Evapotranspiration Estimation in the Arab Region: Methodological Advances and Multi-Sensor Integration Framework

Abstract

Evapotranspiration (ET) estimation is crucial for sustainable water resource management in arid and semi-arid regions, particularly in the Arab world, where water scarcity remains a significant challenge. The objectives of this study were to map dominant ET estimation techniques and their geographic distribution, demonstrate fusion-based ET estimation under data-scarce conditions, and examine their alignment with climate change and food security priorities. The study reviewed 1279 ET-related articles indexed in the Web of Science, highlighting methodological trends, regional disparities, and the emergence of data-driven techniques. The results showed that traditional methods—primarily the Penman-Monteith model—dominate nearly 70% of the literature. In contrast, machine learning (ML), remote sensing (RS), and artificial intelligence (AI) collectively account for approximately 30%, with hybrid fusion frameworks appearing in only 2% of studies. ML applications are concentrated in Morocco, Egypt, and Iraq, while 50% of Arab countries lack any ML or AI-based research on energy transition (ET). Complementing the bibliometric analysis, this study demonstrates the practical potential of ML-based ET fusion using Landsat and the FAO Water Productivity (WaPOR) data within Saudi Arabia. A random forest model outperformed traditional averaging, reducing the mean absolute error (MAE) to 215.08 mm/year and the root mean square error (RMSE) to 531.34 mm/year, with a Pearson correlation coefficient of 0.86. The findings advocate for greater support and regional collaboration to advance ET monitoring and integrate ML-based modelling into climate resilience frameworks.

1. Introduction

Arab countries face a convergence of critical challenges that intensify pressure on their already fragile environmental systems. Chronic water scarcity, exacerbated by arid climates, low annual precipitation, and unsustainable water use, is placing increasing strain on natural and agricultural water resources. Simultaneously, rapid population growth and urbanisation are driving up demand for food and fresh water, intensifying threats to food security. Compounding these issues are the accelerating impacts of climate change, which further disrupt rainfall patterns and increase the frequency of droughts, affecting both crop productivity and water availability. ESCWA [1] reported that despite progress, the Arab region continues to face severe water challenges, with only five countries on track to meet Sustainable Development Goals (SDG) 6 by 2030. Freshwater availability has already dropped below the 1000 m³ per capita threshold and is projected to fall below 500 m³ by 2026, placing 19 countries in water scarcity and 13 in absolute scarcity conditions. These interconnected pressures underscore the urgency of enhancing our understanding of evapotranspiration (ET), a central process in the hydrological cycle that links water use, agricultural performance, and climatic variability.

Despite its critical role, a coherent understanding of ET research in Arab countries remains limited. Existing reviews often overlook the region’s unique combination of ecological fragility, technological disparities, and scarcity of geospatial data. This study, therefore, offers a novel contribution by synthesising ET research within the Arab context, mapping the intersection of methodological approaches with national policy priorities in climate adaptation and food security. Reviewing ET studies across the region is not only timely but essential to support data-informed decision-making and long-term sustainability of agriculture and ecosystems [2,3,4]. This study is the first comprehensive synthesis of ET research across Arab countries, combining bibliometric analysis with machine learning-based fusion modelling. It addresses regional disparities, methodological gaps, and policy relevance.

1.1. ET Concepts

Evapotranspiration (ET) is a key hydrological process that describes the combined loss of water from the land surface through evaporation from soil and water bodies, as well as transpiration from plants [5]. It plays a critical role in the water cycle, agricultural sustainability, and climate regulation, influencing crop water demand, irrigation scheduling, and hydrological modelling [6].

A standardised measure within ET studies is reference evapotranspiration (ETo), which represents the evaporation and transpiration from a well-watered, uniform-height grass surface under specific climatic conditions [5]. ETo provides a baseline for assessing actual crop water requirements and optimising irrigation strategies, making it widely applicable in agricultural water management. The introduction of the FAO Penman-Monteith equation revolutionised ETo estimation, offering a robust, physically based method that enhances comparability and accuracy across different climates [7].

Another widely used term in hydrological and agricultural sciences is potential evapotranspiration (ETp), which differs fundamentally from ETo in conceptual development, equations, and practical applications. Yet, the distinction between ETp and ETo remains a subject of ongoing scientific debate. Their differences lie in conceptual development, mathematical equations, and practical applications, influencing water resource management, irrigation planning, and climate adaptation strategies. ETp, originally introduced by Thornthwaite in 1948 [6], quantifies the maximum possible rate of evapotranspiration under unlimited water availability, independent of vegetation type. It represents atmospheric water demand from an idealised, fully saturated surface, such as open water bodies or a dense crop canopy that completely shades the ground [8]. In contrast, ETo was developed as a standardised reference, defined as the evapotranspiration rate from a well-watered grass surface of uniform height, actively growing under specific climatic conditions [5]. This standardisation was essential in establishing consistent water-use estimations across diverse agro-climatic zones, facilitating comparative analyses, and optimising irrigation strategies.

Although both ETp and ETo quantify atmospheric evaporative demand, their surface assumptions and applications differ significantly. ETp provides a theoretical upper limit, unlinked to specific vegetation types, whereas ETo functions as a calibrated baseline, enabling precise estimations of crop water requirements via crop coefficients [6,8,9]. However, interchangeable usage of these terms in scientific literature and applied research has led to inconsistencies, reinforcing the need for conceptual clarity, methodological standardisation, and harmonised classification frameworks [8].

This study critically assesses ET research across Arab countries (Figure 1), incorporating both ETo- and ETp-focused investigations to provide a comprehensive review of methodological frameworks and regional research trends. Given escalating concerns about water scarcity, integrating advanced ET modelling techniques, including remote sensing, machine learning, and hybrid approaches, into climate adaptation and sustainable water governance will be pivotal for optimising irrigation efficiency, improving hydrological modelling, and enhancing agricultural resilience in the region.

1.2. ET Estimation Methods

While direct measurements using lysimeters or eddy covariance systems are considered precise, they are often costly, labour-intensive, and impractical across large or heterogeneous landscapes. As a result, ET is commonly estimated indirectly as the product of reference evapotranspiration (ETo) and a crop coefficient (Kc) (Table 1) [10]. Although the FAO-endorsed Penman-Monteith (PM) equation offers high accuracy, its application is often constrained by its reliance on quality-controlled meteorological data, including solar radiation, temperature, humidity, and wind speed [11,12]. Wind speed alone can contribute up to 70–80% of the variation in ETp across the Middle East and North Africa (MENA) region [2]. Yet, it remains challenging to monitor consistently due to sparse station coverage in many Arab countries.

