from −20 to +20% at a gap of ±5% (eight situations).

2.4. Geo-Statistics and Semi-Variogram Analysis

Meteorological data and ETo were interpolated using deterministic and geo-statistical approaches for 360 sites in Punjab. Kriging, a geo-statistical method that incorporates probability into data projections, is thought to be the most appropriate technique for interpolation of climatic variables [32,33]. Semi-variograms are essential in GIS-based Kriging, because they quantify the spatial autocorrelation of a variable, enabling accurate interpolation by defining how similarity between data points changes with distance, which directly influences the precision and reliability of the generated maps [34,35]. By merging semi-variogram readings at arbitrary lag intervals, a plot of the semi-variogram is constructed. The semi-variogram models (linear, spherical, circular, exponential, and Gaussian) give information about which semi-variogram model is the best fit model for Kriging interpolation and reporting the spatial autocorrelation study of data. For this purpose, Gamma Design Software (GS+, version 10) was utilized for model selection.

Nugget (Co): The point at which the semi-variogram (almost) heads off the ordinate value. Sill (Co + C): A sill is the variable (the value on the ordinate) at which the semi-variogram model achieves the range. Range: In a semi-variogram, the lag distance at which the semi-variogram model (curve) becomes straight, parallel to the horizontal, is known as range.

3. Results and Discussions

3.1. Serial Correlation and Meteorological Variables

It was found that positive serial correlations were present in South Punjab, North Punjab, Central Punjab, and Punjab as maximums of 0.48, 0.63, 0.46 and 0.48, while 0.29, 0.34, 0.31 and 0.29 were the minimum serial correlation coefficients, respectively. The strongest and the weakest serial correlations were found in the wind speed (North Punjab) and Tmin (Central Punjab) data series, respectively. Tmin showed the weakest serial correlation in Central Punjab (autumn) series. Tmax showed the highest serial correlation in South Punjab (autumn) series. Wind speed and relative humidity showed the weakest serial correlation in summer (South Punjab), while the strongest serial correlation was in Kharif (North Punjab).

The strongest serial correlation for reference evapotranspiration was found in the Kharif season. The negative serial correlations were not significant, while the majority of the study area had significant positive serial correlations for all the variables and ETo that was removed by the pre-whitening approach. The trend results before and after the pre-whitening procedure were different; the positive serial correlations were causing an overestimation of significant trends, and the negative serial correlations were giving an underestimation of significant trends, and the predicted results were not so accurate and reliable. The seasonal and annual serial correlation coefficients of all climatic variables and ETo are shown in

Table 2. Lag-1 serial correlation coefficients for ETo and climatic variables in seasonal and annual time data series.

	ETo				Tmin				Tmax				WS				RH
Period	SP	NP	CP	P	SP	NP	CP	P	SP	NP	CP	P	SP	NP	CP	P	SP	NP	CP	P
Kharif	0.48	0.63	0.45	0.48	0.26	0.23	0.20	0.24	0.02	0.48	0.28	0.30	0.49	0.66	0.58	0.59	0.19	0.47	0.24	0.30
Rabi	0.32	0.37	0.23	0.33	0.23	0.33	0.28	0.27	0.36	0.42	0.36	0.38	0.54	0.36	0.23	0.40	0.11	0.34	0.21	0.22
Winter	0.19	0.21	0.09	0.19	0.16	0.25	0.23	0.22	0.21	0.18	0.16	0.19	0.53	−0.01	0.05	0.14	0.20	0.26	0.04	0.19
Spring	0.25	0.22	0.06	0.25	0.13	0.13	0.16	0.15	0.19	0.27	0.17	0.20	0.59	0.30	0.18	0.36	0.33	0.40	0.30	0.34
Summer	0.29	0.61	0.33	0.30	0.03	0.08	−0.04	−0.01	−0.05	0.37	0.18	0.19	0.29	0.56	0.50	0.48	0.05	0.29	0.06	0.10
Autumn	0.48	0.59	0.35	0.48	0.35	0.34	0.32	0.35	0.51	0.58	0.36	0.48	0.47	0.58	0.33	0.50	0.32	0.46	0.24	0.33
Annual	0.48	0.53	0.46	0.48	0.45	0.44	0.45	0.45	0.20	0.47	0.40	0.38	0.60	0.61	0.56	0.63	0.27	0.52	0.35	0.39

Notes: SP = South Punjab, NP = North Punjab, CP = Central Punjab, P = Punjab.

