Reference evapotranspiration estimation using reanalysis and WaPOR products in dryland Croplands

Abstract

Accurate estimation of the reference evapotranspiration (ET_o) is crucial for determining crop water requirements. However, the lack of appropriate weather stations representing croplands, particularly in drylands, may adversely influence the accuracy of ET_o estimates. To overcome this issue, a promising approach is to use meteorological stations in cropland areas to collect weather data that are representative of actual conditions. However, the number of agrometeorological stations in these areas is limited. Therefore, this study aims to assess the effectiveness of three datasets, including ERA5 and ERA5-Land, and WaPOR (Water Productivity Open-access portal), for estimating ET_o in cropland areas on a basin scale. The land use/land cover (LULC) of the European Space Agency (ESA) was used to identify the sites resembling agrometeorological stations. Data were collected from 2009 to 2022, and the FAO-Penman-Monteith method was used to estimate daily and monthly ET_o. The accuracy and reliability of ET_o estimates with the three datasets were evaluated by comparing them with ET_o estimated by ground measurements. Statistical analysis metrics, normalized root mean squared error (nRMSE), and relative mean bias error (rMBE) were used to assess the performance of the datasets. This study highlights that ERA5 exhibited superior overall performance compared to other datasets in estimating ET_o. However, WaPOR performed better at high-altitude stations with inhomogeneous topography than ECMWF reanalysis (i.e., ERA5 and ERA5-L). Thus, none of the datasets could provide accurate ET_o estimates for all the stations within the basin. Therefore, applying the best-performing data source yielded better results than using a single dataset. These findings are valuable for improving irrigation scheduling and water management practices on a large scale, particularly in regions facing data scarcity challenges.

Graphical abstract

Highlights

• • ECMWF reanalysis and WaPOR have been used for estimating daily and monthly ET _o .
• • More reliable results were obtained for monthly ET _o .
• • The application of a Land Use/Land Cover map can enhance ET _o modeling.
• • A better understanding of ET _o can be obtained by mapping the best-performing data.

1. Introduction

Water scarcity poses a significant challenge to achieving food security, especially in regions with limited access to freshwater resources [[1], [2], [3]]. Accurately determining crop evapotranspiration (ET_c) is essential because agriculture is the largest consumer of water resources [4]. ET_c is a function of two main variables: the reference evapotranspiration (ET_o) and crop coefficient (K_c) [4,5]. An accurate estimation of ET_o, being a critical parameter, is therefore indispensable [6]. ET_o is defined as the rate of evapotranspiration from a well-watered reference surface, which is typically grass, under standardized meteorological conditions [5].

The Penman-Monteith model is a comprehensive and widely accepted approach for estimating evapotranspiration from a vegetated surface [5,7]. It takes into account various meteorological and surface-related factors, such as temperature, humidity, wind speed, solar radiation, and vegetation characteristics, to estimate the rate at which water is evaporated from the soil and transpired by plants into the atmosphere. After parametrizing this model for a reference surface defined in Ref. [5], the Food and Agriculture Organization of the United Nations (FAO) Penman-Monteith (FPM) was developed [6]. While it is a robust approach it requires a considerable amount of input [8]. Hence, the application of FPM may encounter significant limitations in data-scarce areas or regions with limited historical data records [[9], [10], [11]]. Besides data availability, it is worth noting that weather stations may not consistently provide an accurate representation of the “reference surface”, which can lead to notable uncertainties [[11], [12], [13]]. Therefore, carefully selecting weather stations that realistically reflect local agricultural conditions is of paramount importance. In other words, relying solely on meteorological data from non-cultivated areas, such as urban or airport locations, may not provide accurate ET_o estimates because the microclimate of cultivated areas can differ significantly from that of non-cultivated areas [14].