To address these limitations, researchers have explored both empirical and remote sensing-based approaches. While traditional methods are valued for their simplicity and reliance on accessible inputs, they often lack sensitivity to spatial and temporal variability [6]. Ground-based systems, though reliable at the point scale, struggle to capture large-scale evaporative fluxes [13]. In contrast, remote sensing offers enhanced spatial coverage and scalability. For instance, Singh and Senay [14] evaluated four energy balance models, including Mapping Evapotranspiration at High Resolution with Internalized Calibration (METRIC), Surface Energy Balance Algorithm for Land (SEBAL), Surface Energy Balance System (SEBS), and Operational Simplified Surface Energy Balance (SSEBop) using Landsat imagery in the Midwestern U.S. They found METRIC and SSEBop provided the most accurate ET estimates, emphasizing that each model has its unique strengths and limitations depending on scale and climate conditions.

Scientific evidence suggests that the spatial and temporal resolution of input data plays a decisive role in ET estimation accuracy [15]. Coarse spatial scales may obscure small-scale heterogeneity, such as crop variability or microclimate, while low temporal resolution can miss short-term fluctuations in meteorological drivers. Aggregating inputs such as the Normalised Difference Vegetation Index (NDVI), albedo, or surface temperature may introduce errors in key energy balance terms, especially in rapidly changing or complex environments [16]. Therefore, adapting ET models to account for resolution-sensitive variability is essential for reliable application in diverse climatic settings.

Table 1. Major reference evapotranspiration estimation methods used worldwide.

Method	Data Requirements	Strengths	Limitations	Source
Penman-Monteith	Full meteorological data (T, RH, wind, radiation)	High accuracy, widely validated	Needs comprehensive data	Sun et al. [12]
Hargreaves-Samani	Mainly temperature	Simple, minimal data required	Underestimates ET, needs local calibration	Azzam et al. [10]
Thornthwaite	Temperature	Simple, easy to apply	Less accurate in arid climates	Aschonitis et al. [17]
Remote Sensing (SEBS, etc.)	Satellite data (LST, NDVI, etc.)	Large-scale, spatially continuous, multi-sensor	Requires validation, complex processing	Xue et al. [6]

Table 1. Major reference evapotranspiration estimation methods used worldwide.

Method	Data Requirements	Strengths	Limitations	Source
Penman-Monteith	Full meteorological data (T, RH, wind, radiation)	High accuracy, widely validated	Needs comprehensive data	Sun et al. [12]
Hargreaves-Samani	Mainly temperature	Simple, minimal data required	Underestimates ET, needs local calibration	Azzam et al. [10]
Thornthwaite	Temperature	Simple, easy to apply	Less accurate in arid climates	Aschonitis et al. [17]
Remote Sensing (SEBS, etc.)	Satellite data (LST, NDVI, etc.)	Large-scale, spatially continuous, multi-sensor	Requires validation, complex processing	Xue et al. [6]

Remote sensing offers broad-scale monitoring of evapotranspiration (ET), but it faces challenges related to spatial resolution and data quality. Machine learning (ML) and artificial intelligence (AI) techniques improve ET estimation in data-rich environments by detecting complex patterns within large datasets. However, their effectiveness in data-scarce regions is hindered by calibration demands, data gaps, and high computational costs [18,19].

Fusion techniques that integrate satellite imagery, ground observations, and reanalysis data with AI models offer a powerful solution to spatial and temporal limitations. These hybrid frameworks improve model calibration, generalizability, and real-time water management [20]. For example, Sentinel-2 and Landsat fused with ML algorithms trained on lysimeter and eddy covariance data enhance ET predictions, addressing gaps in conventional models like FAO Penman-Monteith, which struggle in regions with missing meteorological inputs. A robust hybrid ET model can integrate:

• Physics-based Penman-Monteith principles for physical consistency.
• Satellite-derived NDVI, surface temperature, and radiation for spatial precision.
• Machine learning models (e.g., Random Forest, LSTM networks) to estimate missing parameters.

This approach boosts accuracy, scales effectively, and reduces dependence on sparse meteorological stations, improving ET modelling for water resource management. Studies validating the fusion of Sentinel-3 thermal, Sentinel-2 optical, and Landsat data demonstrate improved land surface temperature accuracy (RMSE reduced by 1.5 K) and high-resolution (20 m) daily ET estimates, achieving an ET accuracy of up to 0.84 mm/day across sites in Lebanon, Tunisia, Spain, and Senegal [21].

The objectives of this study were to: (i) map dominant ET estimation techniques and their geographic distribution; (ii) demonstrate fusion-based ET estimation under data-scarce conditions; (iii) examine their alignment with climate change and food security priorities.

2. Methodological Framework

The methodological framework consists of two main components:

2.1. Bibliometric Analysis of ET Studies

A structured literature review was performed using the Web of Science database. The search process began with the keyword evapotranspiration combined with specific Arab country names. The initial dataset was refined using key ET estimation methodologies, including the Penman-Monteith (PM), Hargreaves-Samani (HG-S), Hargreaves (HG), Thornthwaite, remote sensing (RS), machine learning (ML), and artificial intelligence (AI) approaches (Table 1). The dataset was further filtered to identify ET studies that intersect with major regional concerns, such as climate change and food security, as well as the extent of cross-border research collaboration and funding mechanisms. Data were tabulated and analysed within spreadsheet environments to extract frequency distributions, regional patterns, and thematic associations.

2.2. Application of Fusion-Based ET Estimation

To complement the bibliometric findings, this study incorporates a practical implementation of evapotranspiration (ET) estimation through multi-source data fusion, utilising Landsat-8 and the FAO Water Productivity (WaPOR) Open-access Portal datasets of the FAO WaPOR in 2024. The Landsat-8 data were collected in 2018 as annual data, calculated from daily and monthly accumulated images. However, the WaPOR data were obtained annually for the year 2018. This methodological component evaluates the performance and relevance of fusion models under typical data-sparse conditions characteristic of arid and semi-arid regions in the Arab world, emphasising their potential for scalable, high-resolution ET monitoring.

Two evapotranspiration (ET) datasets were pre-processed and harmonized for spatial integration: (i) a SEBAL-based ET product derived from Landsat imagery at 30 m resolution, and (ii) FAO’s WaPOR Level 1 actual evapotranspiration (ETa) dataset at ~250 m resolution, covering the target region (25.126342° to 25.79800° N and 49.313754° to 49.903907° E). The traditional fusion method employed a straightforward arithmetic mean of co-registered pixels from both sources to generate a baseline ET estimate. In contrast, the machine learning (ML) fusion approach utilised a random forest regression model trained to predict Landsat-derived ET as the target variable, with co-located WaPOR ETa and ancillary variables (e.g., NDVI, albedo) as predictors.

A spatially explicit difference map was produced to assess the deviations between the traditional and ML-fused ET outputs. Model performance was evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Pearson’s correlation coefficient (r). All data processing and modelling steps were implemented in R version 4.5.1 (open-source software), complemented by spreadsheet tools for tabular analysis and figure generation.