.

3.2. Seasonal and Annual Trends in Climate Variables

Statistical test results for seasonal and annual trends in climatic variables between 1981 and 2020 on the Punjab scale are discussed here. All of the significant trends for Tmax at the 1, 5, and 10% significance levels were increasing. On the annual time scale, the significant increasing trends were detected at the 5% significance level in SP, with an increase of 0.02 °C/year. On the seasonal time scale, at a significance level of 1, 5, and 10%, trends were detected in SP, NP, CP, and P for Kharif, winter, spring, and autumn, respectively. Tmax varied between 0.72 to 0.44 °C/year at a 10% significance level in spring (

Table 3. Annual and seasonal Z and β values of MK test for ETo and climatic variables.

Period		ANNUAL	KHARIF	RABI	WINTER	SPRING	SUMMER	AUTUMN
P	Z	2.039 *	2.155 *	1.43	0.24	1.20	1.53	3.134 **
	β	2.23	1.56	0.80	0.05	0.68	0.72	1.09
SP	Z	1.27	1.41	1.18	0.64	2.272 *	2.668 **	1.46
ETo	β	1.68	1.02	0.62	0.15	1.18	1.49	0.63
NP	Z	1.32	0.87	1.08	0.83	1.946 +	0.29	1.666 +
	β	1.58	0.57	0.64	0.17	0.72	0.09	0.44
CP	Z	1.29	1.04	2.16	0.76	1.36	0.92	2.02
	β	1.64	0.66	1.01	0.17	0.59	0.40	0.58
P	Z	1.53	0.94	1.62	2.319 *	2.621 **	1.62	1.32
	β	0.02	0.01	0.02	0.03	0.05	0.02	0.02
SP	Z	2.249 *	2.225 *	1.57	2.155 *	2.924 **	1.25	1.48
Tmax	β	0.02	0.02	0.02	0.03	0.05	0.01	0.02
NP	Z	1.32	0.87	1.08	0.83	1.946 +	0.29	1.666 +
	β	1.58	0.57	0.64	0.17	0.72	0.09	0.44
CP	Z	1.29	2.272 *	1.62	2.249 *	2.48 *	1.48	1.46
	β	0.02	0.03	0.02	0.03	0.06	0.03	0.02
P	Z	1.736 +	0.45	0.59	−0.01	0.57	0.06	−1.34
	β	0.013	0.005	0.006	0.000	0.010	0.001	−0.017
SP	Z	1.736 +	0.50	0.83	0.13	0.97	0.48	−1.34
Tmin	β	0.013	0.006	0.008	0.003	0.016	0.004	−0.017
NP	Z	1.34	0.19	−0.13	−0.03	0.92	0.22	−1.53
	β	0.012	0.003	−0.002	−0.001	0.012	0.001	−0.018
CP	Z	1.783 +	0.06	0.31	−0.06	0.59	0.13	−1.67
	β	0.013	0.001	0.003	−0.001	0.008	0.001	−0.019
P	Z	0.31	0.17	0.85	1.36	0.34	0.06	0.52
	β	0.000	0.000	0.001	0.002	0.001	0.000	0.001
SP	Z	0.29	0.36	0.90	0.76	0.64	0.22	1.01
WS	β	0.001	0.001	0.001	0.001	0.001	0.001	0.002
NP	Z	0.71	0.13	1.11	0.91	−0.01	0.29	0.59
	β	0.001	0.000	0.001	0.001	0.000	0.001	0.001
CP	Z	0.10	−0.15	1.48	1.22	0.83	0.00	0.22
	β	0.000	0.000	0.001	0.001	0.001	0.000	0.000
P	Z	−0.94	−0.92	−0.36	3.110 **	−1.29	−0.78	−1.83
	β	−0.06	−0.05	−0.02	0.20	−0.11	−0.08	−0.18
SP	Z	−1.60	−1.89	0.45	3.460 ***	−1.50	−0.97	−1.85 +
RH	β	−0.07	−0.08	0.02	0.21	−0.12	−0.04	−0.15
NP	Z	−0.66	−0.76	−0.36	2.971 **	−1.48	−0.90	−2.06 *
	β	−0.05	−0.06	−0.03	0.23	−0.13	−0.07	−0.19
CP	Z	−0.92	−1.71 +	0.01	2.528 *	−1.18	−1.11	−2.44 *
	β	−0.06	−0.10	0.01	0.17	−0.12	−0.08	−0.26