Some eco-physical mechanisms can significantly affect the evapotranspiration from cultivated areas surrounded by non-cultivated regions in drylands. Large irrigated areas often exhibit lower air temperatures and higher humidity compared to their surrounding areas in drylands [[15], [16], [17]]. This is known as the oasis effect, also referred to as the cooling effect, mitigating the adverse impact of elevated temperatures [18,19]. In general, when estimating ET_o in areas with significant vegetation cover in drylands, it is crucial to take into account the potential cooling effect [12,20,21]. Additionally, the microclimates also affect the evapotranspiration pattern in drylands via the clothesline effect. The clothesline effect is most commonly observed in small, isolated patches of vegetation or along the windward edge of a vegetation canopy surrounded by less rugged terrain. In these areas, evapotranspiration rates are notably high because there are few obstacles to block solar radiation and wind, allowing for a substantial influx of both radiation and sensible heat [22].

Several classical approaches are available for estimating ET_o in data-limited areas, including the use of temperature-based equations [[23], [24], [25]] and geostatistical interpolation techniques [9,26,27]. Temperature-based ET_o estimates can exhibit significant errors in extreme cases and may not offer a realistic physical representation of evapotranspiration [28,29]. The latter approach is heavily reliant on the spatial resolution of weather stations and may yield unreliable estimates in complex terrains [9]. The application of gridded datasets has also been proposed as a robust approach for estimating ET_o in regions experiencing data scarcity [9,12,30]. Gridded datasets, with their consistent spatial coverage, bridge data gaps and effectively portray variations in land cover. This, in turn, facilitates the capture of the aforementioned effects and evapotranspiration variations in drylands [31].

Recently, reanalysis datasets have been applied to estimate ET_o over Iran [9,30]. They have been also employed to estimate ET_o in extreme windy conditions [28]. Based on the findings from these research works, it is evident that ERA5 and ERA5-L can be favorable alternatives for modeling ET_o in data-scarce regions of Iran. Additionally, several other research studies have been conducted in various regions worldwide to assess ET_o models forced by reanalysis and remotely-sensed products [26,[32], [33], [34], [35], [36]]. Nevertheless, most of these studies have not been conducted at meteorological stations, which may have resulted in biased results in drylands, as the findings could be affected by cooling and coastline effects. Furthermore, the majority of studies employing gridded products for estimating ET_o tend to lean on a single dataset. In other words, they often conclude by recommending a single dataset, showing an overall better performance, for areas that may exhibit diverse climatic and geographic characteristics. However, a single dataset may not consistently yield reliable results in regions with different climates and geographic settings [37,38]. Put simply, in such cases, a one-size-fits-all approach is inadequate. Hence, comparative analysis of gridded datasets enables us to leverage the synergistic advantages inherent in various data sources. In essence, amalgamating gridded data, based on their performance, presents a promising approach for estimating ET_o in regions facing data limitation. To fill this research gap, this study aimed to (i) evaluate the performance of European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis data, i.e., ERA5 and ERA5-Land (ERA5-L), as well as the FAO Water Productivity Open-access portal (WaPOR), in estimating daily and monthly ET_o within arid and semi-arid croplands; and (ii) create an integrated map to facilitate the selection of the outperforming gridded datasets tailored to specific surveyed areas.

2. Materials and methods

2.1. Study area and observed data

The Persian Gulf and Oman Sea (PGOS) basin is a large basin located in the west and southwest of Iran, covering an area of approximately 500,000 km² (Fig. 1). The region plays a crucial role in agriculture, encompassing the cultivation of crops such as wheat and barley, as well as supporting livestock grazing activities. The mean precipitation and temperature in the PGOS basin can vary depending on the location within the basin. However, this region generally experiences a hot and arid climate [39]. The mean annual precipitation in the region ranges from less than 100 mm in interior desert areas to approximately 500 mm in mountainous areas [[39], [40], [41]]. The combination of high temperatures and dry spells in the region has led to significant water scarcity issues exacerbated by the increasing demand for water due to agricultural expansion [2,42].

Fig. 1.

Location of the study area and all synoptic stations available in PGOS.