3. Results and Discussion

3.1. Evapotranspiration Research Distribution Across Arab Countries

Figure 2 presents the distribution of ET studies indexed in the Web of Science across Arab countries, revealing substantial variations in research volume and methodological approaches. A total of 1279 studies highlighted regional disparities, with Morocco, Egypt, Tunisia, and Saudi Arabia leading in publications, reflecting their strong institutional commitment to water resource management and agricultural sustainability. When normalized by population, Morocco produces approximately 0.9 ET-related publications per million people, followed by Tunisia (0.7) and Egypt (0.6). While the figure focuses exclusively on publication volume, broader bibliometric analysis reveals two distinct dimensions of institutional commitment: the research productivity, measured by normalized publication output (e.g., Morocco: 0.9 per million), and the strategic investment, reflected in domestic funding share as Qatar, in parallel, demonstrates a distinct form of institutional commitment by domestically financing 57% of its ET research portfolio, despite comparatively modest agricultural R&D spending. These complementary indicators, though derived from separate datasets, offer a multidimensional view of national engagement in water-related research across the Arab region. Arab countries were grouped into four regional categories by several international and regional organizations: North Africa, the Arabian Peninsula (including Gulf nations and Yemen), the Middle East and North Africa (MENA), and Sub-Saharan Africa (SSA, including Sudan, Djibouti, and Somalia). It is worth acknowledging that some countries—such as Algeria and Sudan—may appear in more than one grouping due to overlapping definitions in the literature. This overlap reflects the diversity of geographic framings adopted by regional studies. Our classification follows a bibliometric logic: studies were grouped based on the regional labels explicitly used by their authors, allowing us to compare thematic and geographic trends in ET-related research. Among these groupings, North Africa dominates with 95 ET-related studies, followed by the Arabian Peninsula (34), SSA (24), and MENA (19). El-Shirbeny et al. [4] conducted the only region-wide ET study across Arab countries, generating classification maps of ETo, ETc, and vegetation cover using MODIS data. While the study offers valuable spatial insights, its reliance on coarse-resolution inputs and lack of fusion or machine learning (ML) integration limits its applicability for high-resolution ET monitoring. In contrast, the present study advances ET modelling by integrating Landsat and FAO WaPOR datasets with ML (random forest), enabling scalable, spatially explicit ET estimation for climate resilience and agricultural planning. The following findings could be summarized:

Despite these contributions, the total number of ET studies conducted across Arab countries (1428) remains significantly low, considering ET’s critical role in water resource management within water-scarce environments. Arab-region ET publications account for <3% of global ET literature indexed in Web of Science, despite comprising 5% of global arid land area. This limitation can be attributed to several structural challenges:

• Data constraints: The lack of long-term, high-quality meteorological and land-use data restricts researchers’ ability to conduct comprehensive ET assessments, hindering model calibration and accuracy. Commonly missing datasets in the Arab region include, for example, long-term rainfall records, radiation, and temperature data. El-Shirbeny et al. [4] employed simulated meteorological datasets to compensate for the spatial and temporal limitations of ground-based weather observations across the Arab region. Aieb et al. [22] identified missing climatic data as a persistent challenge in climatology and proposed a regression-based imputation framework to reconstruct daily rainfall records in Algeria’s Soummam watershed. In Sudan, soil moisture estimates from the Climate Prediction Center deviated by up to 106% from ground measurements, largely due to mismatches in spatial resolution and rainfall variability. Similarly, the Climate Research Unit’s (CRU) potential ET estimates underestimated values by about 7–37%, affecting water balance accuracy [23]. Technical and expertise gaps: ET studies demand advanced hydrological modelling, meteorological precision, and environmental science expertise, fields that remain underdeveloped in many Arab research institutions. Persistent technical and institutional gaps continue to hinder effective drought and ET monitoring across the MENA region. Bouhioui & Loudyi [24] highlight a significant disparity in modern drought monitoring techniques, largely driven by data inadequacies and insufficient early warning systems, despite growing research interest. These limitations reflect broader challenges in hydrometeorological modeling, environmental data integration, and regional capacity for climate resilience planning. El-Shirbeny et al. [4] highlight persistent technical limitations in ET monitoring across the Arab region, including sparse and unevenly distributed ground meteorological data, limited integration of GIS-based modeling frameworks, and reliance on cloud platforms such as Google Earth Engine to compensate for infrastructural constraints. The validation of satellite-derived ET estimates using only 35 ground stations further underscores the region’s restricted capacity for GIS-enabled monitoring and multi-source data fusion. Low research prioritisation: Socio-economic challenges divert national focus away from environmental research, leading to limited investment in hydrology and climate-related studies.

Institutional and policy limitations: While some Arab nations acknowledge the importance of ET research, policy frameworks rarely translate into substantial research grants or institutional support, restricting scientific innovation and the application of ET methodologies. Molle [25] discusses Egypt’s irrigation evolution in detail but references evapotranspiration only twice, without integrating it into core policy frameworks or water balance assessments. Also, although the Arab Strategy for Sustainable Agricultural Development for the period 2020–2030 issued by AOAD does not explicitly mention ET, its emphasis on the imbalance within the water-energy-food nexus, driven by rising demand and climate risks, highlights the need for integrated resource management in the region. ET, as a key indicator of actual water consumption in agriculture, remains critically underrepresented. Incorporating ET-based metrics into regional planning would enhance demand forecasting, irrigation efficiency, and climate resilience across the nexus in the Arab region.

3.2. Common Methods Used in Arab Countries

Figure 3 provides a summary of the main techniques used to estimate evapotranspiration (ET) across Arab countries, highlighting their diverse data requirements and varying suitability across different environmental settings:

• FAO Penman-Monteith Method (PM): Widely regarded as the standard approach. The PM method has been widely adopted in approximately 14% of evapotranspiration (ET) studies conducted in Arab countries, reflecting its robustness and reliability. Its application is particularly prominent in Egypt (18%), Saudi Arabia (15%), and Morocco (11%) (

  Table 2. Proportional use of various evapotranspiration (ET) estimation methods across Arab countries (%), based on Web of Science-indexed studies. PM = Penman-Monteith; HGS = Hargreaves-Samani; HG = Hargreaves; TH = Thornthwaite; RS = Remote Sensing; ML = Machine Learning; AI = Artificial Intelligence.