Notes: Z: Mann–Kendall test statistic, β: Sen’s slope estimator, SP: South Punjab, NP: North Punjab, CP: Central Punjab, P: Punjab. *** Statically significant trends at the 0.1% significance level. ** Statically significant trends at the 1% significance level. * Statically significant trends at the 5% significance level. + Statically significant trends at the 10% significance level.

). On annual and seasonal time scales, wind speed showed no significant trends, while positive and negative trends were detected for RH in Punjab. For RH negative trends at a 5% significance level were found in autumn for Punjab, South Punjab, North Punjab, and Central Punjab, with a decrease of −0.18, −0.15, −0.19, and −0.26%, respectively. In Kharif, decreases of −0.08 and −0.10% in RH for South Punjab and Central Punjab were also observed (Table 3). Positive trends were detected for Tmin at an annual time scale in Punjab, South Punjab, and Central Punjab, while negative trends appeared in Central Punjab for autumn, and all trends of Tmin were at a 10% significance level (Table 3).

Increasing trends results of Tmax were in line with the results of [36], who reported an increase in Tmax in winter; increasing trends have been observed across seven sub-regions of Asia; ref. [37] found a significant increase in temperature in all climatic regions of Pakistan during summer. Positive trends have been found by [38] in the middle and lower Indus basin; ref. [39] noted an increase in Tmax and Tmin at seasonal scales and the annual resolution. The annual increasing trends findings compared well with trends reported at the lower Indus Basin of Pakistan by [38] with different slopes; ref. [40] reported an increase of +0.15 and +0.02 °C/decade in annual Tmax and Tmin.

This increase can be related to several factors such as increased concentrations of GHGs, emission of aerosols, increased cloud cover, and urbanization, as a result causing climate variation. The decreasing trends results of RH were consistent with the findings of research conducted in Malaysia by [40], who found that RH decreased by −0.069 to −0.208%; ref. [19] found downward trends for relative humidity in arid regions of China. Ref. [28] found downward trends in relative humidity in winter in Punjab, Pakistan; ref. [26] demonstrated downward trends in RH for arid and semi-arid regions of Iran. Negative trends in RH at the rate of −0.007%/year were observed in arid and semi-arid regions of China by [15]. Ref. [31] found both non-significant increasing and decreasing trends in relative humidity in the Upper Chanab Canal command area of Pakistan. The decreasing change in RH can be due to various factors, such as urbanization in the province, causing an increase or decrease in surface water or cultivated area and the number of plants [41].

It is particularly worth noting that the results showed a relationship between relative humidity and temperature, as the increase in temperature can cause an increase in the acceptation capacity of steam in the air that causes a decrease in RH. Increasing trends in temperature with decreasing trends in relative humidity were observed, which shows a relationship between both of them; the inverse relationship between RH and temperature was observed by [28], which is in line with current study results.

3.3. Seasonal and Annual Trends in ETo

Temporally averaged ETo showed a general increase trend in all regions of the study area over the past 40 years, as shown in Table 3. Trend analysis indicated that ETo in SP has increased significantly at 1 and 5% significance levels in spring and summer by 1.18 and 1.49 mm/year, respectively. In NP small upward trends were detected at α = 0.05 and 0.10, with Z values of 1.95 and 1.67 in spring and autumn. ETo also obviously increased with multiple trends in Punjab and its regions; NP, CP, and Punjab ETo increased by 0.58, 0.44, and 1.09 mm/year in autumn on significance levels of 10, 5, and 1%. On the Punjab scale, significant upward trends were detected annually and for Kharif and autumn, with Z values of 2.04, 2.16, and 3.13 at α = 0.05, 0.05, and 0.01, respectively.