Fig. 1 shows 106 synoptic weather stations in the PGOS. However, 59 stations were excluded from the analysis because they were not located in croplands. The remaining 47 stations were chosen for analysis because they are surrounded by cultivated land (Fig. 2 and Table 1). We employed the European Space Agency(ESA) LULC to determine the proportion of cultivated land for each of the 47 stations. The proportion of cultivated land varied from 50 to 90% across the selected stations. The data from 47 synoptic weather stations located in croplands were retrieved from the Iran Meteorological Organization (IRIMO) during 2009–2022. The dataset included the minimum and maximum temperatures (Tmin and Tmax), dew point temperature (Tdew), wind speed (U), and sunshine hours. The sunshine hours were converted to solar radiation (SR) using the Angstrom equation. The PGOS experiences a range of temperature values, including Tmin (minimum temperature), Tmax (maximum temperature), and Tdew (dew point temperature) between −4.5 °C and 30.8 °C, 6.80 °C and 46.5 °C, and −18.3 °C–24.1 °C, respectively (Fig. 3). The near-surface wind speed also ranges from 0.5 to 3.6 m s⁻¹. Moreover, SR varies within the range of 5.5–26.6 MJ m⁻² d⁻¹.

Fig. 2.

Location of the sites selected for the current investigation.

Table 1.

The name of the surveyed meteorological sites.

Number	Station	Number	Station	Number	Station
1	Tazehabad	17	Hajiabad	33	Minab
2	Sonqor	18	Darreh Shahr	34	Nurabab Mamasani
3	Sararood	19	Dehdasht	35	Nurabad Lorestan
4	Kermanshah	20	Rask	36	Hendijan
5	Harsin	21	Ramhormoz	37	Shahr-e Kord
6	Qasr-e Shirin	22	Roomeshkan	38	Izeh
7	kangavar	23	Rimeleh	39	Dehdez
8	Eslamabad-e-Gharb	24	Sarableh	40	Safiabad
9	Ardal	25	Sanandaj	41	Shushtar
10	Asadabad	26	Farashband	42	Gotvand
11	Aleshtar	27	Fasa	43	Lali
12	Emamzadeh Jafar	28	Qir-Karzin	44	Masjed Soleyman
13	ImanAbad	29	Kazerun	45	Yasuj
14	Eyvan	30	Kuhdasht	46	Sisakht
15	Borazjan	31	Likak	47	Silakhor
16	Bostan	32	Marivan

Fig. 3.

Seasonal boxplots depicting meteorological observations. A) Minimum and maximum temperatures, B) dew point temperature, C) wind speed, and D) solar radiation (The physical unit for Tmax (maximum temperature), Tmin (minimum temperature), and Tdew (dew point temperature) is °C, for solar radiation (SR) is MJ m⁻² d⁻¹, and for wind speed (U is m s⁻¹).

2.2. The ET_o modeling

The FPM was applied to calculate ET_o [5], represented by Equation (1):

$$E{T}_o=\frac{0.408\Delta \left(R_n-G\right)+\gamma \frac{900}{T_{mean}+273}U\left(e_s-e_a\right)}{\Delta +\gamma \left(1+0.34U\right)}$$ (1)

where ET_o: reference evapotranspiration (mm.day⁻¹), Δ: slope of the vapor pressure curve with respect to temperature (kPa.°C⁻¹), R_n: net radiation (MJ.m⁻². d⁻¹), G: soil heat flux density (MJ.m⁻². d⁻¹), γ: psychometric constant (kPa.°C⁻¹), T_mean: mean daily air temperature (°C), e_s: saturation vapor pressure (mbar), e_a: actual vapor pressure (mbar), U: wind speed at 2 m above sea level (m.s⁻¹).

2.3. The gridded data sources

The WaPOR version 2 uses the ETLook algorithm to estimate evapotranspiration, which incorporates modifications to FPM equation to effectively leverage satellite data [43]. The WaPOR applies the Modern-Era Retrospective analysis for Research and Applications (MERRA) for estimating ET_o before February 21, 2014, and subsequently switched to the Goddard Earth Observing System (GEOS-5) after that date [44,45]. Gridded ET_o estimates are freely available on the FAO website at a spatial resolution of 20 km (https://wapor.apps.fao.org/home/WAPOR_2/1).