  Scale	PM	HGS	HG	TH	RS	ML	AI	Other
  Algeria	9	11	7	17	4	12	12	6
  Bahrain	0	0	0	0	0	0	0	0
  Comoros	0	0	0	0	0	0	0	0
  Djibouti	0	0	0	0	0	0	0	1
  Egypt	18	11	12	8	12	19	18	16
  Iraq	10	21	17	0	5	18	18	4
  Jordan	7	0	7	0	6	2	0	6
  Kuwait	3	0	0	0	0	0	0	2
  Lebanon	1	0	0	0	6	5	0	4
  Libya	2	5	5	0	0	0	6	0
  Mauritania	1	0	0	0	1	0	0	0
  Morocco	11	16	14	25	32	25	12	11
  Oman	3	0	0	0	2	2	0	3
  Palestine	1	0	0	0	1	0	0	1
  Qatar	2	5	7	0	0	0	6	1
  Saudi Arabia	15	11	14	8	9	5	18	8
  Somalia	0	0	0	0	1	0	6	1
  Sudan	7	11	7	17	10	7	6	8
  Syria	2	0	0	8	1	2	0	8
  Tunisia	7	11	10	0	9	0	0	16
  UAE	1	0	0	8	2	4	6	1
  Yemen	1	0	0	0	0	0	0	1
  Average	5	5	5	4	5	5	5	4
  Standard deviation	5	6	6	7	7	7	7	5
  Coefficient of variation (%)	113	140	126	174	157	159	134	111

  and Table 3).
• Hargreaves-Samani Equation (HS): This empirical method is commonly used when comprehensive meteorological data are unavailable. It has been used in less than 2% of ET studies conducted across Arab countries, especially in Iraq (21%) and Morocco (16%). Integrating wind speed into the HS equation, coupled with an optimised coefficient set, significantly enhances its predictive reliability across diverse Iraqi regions, including high-altitude areas [22,26]. Ahmed et al. and Hadria et al. [23,27] claimed that the HS equation demonstrated moderate agreement with the PM method across Morocco’s arid and semi-arid zones. While practical in data-scarce settings, its accuracy declined in high-altitude or variable-climate regions. Recent calibration efforts of the HS equation across the Arab region emphasize the necessity of empirical refinement and contextual sensitivity in reference ET modeling. In Egypt, Anwar et al. [28] integrated a recalibrated HS formulation into the RegCM4 framework, demonstrating strong agreement with ERA5-derived PET estimates. Notably, modifying the radiation coefficient from 0.0135 to 0.0105 yielded superior bias reduction compared to adjustments in the temperature coefficient (17.8), particularly enhancing performance in arid and semi-arid zones with limited station data. Complementing this, Al-Alsadi et al. [26] assessed three HS variants across 103 Iraqi stations, revealing that the default global coefficients were inadequate for reliable ET₀ estimation. The wind-adjusted variant also achieved full calibration and validation success. Collectively, these findings underscore the value of region-specific calibration, especially when incorporating local climatic drivers such as wind speed, in strengthening the operational reliability of HS-based ET₀ estimation under data-scarce conditions across the Arab domain.
• Hargreaves Equation: This simplified empirical method estimates evapotranspiration using only temperature and extraterrestrial radiation, making it especially suitable for regions where meteorological data are scarce or incomplete. This equation has been used in approximately 3% of ET studies conducted across Arab countries, particularly in Iraq (17%), Morocco (14%), and Saudi Arabia (14%). In Morocco’s semi-arid region, unlike Priestley–Taylor and Makkink, which underestimated ET₀, Hargreaves showed minimal bias, proving effective under data-scarce conditions [24,29].
• Thornthwaite Equation (TH): This temperature-based empirical method estimates potential ET using only mean monthly air temperature and latitude [17]. This model has been used in about 1% of ET studies conducted across Arab countries, particularly in Morocco (1.4%), Algeria (2.3%), and Sudan (1.9%). The TH equation was evaluated alongside other temperature-based models for estimating ET₀ in Al-Baha, Saudi Arabia. While it offered simplicity, its performance was less accurate than the Linacre method (a temperature-based model), highlighting the need for local calibration under arid conditions [30]. The TH equation produced consistently higher evapotranspiration (ET₀) estimates than the Blaney–Criddle, Pan evaporation, and Penman methods under desert conditions of Qatar [31]; during winter, TH values ranged from 2.3 to 6.2 mm/day, exceeding the 0.67 to 3 mm/day range observed in the other methods. Similarly, in summer, TH estimates spanned 11.1 to 12.9 mm/day, compared to 5.2 to 7.9 mm/day for the remaining approaches. These results highlight the tendency of the TH method to overestimate ET₀, particularly under arid conditions, reinforcing the need for regional calibration [31].
• Remote Sensing Approaches: Given the sparse and uneven distribution of ground-based meteorological stations in many Arab countries, remote sensing methods have become increasingly important. Common satellite sources include MODIS, Landsat-8, and Sentinel-2. Validation protocols typically involve field observations and benchmarking against the FAO–Penman–Monteith method. Validation protocols involve field observations and the PM model [4,32,33]. These remote sensing-based models are effective for regional-scale ET mapping and have been validated using ground-based PM estimates in 24% of all ET studies in Arab countries, with extensive validation in Morocco (32%), Egypt (12%), Sudan (10%), and Tunisia (9%). Using Landsat-8 imagery and the SEBAL model, actual ET for winter wheat in Sudan’s central clay plain was accurately estimated [34].
• Machine Learning (ML): Machine learning techniques are increasingly used to estimate evapotranspiration (ET) by identifying complex, non-linear patterns in environmental data. ML approaches, including decision trees, support vector machines, and neural networks, offer high flexibility and can outperform conventional methods in data-rich environments. However, they require large, well-curated datasets and careful calibration to prevent overfitting, and their performance can vary depending on the model type and regional characteristics. Machine learning (ML) has been applied in only approximately 4% of evapotranspiration (ET) studies conducted in Arab countries. Notably, Morocco (25%), Egypt (18%), and Iraq (18%) account for the highest shares of these applications. However, nearly 50% of Arab countries have yet to produce a single ML-based ET study, indicating significant regional disparities and untapped research potential. Ahmed et al. [23] applied a bagging machine learning algorithm to assess evapotranspiration dynamics across Sudan, revealing a declining trend in actual evapotranspiration (AET) at −0.2% per year since 1960. This reduction, linked to rising temperatures and recurrent droughts, underscores the need for adaptive water management strategies. Despite the growing interest in ML for satellite-based ET modeling, its adoption remains limited due to several technical challenges. Overfitting is a persistent concern, particularly when models are trained on region-specific datasets with sparse ground truth, leading to poor generalization across diverse climatic zones [4,35]. Data preprocessing also poses a significant hurdle, as satellite inputs from MODIS, Landsat-8, and Sentinel-2 require rigorous cloud masking, radiometric correction, and temporal harmonization to ensure consistency [36]. Furthermore, the interpretability of complex ML models, especially deep learning architectures, remains a barrier to institutional and policy uptake, as black-box outputs lack transparency and hinder stakeholder trust [37]. Computational constraints further limit experimentation, particularly in resource-constrained settings where access to cloud platforms or GPU clusters is restricted [4]. Lastly, temporal drift due to climate variability challenges model robustness, as shifting vegetation dynamics and land-use patterns require frequent retraining to maintain predictive accuracy. Addressing these barriers through explainable AI, standardized preprocessing pipelines, and scalable validation frameworks is essential for advancing ML integration in ET monitoring across arid and semi-arid regions.
• Artificial Intelligence (AI): AI-based methods for estimating evapotranspiration (ET) utilise advanced computational algorithms that surpass traditional rule-based or empirical models. AI models offer high predictive accuracy when trained with robust, high-resolution data; however, they can be computationally intensive and require careful validation to ensure generalizability across diverse environmental conditions [4,38]. The application of AI methods in ET research in Arab countries remains in its infancy, accounting for only 1% of the total studies. Among these, Egypt, Iraq, and Saudi Arabia each contribute approximately 18%, reflecting early but concentrated efforts in select regions. Deep learning models, such as LSTM and GRU, demonstrated strong predictive performance for daily ET₀ estimation in Sudan’s coastal Red Sea region, especially when combined with pseudo-labelling [39]. Explainability techniques such as SHAP and LIME, i.e., two widely adopted post-hoc interpretability methods, are increasingly recognized as critical enablers of policy uptake in AI-driven evapotranspiration (ET) modeling. By attributing model predictions to specific input features, such as vegetation indices, surface temperature, and net radiation, these tools provide transparent, instance-level insights into model behavior. This interpretability not only enhances stakeholder confidence but also facilitates regulatory validation and integration into national water governance frameworks. As highlighted by Linardatos et al. [37], such transparency is essential for managing uncertainty and ensuring the responsible deployment of complex AI and ML systems in sensitive environmental domains.
• Other Empirical and Semi-Empirical Methods: Additional approaches, such as the Blaney-Criddle, Priestley-Taylor, Kimberly-Penman, and Pan Evaporation methods, are also referenced in about 51% of ET studies across Arab countries. While less commonly used, these techniques are often adapted to specific local conditions or applied when data availability is constrained [30], particularly in Egypt (16%) and Tunisia (16%). A comparison of ET₀ estimation methods against the PM standard found that the Makkink and Priestley–Taylor models could serve as viable alternatives under semi-arid conditions in Tunisia [40]. However, in Morocco’s Tensift Al Haouz region, both models significantly underestimated ET₀, highlighting the location-dependent nature of their performance [29]. This variability underscores the need for regional calibration and adaptation when applying empirical ET models across diverse climatic zones. A cross-regional comparison of empirical ET₀ models reveals significant variability in their performance under semi-arid conditions. In Tunisia, Latrech et al. [41] found that the Makkink and Priestley–Taylor models closely aligned with the FAO-PM standard, yielding high R² values (0.80 and 0.89) and low RMSE values (0.93 and 1.04 mm day⁻¹), indicating strong predictive reliability. However, in Morocco’s Tensift basin, Er-Raki et al. [29] observed substantial underestimation by the same models, 18% for Makkink and 20% for Priestley–Taylor, when using default coefficients. Calibration of the empirical parameters α and Cm via linear regression changed the bias in relative humidity to 4% and 9%, respectively, improving model accuracy by up to 76%. These findings underscore the necessity of localized calibration when applying radiation-based ET₀ models across diverse Mediterranean semi-arid zones.