The results of this study are in line with [27]’s findings of significant increasing trends in ETo in Peninsular Malaysia; ref. [42] found increasing trends in ETo on a seasonal and annual basis in Iran; ref. [43] also found mostly increasing trends in ETo in Solvinia (Europe) with a selected significance level of 0.05. Significant increasing trends in ETo for different time periods in China have been detected by several researchers in arid and semi-arid regions since 1980; the findings of current research are in agreement with the results of [30,44,45,46] who detected a significant increase in ETo up to 3.5 mm/year in Spain. Ref. [47] analyzed ETo in a French Mediterranean region and found increasing trends, along with increasing trends in temperature; ref. [48] found a net change of 0.39 mm/year, along with the results of significant increasing trends in ETo in Madhya Pradesh (India) having the same climatic characteristics as the current study’s area of investigation. Ref. [16] identified more pronounced statistically increasing trends in reference crop evapotranspiration in Iran; ref. [49] observed an increase in ETo by 21.1 mm/decade due to a decrease in RH after 1993 in mainland China.

The findings of the current study may be explained by the offsetting impacts of the significant downward trends in RH and the significant upward trend in T (max and min) on ETo in Punjab. The decrease in RH will directly affect the transpiration of plant stomata; the lower the RH of the atmosphere, the smaller the vapor pressure at the same temperature. It means a high vapor pressure difference between plants; such a high difference will cause more transpiration, as water vapor will be able to escape from stomata easily; high temperatures provide more energy for evapotranspiration, hence an increase in reference evapotranspiration. The main impacts of the increase in ETo can be the reduction in the water balance in Punjab, Pakistan which may be a major reason of high irrigation requirements, hence the increase in crop water requirements for the growth of crops.

3.4. Sensitivity of ETo to Climatic Variables

The influence of meteorological variables on ETo was assessed through sensitivity analysis. This method has been used by many researchers due to its simplicity and practicability [19,50,51]. The values of meteorological variables were varied by ±20% to understand the influence of different variables on ETo. Eight climate change scenarios (Δx = ±20%, ±15%, ±10%, ±5%), where x is the meteorological variable) were used as the input for the FAO PM model. The analysis was performed at the seasonal (six seasons), and annual scales for the study area and its regions. Sensitivity coefficients for each meteorological variable on regional and Punjab scale for the quarter periods are summarized in

Table 4. Summary of sensitivity coefficients of ETo against climatic variables.

Variable	−20%	−15%	−10%	−5%	+5%	+10%	+15%	+20%
Tmax	−0.47	−0.50	−0.53	−0.55	0.60	0.62	0.64	0.66
Tmin	−0.08	−0.09	−0.10	−0.11	0.13	0.14	0.16	0.17
WS	−0.33	−0.34	−0.35	0.35	0.39	0.39	0.40	0.40
RH	0.06	0.07	0.08	0.08	−0.10	−0.10	−0.11	−0.12

.

The sensitivity analysis reveals that reference evapotranspiration (ETo) is most responsive to changes in maximum temperature (Tmax) and wind speed (WS). Specifically, a 20% increase in Tmax raises ETo by a coefficient of 0.66, while a similar increase in WS results in a 0.40 coefficient, indicating that higher temperatures and wind speeds significantly boost ET_o. Minimum temperature (Tmin) has a smaller impact, with a 20% increase leading to a sensitivity coefficient of 0.17. Relative humidity (RH) has an inverse effect: a 20% increase in RH lowers ETo by 0.12, while a decrease in RH slightly raises ETo. These results suggest that a higher Tmax and WS increase ETo, while RH reduces it, and Tmin has a relatively modest influence. The high sensitivity of ETo to Tmax and wind speed, and its inverse relationship with RH, can be linked to both climatic trends and significant land use/land cover (LULC) changes in Punjab over the past four decades [52]. Urban expansion in Central Punjab has intensified urban heat island effects, which raise temperatures and lower humidity. Agricultural intensification and cropland expansion in the south have altered surface albedo, evapotranspiration, and wind regimes [53,54]. Deforestation and reduced vegetative cover have further decreased moisture recycling. These combined factors have amplified Tmax impacts and contributed to the spatial heterogeneity in ETo sensitivity across the province [55,56].