ERA5 reanalysis, short for the Fifth Generation of the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis, is used as part of the Copernicus climate change service in Europe https://www.ecmwf.int/. It is based on a unified prediction system that uses physical laws to combine the modeled outputs with global observations. The horizontal resolution of ERA5 is 0.25°, corresponding to approximately 31 km, and extending from 1979 to near real-time [46]. ERA5 offers hourly, 3-hourly, and monthly data, enabling users to examine weather and climate phenomena on various time scales, from short-term weather events to long-term climate trends.

ERA5-L is the land component of ERA5 that specifically focuses on land surface conditions and processes [47]. Unlike ERA5, which assimilates direct observations from various sources, ERA5-L does not use direct observations. Instead, it relies on a set of forcing data from the ERA5 reanalysis to model and provide hydroclimatological variables [47]. Then, the forcings were interpolated, providing data at a horizontal resolution of approximately 0.1° (≈9 km). The ERA5-L and ERA5 forcings included temperature (K), 10 m u- and v-components of wind speed (m s⁻¹), surface solar radiation downwards (J m⁻²), and dew point temperature (K). The gridded products were obtained from Google Earth Engine (GEE).

2.4. Statistical evaluation

The accuracy of the ET_o estimates obtained from the three datasets was evaluated by comparing them with the ET_o calculated using observations, based on the normalized root mean square error (nRMSE) and relative Mean Bias Error (rMBE) indices. These statistics were computed using Equations (2), (3):

$$nRMSE=\frac{100}{\bar{O}}\times (\sqrt{\sum _{i=1}^n\raisebox{1ex}{${\left(P_i-O_i\right)}^2$}\!\left/ \!\raisebox{-1ex}{$n$}\right.}$$ (2)

$$rMBE=\frac{100}{\bar{O}}\times \frac{\sum _{i=1}^n\left(P_i-O_i\right)}{n}$$ (3)

Where $O_i$ represents the observed values of ET_o at different time scales, $P_i$ represents the predicted values using the alternative datasets, $\bar{O}$ denotes the average of observed values, and n stands for the number of pair comparisons. The nRMSE is a metric employed to assess the absolute modeling error. An ideal nRMSE value is typically below 10%, signifying a high level of accuracy. If the nRMSE falls within the range of 10%–20%, it suggests a reasonable accuracy. However, the nRMSE exceeding 30% indicates unreliable modeling [40,48]. A positive rMBE indicates that the model overestimates the variable, while a negative rMBE illustrates underestimation [40]. The magnitude of the rMBE indicates the extent of the overestimation or underestimation. According to Ref. [49], if the absolute value of the rMBE exceeds 25%, it indicates that the model significantly overestimates or underestimates the variable. In other words, a deviation of more than 25% suggests considerable overestimation or underestimation.

After evaluating the performance of the FPM using the alternate data sources, we also presented a map illustrating the best-performing gridded dataset (i.e., the one with the lowest nRMSE) at each station location (Fig. 4). presents the outlined steps taken in the current investigation.

Fig. 4.

The research framework.

3. Results

3.1. Evaluation of WaPOR-estimated ET_o

The WaPOR estimated ET_o reliably (i.e. nRMSE <30%) for around 47% of surveyed sites (Fig. 5). The average daily nRMSE for WaPOR is 35% at the level of basin. The minimum and maximum daily nRMSE were determined for Shahr-e Kord (20%) and Sararood (64%), respectively. On monthly scale, nRMSE was below 30% for around 66% of stations.

Fig. 5.

The nRMSE obtained for WaPOR-estimated ETo. A) Daily scale, and B) Monthly scale (The values are given in Table S1 in supplementary material).