Methods such as Thornthwaite (TH), ML, and RS exhibit the highest variability (CV > 150%), suggesting uneven adoption across the region, compared to PM and other methods, which showed relatively lower CVs (~110%), indicating more consistent usage across the Arab region (Table 2). This may be attributed to differences in technical capacity, research priorities, or data availability, as discussed above.

Table 3. Overview of Evapotranspiration (ET) Models Commonly Applied in the Arab Region.

Model	Inputs	Strengths	Limitations	Typical Applications	Reference
Penman-Monteith	Full meteorological data	Physically robust; internationally standardized	Requires complete, quality-controlled data	Irrigated agriculture; climate zones with full data	Allen et al. [5]
Hargreaves–Samani	Minimum and maximum temperature, extraterrestrial radiation	Simple, low data demand; suitable for arid regions with sparse observations	Sensitive to calibration; less accurate in humid climates	Regional climate modeling, data-scarce environments	Pandey and Pandey [42]
Makkink	Solar radiation, air temperature	Good performance in temperate climates; fewer inputs than Penman–Monteith	Less reliable in arid zones; assumes constant psychrometric conditions	Irrigation scheduling, temperate zone hydrology	Zhang et al. [43]
Priestley–Taylor	Net radiation, temperature, humidity	Strong performance in humid regions; physically grounded	Overestimates ET0 in arid zones; assumes ample moisture availability	Wetland hydrology, humid climate ET0 estimation	Xie et al. [44]
Thornthwaite	Mean monthly temperature, latitude	Minimal data requirement; useful for long-term climatological studies	Oversimplified; poor accuracy in arid and tropical climates	Historical reference ET trends, water balance studies	Song et al. [45]
Remote Sensing (RS)	Satellite data (LST, NDVI, albedo, etc.)	Scalable; spatially continuous; multi-sensor integration	Requires validation; complex processing	Regional ET mapping; crop monitoring	Elnashar et al. [46]
Machine Learning (ML)	Multi-variable datasets; RS + meteorological	Captures non-linear patterns; flexible	Needs large training datasets; risk of overfitting	ET prediction in data-rich zones; fusion modeling	Amani et al. [38]
Artificial Intelligence (AI)	Big data; RS inputs; climate variables	High precision; dynamic modeling	Computationally intensive; limited regional adoption	Climate resilience modeling; urban hydrology	Goyal et al. [47]
Fusion (RS + ML/AI)	RS imagery + ground data + ML algorithms	High-resolution; reduces error; scalable	Sparse adoption; requires interdisciplinary expertise	ET monitoring in arid zones; policy integration	Guzinski et al. [21]

Table 3. Overview of Evapotranspiration (ET) Models Commonly Applied in the Arab Region.

Model	Inputs	Strengths	Limitations	Typical Applications	Reference
Penman-Monteith	Full meteorological data	Physically robust; internationally standardized	Requires complete, quality-controlled data	Irrigated agriculture; climate zones with full data	Allen et al. [5]
Hargreaves–Samani	Minimum and maximum temperature, extraterrestrial radiation	Simple, low data demand; suitable for arid regions with sparse observations	Sensitive to calibration; less accurate in humid climates	Regional climate modeling, data-scarce environments	Pandey and Pandey [42]
Makkink	Solar radiation, air temperature	Good performance in temperate climates; fewer inputs than Penman–Monteith	Less reliable in arid zones; assumes constant psychrometric conditions	Irrigation scheduling, temperate zone hydrology	Zhang et al. [43]
Priestley–Taylor	Net radiation, temperature, humidity	Strong performance in humid regions; physically grounded	Overestimates ET0 in arid zones; assumes ample moisture availability	Wetland hydrology, humid climate ET0 estimation	Xie et al. [44]
Thornthwaite	Mean monthly temperature, latitude	Minimal data requirement; useful for long-term climatological studies	Oversimplified; poor accuracy in arid and tropical climates	Historical reference ET trends, water balance studies	Song et al. [45]
Remote Sensing (RS)	Satellite data (LST, NDVI, albedo, etc.)	Scalable; spatially continuous; multi-sensor integration	Requires validation; complex processing	Regional ET mapping; crop monitoring	Elnashar et al. [46]
Machine Learning (ML)	Multi-variable datasets; RS + meteorological	Captures non-linear patterns; flexible	Needs large training datasets; risk of overfitting	ET prediction in data-rich zones; fusion modeling	Amani et al. [38]
Artificial Intelligence (AI)	Big data; RS inputs; climate variables	High precision; dynamic modeling	Computationally intensive; limited regional adoption	Climate resilience modeling; urban hydrology	Goyal et al. [47]
Fusion (RS + ML/AI)	RS imagery + ground data + ML algorithms	High-resolution; reduces error; scalable	Sparse adoption; requires interdisciplinary expertise	ET monitoring in arid zones; policy integration	Guzinski et al. [21]