Findings of current studies were in line with several studies of ETo sensitivity globally; ref. [15] witnessed maximum temperature as the second most sensitive variable to ETo followed by wind speed; ref. [49] found Tmax to be the most sensitive climatic variable to ETo followed by relative humidity. Ref. [57] investigated the sensitivities of ETo to wind speed and temperature and found the same order of sensitivity of temperature and wind speed as in the current study; ref. [58] found Tmax to be the second most sensitive climatic parameter to ETo in the Upper Chenab Canal command area of Pakistan. Ref. [59] found temperature (Tmax, Tmin) to be the most important meteorological factor affecting ETo in arid and semi-arid zones of Pakistan; ref. [19] found Tmax the second most influencing parameter for Eto; ref. [48] also represented findings parallel to the current study’s findings, showing Tmax as primarily sensitive to Eto; the current study results are in line with the finding of [50,51].

3.5. Semi-Variogram Analysis

Semi-variogram analyses were conducted in GS+ software version 10 to find the best fit model for Kriging interpolation of average annual data of climatic variables, and the findings of the analysis are presented in

Table 5. Best fit model for interpolation of ETo and climatic variables in GS+.

Variable	Best Model	Nugget	Sill	Proportion	RSS	R2
Tmax	Gaussian	0.38	8.76	0.957	2.66	0.969
Tmin	Gaussian	0.53	3.07	0.827	0.125	0.979
WS	Exponential	0.011	0.21	0.944	0.0002	0.98
RH	Spherical	0.1	51.2	0.98	61.3	0.998

. The analysis showed that the best fit model for the interpolation of Tmax and Tmin data series was the Gaussian model, with R² values of 0.969 and 0.979, respectively, which was evidence of the best fitting model. Wind speed data series fitted best in the exponential model, with an R² value of 0.98, while on the other hand, the relative humidity data set fitted best in the spherical model, with an R² of 0.998. The above-mentioned models were used in ordinary Kriging (OK) for all climatic variables.

Several researchers used ordinary Kriging for climatic variables’ interpolation [60,61]. Exponential kriging was found to outstanding, with ordinary kriging, for climatic variables, and it gave the lowest errors [38]; ordinary kriging was considered the optimal method for interpolation of air temperature [62]. The Kriging method with spherical semi-variogram gave a good performance in meteorological interpolations [63].

3.6. Spatial Variability in Climatic Variables

3.6.1. Spatial Variability in Annual Average Tmax and Tmin

The multi-year average of Tmax in the study area is about 30 °C at the annual time scale. The fluctuation range is from 26 °C to 37 °C. Tmax varied spatially throughout the Punjab. The spatial distribution of average annual maximum temperature across the Punjab province is shown in Figure 4. It shows that Tmax is high in South Punjab in the districts of Layyah, Muzaffargarh, and Multan and gradually decreases toward the surrounding areas (Figure 4). Annual change in maximum temperature shows that the value in the plain areas is higher than that of the northern areas having good vegetation. From NP to SP, Tmax increased gradually, considering the spatial distribution of inter-decade annual Tmax for 40 years, to evaluate the variation in Tmax with time. It can be seen that with time Tmax increased gradually from lower South Punjab to Central Punjab, with areas having an annual average temperature of 33–34 °C in 1981 facing up to 35 °C and above.

Spatial distribution of Tmin in Punjab can be seen in Figure 5, consisting of five maps based on per decade average values of minimum temperature. Tmin increased gradually/decade in all regions of Punjab, especially in South and Central Punjab. It is shown that it increased from Multan and Bahawalpur with time, and Lodhran, Vehari, Khanewal and Bahawalnagar faced high minimum temperature values after 40 years. Bahawalpur, Rahim Yar Khan, Rajanpur, and Muzaffargarh faced the highest minimum temperatures while Attock and Rawalpindi faced low minimum temperatures.