As the time scale increases, the number of stations with an nRMSE of more than 30% decreases by almost half. The accuracy of ET_o estimates can be improved by considering longer time periods, as nRMSE tends to be averaged out. Thus, considering longer time periods can lead to more realistic ET_o results. These findings are in line with the study conducted by Ref. [50], which also suggests that reanalysis on a larger time scale can provide more reliable results. Correct monthly ET_o estimation has potential benefits to guide decision-making for different purposes, including irrigation scheduling for each stage of crop growth and agricultural water resources planning [51]. The trend of nRMSE reveals distinct regional patterns within the basin, categorizing the stations into two regions (Fig. 5). The stations located in the north and northeast consistently display lower nRMSE values (<30%) on daily and monthly scales. However, the stations situated in the south and southwest exhibit generally higher nRMSE values, even at monthly resolution.

It can be concluded that WaPOR demonstrates reliable performance in regions characterized by topographic features exhibiting notable spatial variability, including mountain ranges or elevated topography near coastal areas. The reason behind the successful performance of WaPOR in areas with pronounced topographic variability lies in the predominant influence of wind speed on ET_o in arid and semi-arid regions. Furthermore, the application of the MERRA data in WaPOR contributes to its reliability, as numerous studies have validated the accuracy of MERRA in representing wind speed in regions with diverse topography [52,53].

The rMBE analysis revealed that out of the 47 investigated stations (Fig. 6), ET_o was underestimated in 9 stations and overestimated in 38 stations. Among these, 13 stations had a rMBE exceeding 25%. However, no station exhibited a rMBE value lower than −25%.

Fig. 6.

The rMBE values of the WaPOR-estimated ET_o (The values are given in Table S1 in supplementary material).

Thus, FPM fed by WaPOR tended to overestimate ET_o. This is consistent with the findings of [54]. The ET_o overestimation can lead to an excessive allocation of water resources, with adverse consequences for environmental flow and water availability for local communities.

3.2. Evaluation of ERA5-estimated ET_o

The average daily nRMSE was calculated to be 31% by employing ERA5 forcings (Fig. 7). The FPM estimated ET_o with acceptable accuracy (nRMSE <30%) in approximately 49% of areas. The lowest and largest nRMSE were determined for Shahr-e Kord (17%) and Asadabad (54%), respectively. There was an nRMSE value below 30% for monthly ET_o in around 77% of investigated sites.

Fig. 7.

The nRMSE obtained for ERA5-estimated ET_o. A) Daily scale, and B) Monthly scale (The values are given in Table S2 in supplementary material).

The ET_o was underestimated in 79% of stations (Fig. 8). Thus, contrary to WaPOR, ERA5 tended to underestimate ET_o for most cases. Underestimation of ET_o values at stations including agricultural fields can have significant negative impacts on crop production and water management in drylands. It can lead to inadequate irrigation scheduling and water allocation, which may cause water stress and reduced crop productivity. Overall, ERA5 provided slightly more accurate daily and monthly ET_o estimates relative to WaPOR. The FPM forced by ERA5 products performed better in flat and less complex terrains and coastal areas than the mountainous sites characterized by highly complex terrains.

Fig. 8.

The rMBE values of the ERA5-estimated ET_o (The values are given in Table S2 in supplementary material).

Wind speed variations control the ET_o dynamics in arid and semi-arid areas in Iran [9]. Therefore, errors in wind speed products appear to have a significant impact on the accuracy of ET_o estimated by gridded datasets. The type of cultivation (rain-fed or irrigated) can influence ET_o estimation, emphasizing the need to consider this factor for a more reliable understanding.

3.3. Evaluation of ERA5-L-estimated ET_o

The average nRMSE found for daily ET_o was 35% at the level of basin (Fig. 9). There was an nRMSE of less than 30% for daily ET_o in around 40% of sites using ERA5-L reanalysis. The lowest and highest nRMSE calculated for daily ET_o was for Shahr-e Kord (20%) and Sararood (64%), respectively. On monthly resolution, a reliable monthly ET_o was modeled using ERA5-L in 64% of stations (Fig. 9).