The analysis of ET estimation methods across Arab countries reveals notable disparities in research volume and methodological preferences (Appendix A). Countries such as Morocco, Egypt, Tunisia, and Saudi Arabia lead in publication output, reflecting their strategic emphasis on water resource management and agricultural sustainability. Recent advancements in ET modelling, particularly through remote sensing (RS), machine learning (ML), and artificial intelligence (AI), have gained traction, now comprising approximately 30% of all ET-related studies in the region. This trend signals a growing reliance on satellite-based data and spatial analytics to monitor water loss across varied landscapes. Nonetheless, traditional standalone models, especially the PM method, continue to dominate, featuring in nearly 70% of the studies.

While there is a gradual shift toward hybrid approaches that integrate RS, ML, and AI to enhance model performance and climate resilience, the actual adoption of fusion methodologies remains limited. Noureddine et al. [40] showed that hybrid modelling (ET—aridity coupling) markedly improves ET accuracy, reducing MAE from 39% to 2% in Mekkah and Medinah, Saudi Arabia. These metrics enhance and support early flood risk alerts. However, only 8 studies, representing about 2% of those utilizing RS, ML, or AI, have implemented multi-source data integration for ET estimation, especially in Lebanon (hosting 63% of fusion-based studies). This limited uptake underscores an ongoing reliance on single-method frameworks, which constrain the potential benefits of combining complementary datasets. Barriers such as restricted data availability, computational demands, and the need for interdisciplinary expertise may continue to impede the broader adoption of fusion-based modelling in the region. The limited adoption of hybrid RS–ML–AI frameworks for ET estimation reflects not only methodological preferences but also deep-rooted computational and institutional constraints. As noted by Amani et al. [38], ML models demand intensive hyperparameter tuning and are sensitive to user expertise and input data characteristics, challenges exacerbated by high-resolution imagery that increases dimensionality and computational load. Their limited extrapolation capacity further restricts generalization in heterogeneous or data-scarce regions such as the Arab domain. These technical barriers are compounded by limited access to cloud infrastructure, sparse ground truth data, and the absence of standardized RS–ML integration workflows. Addressing these gaps requires both algorithmic innovation and institutional capacity-building in reproducible modeling and data governance. In this context, El-Shirbeny et al. [4] demonstrated the utility of Google Earth Engine (GEE) for vegetation classification across the Arab region using MODIS datasets. GEE’s cloud-based architecture mitigates hardware and data access limitations by enabling scalable, multi-temporal processing and reproducible scripting, making it particularly valuable for ET mapping in resource-constrained environments.

Future research should prioritise advancing fusion methodologies, particularly refining ML algorithms for region-specific ET forecasting, integrating RS with AI-driven climate models, and ensuring that hybrid frameworks are adapted to local hydrological and agricultural conditions. Strengthening institutional collaborations and increasing funding for multi-method research could accelerate the adoption of fusion techniques, ultimately supporting sustainable water management solutions across drought-prone environments. Recommended frameworks include FAO WaPOR, SEBAL-ML fusion workflows, and Sentinel-Landsat hybrid pipelines for ET estimation. FAO WaPOR provides robust, open-access ET datasets validated across arid zones, while SEBAL offers physically grounded ET modeling using satellite-derived surface energy balance components. Coupling SEBAL outputs with random forest machine learning enhances spatial generalization and allows for dynamic calibration using vegetation indices and land surface temperature. This hybrid configuration is particularly effective in data-scarce environments and aligns with cloud-based platforms such as Google Earth Engine, facilitating regional-scale ET monitoring and decision support.

Despite the growing adoption of advanced methodologies, such as remote sensing (RS), machine learning (ML), and artificial intelligence (AI), the lack of extensive ET research could hinder the development of effective climate adaptation strategies. More collaborative studies integrating field observations, modelling approaches, and policy-driven frameworks are needed to ensure sustainable water management in these highly vulnerable environments in the Arab region. This highlights the urgent need for expanded research efforts, cross-border collaborations, and increased funding to enhance regional ET studies, providing data-driven solutions for long-term climate adaptation and water security.

3.3. Fusion

Table 4. Statistical evaluation of machine learning-based and traditional evapotranspiration (ET) fusion methods using mean absolute error (MAE), root mean square error (RMSE), and Pearson correlation (r).

Metric	Value	What It Means
MAE	215 mm	On average, there is a 215 mm/year difference between ML-fused and traditional ET.
RMSE	531 mm	Sensitive to outliers—high values in some regions create larger deviations.
r	0.86	Strong positive relationship—ML fusion follows the same spatial trend as traditional, but with improved precision.

presents the evaluation metrics used to compare the performance of the machine learning-based and traditional ET fusion approaches. The machine learning fusion method outperformed the traditional averaging technique in estimating evapotranspiration (ET), aligning with earlier studies that have demonstrated the advantages of random forest models in capturing spatial variability in hydrological applications [48]. The ML-fused ET product achieved a mean absolute error (MAE) of 215.08 mm/year and a root mean square error (RMSE) of 531.34 mm/year, indicating a substantial reduction in prediction error. A Pearson correlation coefficient of 0.86 between the ML and traditional fusion outputs reflects strong spatial agreement, while the difference map (Figure 4) highlights localised corrections introduced by the ML model. These improvements were particularly pronounced in regions with nonlinear relationships between ET drivers, suggesting the model’s capacity to accommodate complex data patterns [49]. Visual assessment of the ML-derived ET map (Figure 5) further confirmed an enhancement in spatial detail and heterogeneity, reinforcing the suitability of data-driven models for ET monitoring in data-scarce, arid regions [50].

The ML-Fused product exhibited a small fraction of negative evapotranspiration (ET) values, i.e., 3.508% of valid pixels, which do not represent actual water fluxes. These anomalies are interpreted as diagnostic artifacts resulting from (i) model extrapolation over low-confidence surfaces (e.g., bare soil, urban zones), (ii) fusion residuals at spatial boundaries, and (iii) the lack of embedded physical constraints in the ML architecture. All inputs underwent rigorous atmospheric correction, including cloud screening and radiometric normalization, minimizing residual effects prior to fusion. Accordingly, negative ET values were masked during post-processing to preserve hydrological realism while retaining their diagnostic value.

Such artifacts are consistent with behaviors observed in both physical-based and data-driven ET models under sparse training or low-signal conditions. Hu et al. [51] highlight structural limitations and interpretability challenges in these approaches. Their comparative analysis showed that while data-driven models (e.g., Random Forest) achieved lower RMSE values (32–53 W·m⁻²) than the SEBS model (121 W·m⁻²), they remain susceptible to spurious correlations and nonphysical outputs when trained on biased or incomplete datasets. The absence of physical constraints in ML architectures can exacerbate instability, particularly in radiatively unstable or sparsely vegetated zones—underscoring the importance of anomaly detection and hydrological realism enforcement in fusion workflows.