The spatial and temporal analysis of annual average temperature (maximum and minimum) in Punjab shows a clear warming trend over the past four decades, with greater increases in southern and central districts. Ref. [64] found that Tmax rose by about 0.10 °C/year and Tmin by 0.04 °C/year from 1981 to 2018, with similar upward trends reported across Pakistan [65]. Higher values in the arid south and lower ones in the vegetated north reflect the influence of latitude, vegetation cover, and land use [66]. Tmin increases, particularly in Bahawalpur, Rahim Yar Khan, Rajanpur, and Muzaffargarh, indicate a narrowing diurnal temperature range (DTR), a trend also observed elsewhere [67,68]. Reduced DTR and rising temperatures heighten crop evapotranspiration, water demand, and heat stress, potentially lowering yields, especially in heat-sensitive crops [69]. Persistently cooler northern areas highlight the role of vegetation and lower urbanization in moderating warming. These trends align with national climate projections, which forecast continued warming and more extreme temperatures in coming decades [70].

3.6.2. Spatial Variability in Annual Average Wind Speed and RH

Spatial variability in wind speed and relative humidity has an increased pattern for the study time period. South Punjab cities Bahawalpur, Rahim Yar Khan, Rajanpur, and Vehari have faced a high annual avg. relative humidity. In contrast, Central Punjab has faced low wind speed in past decades and high values, as shown in Figure 6 and Figure 7, respectively. Relative humidity has decreased in all parts of Punjab. In South Punjab relative humidity decreased more prominently in the areas of Bahawalpur, Rahim Yar Khan, Rajanpur and Muzaffargarh. The change in the wind speed has been very dramatic, as in 1981 parts of North Punjab were having wind speeds in the range of 1.4–1.6 m/s, and they have started facing wind speeds up to 1.8 m/s in the last decade (2011–2020). Areas in the Attock district have not faced any highlighted changes in wind speed in the study period. Central Punjab also faced an increase in wind speed up to 1.9 m/s. The change was more visible in lower Central Punjab as compared to upper Central Punjab. Areas in upper Central Punjab (Gujrat, Narowal, and Sialkot) faced a decrease in wind speed in the 2011–2020 period.

The long-term spatial assessment of wind speed and relative humidity (RH) across Punjab from 1981 to 2020 demonstrated regional contrasts and temporal changes. A consistent decline in RH was observed throughout the province, with the steepest reductions in southern districts such as Bahawalpur, Rahim Yar Khan, Rajanpur, and Muzaffargarh. Similar moisture decline patterns have been noted in other South Asian regions, driven by rising air temperatures and altered monsoon circulation [28,71]. Wind speed trends were more spatially variable. In the early 1980s, much of Northern Punjab recorded lower wind speeds (1.4–1.6 m/s), which increased to around 1.8 m/s in the 2011–2020 period, while Central Punjab reached up to 1.9 m/s. These upward trends correspond with findings from other parts of Pakistan, where strengthening pressure gradients have enhanced wind flow [72]. Conversely, upper central districts including Gujrat, Narowal, and Sialkot experienced a reduction in wind speeds in recent years, a shift that may be linked to localized land cover modifications and vegetation changes [73]. Such concurrent decreases in RH and variable wind speed patterns are likely to influence evapotranspiration, alter soil moisture retention, and increase irrigation water demand, thereby affecting agricultural productivity and resource management strategies in Punjab’s agro-climatic zones [74].

3.7. Spatial Variability in ETo

Spatial variability in ETo showed a wide range in the study area for a selected period, with a maximum value of 2000 mm/year to a minimum of less than 1300 mm/year, as shown in Figure 8. In 1981, ETo was in the range of 1300–1450 mm in North Punjab, while it was up to 1600–1750 mm in 2020. In Central Punjab an increase in ETo can be observed spatially, with a maximum value of 1850 mm in Jhang, Faisalabad, Okara, and adjacent districts in the 2011–2020 period. It was an increase of up to 9%. In Central Punjab, ETo increased spatially, specifically in districts adjacent to South Punjab. Reference evapotranspiration in South Punjab increased dramatically, and areas having an ETo of 1600–1700 mm in 1981 faced greater evapotranspiration losses in 2011–2020 by up to 12%. Overall, in Punjab ETo increased in all districts from north to south.