Fig. 9.

The nRMSE obtained for ERA5-L-estimated ETo. A) Daily scale, and B) Monthly scale (The values are given in Table S3 in supplementary material).

The ERA5-L exhibits a higher frequency of stations with nRMSE exceeding 30% compared to the ERA5 but showed a similar pattern (Fig. 9). The ERA5-L dataset offers hourly time resolution, enabling the extraction of minimum temperature (Tmin) and maximum temperature (Tmax).

Fig. 10 shows that the ERA5-L often tends to underestimate, particularly in arid and semi-arid regions where wind and solar radiation are the primary drivers of ET_o. Previous studies have shown that high wind speed variations are a significant source of error in the ECMWF reanalysis [9,28].

Fig. 10.

The rMBE values obtained for the ERA5-L-estimated ET_o (The values are given in Table S3 in supplementary material).

The results indicate that no single dataset alone can provide reliable ET_o estimates. Therefore, offering a map of the best-performing data sources appears to significantly reduce the uncertainty in estimating ET_o on a basin scale. It allows taking advantage of multiple data sources and enhancing the accuracy and reliability of estimates [55]. Fig. 11 shows that ERA5, ERA5-L, and WaPOR outperformed in 23, 8, and 16 sites, respectively. The WaPOR provided more accurate ET_o estimates than ERA5 and ERA5-L in mountainous regions. However, ERA5 outperformed WaPOR in coastal and low-altitude areas. These findings have important implications for applying datasets in various applications, particularly agricultural water management.

Fig. 11.

The top-performing dataset for estimating ET_o.

The superiority of ERA5 and ERA5-L has been demonstrated in Iran [9,28]. However, it has also been found that in some cases, other alternative data sources, such as MERRA2, outperformed ERA5 across Iran [9]. This suggests that, while ERA5 performs well overall in Iran, it may not always be the best-performing alternative data source in all cases. Our findings suggest that the selection of the best-performing data for each specific region can be a more effective methodology for estimating ET_o, rather than relying on a single data source.

4. Discussions

Incorporating LULC map appears to be essential for improving ET_o estimates [[56], [57], [58], [59], [60]]. Studies have indicated discrepancies in ET_o values, ranging from 17% to 25%, based on the type of cultivation due to the cooling effect [61]. Moreover, differences in irrigation practices can also impact the cooling effect, leading to considerable variation in ET_o estimation error. For example, spray or Low Energy Precision Applicator (LEPA) irrigation systems generate a greater cooling effect compared to subsurface irrigation systems, consequently leading to lower canopy temperature [62]. The aerodynamic resistance plays a significant role in estimating ET_o, particularly in arid and semi-arid regions where cropland weather stations are often situated near urban areas [63,64]. The sensible heat generated by the non-cropped area may affect the temperature in croplands. Considering the aerodynamic impact of neighboring pixels on ET_o can lead to more accurate estimations. In PGOS where the arid and semi-arid climates prevail, there are cropland weather stations surrounded by the non-cropped regions. Thus, the incorporation of this factor may improve the ET_o estimates in the PGOS.

The quality, distribution, and representativeness of ground measurements should also be taken into consideration when assessing the accuracy of reanalysis data sources. The spatial distribution of the stations can somewhat affect the validation results [65,66]. For instance, the distribution of the weather stations was denser in the north and northwest parts of the PGOS basin, which might affect the accuracy evaluation. The observed data may be recorded at specific intervals, e.g., every 3 h. If a meteorological station collects data at times that do not align with the time resolution of the reanalysis, it can lead to erroneous results [46,47]. These errors may occur consistently across all the collected data due to the mismatch in temporal intervals. This discrepancy in data collection times can affect the accuracy and reliability of the analysis, potentially introducing biases or inconsistencies in the results.