Additional studies [52,53] attribute fusion uncertainty to input retrieval errors, temporal interpolation, and pixel selection logic. Lun et al. [54] reported diminished model performance over sand dune ecosystems (R² = 0.72), reinforcing the need for robust masking protocols in unstable terrains. Collectively, these findings affirm that nonphysical ET values should be flagged, visualized, and excluded from downstream analyses.

To assess pixel-wise fusion behavior, binary masks were applied to identify regions where ET_fused was lower than ET_SEBAL and ET_WaPOR. Results indicated that ET_fused was lower than SEBAL in 48.66% and lower than WaPOR in 48.72% of valid pixels, with 31.78% falling below both inputs simultaneously (Figure 5). This pattern suggests active downscaling by the fusion algorithm, indicating it does not merely replicate inputs but adjusts ET estimates to reconcile potential overestimations, i.e., resulting in consistent and comprehensive ET estimates. The spatial distribution of these masks could provide a diagnostic layer for evaluating fusion performance and guiding land cover—specific QA. The study of Jamshidi et al. [55] proposed a modified Landsat-MODIS fusion model to estimate daily high-spatial-resolution ET over heterogeneous, sparsely vegetated regions. By incorporating pixel-wise filtering and NDVI-based coefficients, the model reduced bias and improved accuracy (RMSE: 0.37 mm/day, Relative error, RE: 3.5%). It enabled efficient ET mapping for irrigation scheduling in data-scarce semi-arid landscapes.

3.4. ET Implications for Climate Resilience and Agricultural Sustainability in the Arab Region

Food security and climate change remain key areas of concern in Arab countries [56]. Evapotranspiration (ET) modelling is fundamental to sustainable water management in Arab countries, where severe water scarcity and climate change pose significant threats to agriculture, urban development, and ecosystem stability. Given the region’s heavy dependence on irrigation for food security, advanced ET research remains crucial for enhancing water use efficiency and ensuring long-term agricultural sustainability [57,58]. Beyond agriculture, ET modelling plays a vital role in urban hydrology, helping cities manage evaporation losses, optimise water allocation, and support green infrastructure development to improve climate resilience [59].

3.4.1. ET Research Addressing Climate Change in Arab Countries

The analysis of evapotranspiration (ET) studies related to climate change reveals pronounced disparities in research engagement across the Arab region (Appendix B). Morocco leads with the highest number of climate-related ET studies (n = 64), followed by Egypt (48), Iraq (40), Tunisia (37), and Algeria (37). This concentration reflects a strong alignment between ET modelling and national concerns over climate change and vulnerability, particularly in agriculturally intensive and drought-prone countries [60,61]. The common feature among Morocco, Egypt, Tunisia, and Iraq is their high vulnerability to water scarcity and climate change, which drives extensive ET modelling research in these countries [62,63]. Their heavy reliance on irrigated agriculture and fragile water resources makes ET estimation critical for sustainable water management, food security, and climate adaptation policies [64]. This research focus reflects their commitment to optimising irrigation efficiency, mitigating drought impacts, and enhancing climate resilience through advanced hydrological studies [65].

Despite the significant economic differences, both low-income nations, such as Comoros, Mauritania, and Djibouti, and high-income ones like Kuwait and Bahrain, exhibit limited climate-related ET studies [66]. This disparity highlights that strong research infrastructure and supportive policy frameworks are crucial for advancing ET modelling efforts, regardless of the country’s economic classification. Without these foundations, countries are likely to fall behind in climate-related environmental technology (ET) research, undermining regional climate mitigation and adaptation strategies. Moreover, this imbalance suggests that institutional capacity, research prioritization, and funding availability, rather than financial wealth alone, are the key drivers of scientific output in the Arab region [67]. Al-Sarihi and Luomi [68] emphasize that regional governance and cooperation are vital for addressing climate change in the Arab region, as transboundary challenges like climate change require coordinated responses beyond national borders. Their findings also highlight the necessity of policy coordination, knowledge exchange, technical assistance, and financial mobilization to establish effective climate governance frameworks among Arab countries.

However, persistent weaknesses, including fragmented institutions, unclear implementation targets, and limited transparency, continue to impede progress. The Arab region has yet to establish any significant action-oriented region-wide climate initiatives/partnerships, further delaying coordinated efforts to address climate challenges [68].

Studies at broader scales (e.g., North Africa, MENA, and Arabian Peninsula) contribute modestly to overall evapotranspiration (ET) research, with regional syntheses remaining underutilised [69]. North Africa accounts for only 58 studies, emphasising country-specific rather than comprehensive regional analyses. The lack of climate-related ET studies labelled under “Arab countries” or as pan-Arab collaborations further reflects the fragmented nature of scientific initiatives, limiting the ability to develop a holistic understanding of hydrological responses to climate stressors [48,68].

Enhancing sub-regional cooperation, integrating ET modelling into climate policies, and ensuring sustained funding mechanisms are crucial for enhancing climate resilience and effective mitigation strategies across the Arab world. While 32% of ET studies indexed in the Web of Science in the region address climate change, gaps in large-scale research hinder the development of shared adaptation and mitigation strategies. Research disparities suggest that institutional capacity and funding availability, rather than environmental urgency alone, drive scientific production [67].

3.4.2. ET Research in the Context of Food Security in Arab Countries

Agriculture remains the largest consumer of freshwater in Arab countries, accounting for 85% of total water use based on the Arab Water Security Strategy for the period 2010–2030. The strategy indirectly supports evapotranspiration (ET) research by prioritising reduced agricultural water loss through modern irrigation techniques and drought-resistant crops. Another important regional strategy to mention here is the Arab Strategy for Sustainable Agricultural Development 2020–2030, which focuses on food security, resource efficiency, and climate adaptation by modernising agriculture, enhancing trade policies, and strengthening governance across the region. However, the 2025 Global Report on Food Crises highlights worsening food insecurity across several Arab nations due to climate extremes, economic instability, and conflicts, reinforcing the urgency of integrating ET research into food security strategies.

The water-climate-energy-food security nexus remains a pressing challenge in the Arab region, as demonstrated by the varying emphasis on ET-related studies addressing food security concerns. Egypt (15 studies), Sudan (11 studies), Morocco (9 studies), Jordan (7 studies), and North Africa as a region (7 studies) lead in research engagement, reflecting a growing recognition of ET’s role in agricultural sustainability and food production. These countries rely heavily on irrigated agriculture and are particularly vulnerable to climatic stressors, which likely contribute to their high ET in food security research.