The progressive rise in ETo across Punjab reflects broader regional and global patterns, where climate warming, coupled with changing atmospheric circulation, has intensified evaporative demand. Such increases are often driven by synergistic effects of higher mean and maximum temperatures, reduced relative humidity, and altered wind regimes [75,76]. In irrigated agroecosystems like Punjab, elevated ETo directly translates into higher crop water requirements, amplifying irrigation demand and potentially stressing groundwater and canal systems already operating at or beyond sustainable limits [77]. This trend aligns with projections under CMIP6 scenarios indicating that South Asian semi-arid regions will face 8–15% increases in ETo by mid-century if warming continues unchecked [78]. Moreover, intensified evapotranspiration may exacerbate soil salinization risks in lower Indus Basin zones, where increased water use could lead to greater salt accumulation through capillary rise [79]. From a management perspective, the spatial heterogeneity in ETo trends emphasizes the need for localized irrigation scheduling, adoption of climate-resilient crop varieties, and integration of deficit irrigation practices to optimize water productivity under growing climatic stress.

4. Conclusions

ETo showed significant increasing trends in Punjab, with Z-statistics of 2.04, and non-significant increasing trends were shown by Tmax and Tmin, with Z-statistics of 0.02 and 0.013, respectively. Wind speed showed no significant trends, while a non-significant decreasing trend was observed in relative humidity, representing a Z value of 0.06. Only relative humidity showed negative trends at a monthly, seasonal, and annual time scale with Z values of −0.18, −0.15, −0.19, and −0.26. At an annual scale, Tmax showed a high number of trends, while Tmin and wind speed showed no trends; on the other hand, relative humidity showed positive and negative trends. Sensitivity of Tmax towards ETo was in a range of 0.68 to −0.58, and an increase in Tmax induced an increase in ETo and vice versa. Wind speed showed a wide range of coefficients from −0.35 to 0.57 and had a great impact on Eto, while relative humidity showed an inverse relation with Eto; an increase in relative humidity caused a decrease in ETo. On the Punjab level, seasonal sensitivity coefficients for Tmax, Tmin, wind speed, and relative humidity with a high influence were 0.69, 0.25, 0.55, and 0.13, respectively. It was also found that ETo was sensitive to climatic parameters in the following order for Punjab annually: SCTmax > SCWS > SCTmin > SCRH. The spatial variation in climatic variables revealed an increase of 1.5 to 2.0 °C temperature increase in the last 40 years, and a 10–15% decrease in relative humidity was observed in NP and South Punjab, with a range of 35–50%. An approximate increase of 70–80 mm/decade was observed in ETo in South Punjab. For the study period, ETo has an increased trend in the study area from North Punjab to South Punjab.

This study presents the first multi-decadal spatiotemporal analysis of reference evapotranspiration (ETo) and its climatic drivers across Punjab, Pakistan, integrating advanced trend detection, sensitivity assessment, and optimized semi-variogram-based spatial interpolation. It identifies Tmax as the dominant driver, reveals pronounced north–south gradients and ETo hotspots, and offers a transferable framework for semi-arid irrigated regions facing climate-driven water demand challenges. However, this research work has certain limitations. Satellite-derived data sets may contain interpolation errors and sensor biases, especially in areas with complex microclimates or limited ground validation. The analysis considers only four climatic variables, omitting factors like solar radiation and land use change. Spatial interpolation assumptions may overlook local heterogeneities, and the focus on historical trends without future climate projections limits predictive applicability.

It is recommended to develop water demand management strategies to mitigate the impact of climate variability on reference evapotranspiration and to increase water allocations in areas experiencing higher ETo to meet agricultural water needs. Advanced water techniques such as high-efficiency irrigation systems (HEISs) should be adopted to minimize the impact of climate variation on crop water requirements and to save water. ETo and reference evapotranspiration should be calculated with other methods under various predicted climatic variables to support the verification of the current findings and to mitigate future water demands.