In our study, we recommend not relying solely on a single alternative dataset but rather using multiple outperforming datasets. In other words, we advocate for the use of the most suitable data for each specific area, taking into account its distinct climatic and geographical characteristics. Synoptic observations are conventionally expected to provide a representative picture of conditions within an approximate 100-km radius around an observation station. In situations where applications are focused on smaller, more localized settings, the area of consideration may be as limited as 10 km or less [67]. Consequently, employing top-performing datasets for individual points can be helpful for estimating ET_o for various purposes within the mentioned ranges. This conclusion holds promise for assisting decision-makers and farmers in making informed decisions about agricultural development in data-scarce regions of Iran.

Selecting an alternative data source can also depend on the characteristics of datasets. Among the three datasets analyzed, the ERA5-L has the finest spatiotemporal resolution compared to the other two datasets. ERA5 and ERA5-L provide global coverage starting from 1950, whereas the WaPOR version 2 dataset only covers the Middle East and North Africa (MENA) region since 2009. These attributes are vital factors to take into account when selecting a dataset for a particular purpose. The ERA5 and WaPOR datasets have different update frequencies. ERA5 and ERA5-L receive updates approximately every 5 days. Furthermore, WaPOR version 2 has a data latency of 2 days. The relatively high update frequency offers distinct advantages for applications that require the use of up-to-date data, such as crop water requirement calculations and irrigation scheduling.

In this study, we compared point-scale records with gridded data from reanalysis and remote sensing, covering a relatively large area. Although this methodology is a valid approach for evaluating gridded data sources [9,26,68], it introduces a potential source of uncertainty, which can be considered as a limitation in this research work. However, the sparse distribution of meteorological stations in Iran poses a significant challenge in obtaining gridded observations for the purpose of evaluation. Moreover, gridded datasets typically represent the average weather conditions at the average elevation of a specific grid cell [69]. However, ET_o can significantly vary with elevation, especially in mountainous regions [70]. In greater detail, ET_o is influenced by elevation-related factors that can lead to notable variations in ET_o values, especially in mountainous regions. Higher elevations typically yield lower ET_o values due to the combined impact of lower temperatures, reduced atmospheric pressure, and potential alterations in solar radiation and relative humidity [70,71]. Therefore, if the elevation of a weather station differs substantially from the average elevation of the corresponding grid cell, the gridded data may not accurately reflect the weather conditions measured at that station [72].

We also considered that the cropland areas resemble, to some extent, the reference surface for estimating ET_o. However, as previously stated, the proportion of cultivated area in each pixel ranged from 50% to 90%. Moreover, the croplands may deviate from the reference surface due to water and or nutrient stresses and variations in crop types; which can be acknowledged as another limitation.

5. Conclusions

This study compared the performance of three different datasets, i.e. WaPOR, ERA5, and ERA5-L, in estimating ET_o using the Penman-Monteith equation at 47 stations located in cropland areas across the PGOS basin. In general, the datasets exhibited higher accuracy for monthly ET_o estimates. The ET_o values computed using WaPOR showed relatively good agreement with those computed using observations for mountainous areas, where the topography exhibits high variability. However, unsatisfactory results were found for coastal areas in the southern basin. In addition, using ERA5 and ERA5-L for ET_o estimation led to acceptable results for most coastal stations in the basin.

Since no single dataset can guarantee accurate ET_o estimates for all stations in the basin, the application of top-performing datasets can be beneficial for decision-makers. By leveraging outperforming datasets, decision-makers can access reliable information for effective agricultural water resources management. It is recommended to enhance the accuracy of ET_o estimation in large-scale applications by establishing a unified archiving system for ET_o products based on various remotely sensed and reanalysis datasets.

Funding

This research was funded by grants from the Iran Soil and Water Research Institute (Projects No. 014-53-10-019-98045-990959 and 134-10-1051-005-98045-010338).

Ethical approval

N/A.

Data availability statement

Data will be made available on request.

CRediT authorship contribution statement

Shadman Veysi: Writing – original draft, Methodology, Conceptualization. Milad Nouri: Writing – review & editing, Validation, Formal analysis. Anahita Jabbari: Visualization, Software.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following is/are the supplementary data to this article.