Conversely, several countries, including Bahrain, Comoros, and Mauritania, show no documented ET studies linked to food security. Others, such as Djibouti, Kuwait, Palestine, and Yemen, have only one study each, highlighting apparent gaps in research investment despite shared concerns over water scarcity and food insecurity (Appendix B). These findings suggest that institutional capacity, national research agendas, and funding mechanisms have a significant influence on whether ET research aligns with food security priorities.

At broader regional scales, studies in Sub-Saharan Africa (SSA) and the Middle East and North Africa (MENA) contribute modestly to overall ET-food security research, underscoring the importance of transboundary cooperation in addressing shared food security challenges [70]. The limited food security-linked ET studies under the label “Arab countries” further underscore the fragmentation of regional research efforts.

The land-water linkage is fundamental to sustainable resource management, particularly in arid and semi-arid regions such as the Arab world. Effective land use planning has a direct influence on water availability, as deforestation, urbanisation, and agricultural expansion alter hydrological cycles [71]. The analysis of food security-focused ET studies indicates no strong correlation between arable land area and research intensity across the Arab region. Despite possessing vast agricultural land, approximately 42 million hectares (Mha) in Algeria and 69 Mha in Sudan, these countries exhibit limited engagement in food security-specific ET research, with Algeria conducting only five studies and Sudan eleven. In the Arab region, transboundary water cooperation is critical, as shared water resources account for two-thirds of freshwater availability [72]. However, weak governance and fragmented policies hinder integrated land-water management, exacerbating water scarcity and food insecurity [73]. Strengthening regional collaboration and adaptive policies can enhance water conservation and agricultural sustainability in climate-vulnerable landscapes. The current disconnect between land and water research underscores a critical gap in understanding food sustainability, raising concerns over the long-term viability of current land and water resource management strategies and climate resilience in the Arab region.

Overall, the findings demonstrate an urgent need to integrate ET estimation more directly into food security strategies, particularly in underrepresented countries. Scaling up research through multi-country collaborations, targeted funding, and cross-disciplinary integration would be instrumental in enhancing food system resilience across the Arab world.

3.5. Strategic Alignment of Evapotranspiration (ET) Research with Research and Development Investment in the Arab Region

Evapotranspiration (ET) research is critical for water resource management and agricultural sustainability, particularly in the Arab region, where water scarcity and climate change pose significant challenges. The extent to which ET studies align with national Research and Development (R&D) investment influences research output, funding priorities, and policy integration, shaping long-term sustainability efforts.

Strengthening sub-regional cooperation, integrating climate policies, and ensuring sustained funding mechanisms are vital for enhancing climate resilience and advancing mitigation strategies across the Arab world. However, substantial disparities exist in research support for ET studies (Figure 6). Countries such as Qatar exhibit high levels of ET research funding (57%), reflecting a strong institutional commitment to water management and agricultural sustainability. In contrast, 32% of Arab countries show no evidence of funded ET studies, largely due to limited financial resources, competing policy priorities, or a lower perceived importance of ET research. Moussa et al. [57] emphasise the urgent need for sustained research funding and regional collaboration, particularly in GCC countries, where severe water scarcity poses a significant threat to long-term food security. Investing in regional-scale research initiatives is crucial for developing sustainable water management strategies, enhancing agricultural resilience, and ensuring food security.

The gross domestic expenditure on research and development (R&D) in the Arab region, measured as a percentage of GDP, averaged 0.7% (1996–2024), compared to the global average of 2.68%, based on World Bank datasets. Across countries, this figure varies significantly, ranging from 1.06% in the UAE, indicating relatively substantial investment in R&D, to 0.01% in Mauritania, highlighting substantial disparities in research funding. The analysis reveals a moderate positive linear relationship (R² = 0.51) between agricultural R&D expenditure and ET research volume, indicating that higher investment in agricultural innovation generally correlates with an increase in ET studies (Figure 7). However, the weaker correlation (R² = 31%) between agricultural R&D expenditure and climate-focused ET research highlights a funding misalignment, where financial allocations for agricultural innovation do not necessarily prioritise ET-related studies.

Disparities between investment and research output are evident across Arab nations. For example, Morocco allocated $99.94 million to agricultural R&D, yet recorded no funded ET studies, while Qatar, with only $1.55 million in agricultural R&D investment, supported eight financed studies. This inconsistency reflects institutional prioritisation differences, policy frameworks, and research agendas, indicating that funding availability alone does not determine ET study engagement. Instead, governance structures, national policies, and research integration mechanisms play a crucial role in shaping the trajectories of ET research.

Despite contributions from select countries, only 3.9% of ET studies in the Arab region receive dedicated funding, underscoring a significant gap in environmental research investment. This limitation restricts the application of advanced methodologies—such as artificial intelligence (AI) and machine learning (ML)—in ET modelling, while also reducing international collaboration opportunities and hindering policy-driven advancements in climate resilience.

The correlation between R&D investment and the number of ET-related food security studies (R² = 59%) suggests a strong connection, indicating that higher financial support for agricultural R&D translates into more extensive ET research linked to food security. This aligns with the fact that Arab-region agriculture is heavily dependent on effective water management, where ET plays a crucial role in irrigation optimisation and crop productivity. Nations with higher R&D expenditures prioritise ET research to improve food security, integrating advanced irrigation systems, precision agriculture, and hydrological modelling to mitigate the impacts of water scarcity on food production. However, the lower correlation (R² = 39%) between R&D investment and ET-related climate change studies suggests that climate adaptation research remains underfunded compared to research on food security. Several factors contribute to this discrepancy:

• Research priorities—Many Arab countries prioritise food security over long-term climate adaptation, given immediate socio-economic challenges.
• Policy misalignment—While agricultural innovation is widely supported, climate-focused ET studies often lack dedicated funding mechanisms and institutional backing.
• Data scarcity issues—Climate change research, particularly ET modelling for adaptation strategies, requires high-resolution, long-term datasets, which are limited across the Arab region, thereby restricting the study scope.
• Technological accessibility—AI-driven ET fusion models for climate adaptation are cost-intensive and require advanced expertise, leading to slower adoption and fewer studies in climate resilience domains.

To enhance R&D contributions to climate resilience, governments and research institutions should:

• Increase targeted funding for ET-based climate resilience studies.
• Integrate ET research into national climate policies, bridging the gap between food security and environmental sustainability.
• Enhance regional collaboration to facilitate shared research efforts and cross-country adaptation initiatives.
• Improve data collection and accessibility to support AI-driven ET modelling and advanced fusion techniques.

Aligning R&D investment with both food security and climate adaptation priorities is crucial for establishing a sustainable agricultural framework in the Arab region, thereby ensuring water-efficient crop production and long-term environmental resilience.

4. Conclusions

This study synthesizes ET research across Arab countries, revealing dominance of traditional methods and emerging data-driven approaches. It demonstrates the viability of ML-based fusion modelling and highlights strategic gaps in funding, collaboration, and policy integration.

Importantly, the detection of nonphysical ET artifacts in the ML-Fused product, affecting 3.5% of valid pixels, reinforces the need for hydrological realism enforcement and anomaly masking in fusion workflows. These artifacts, consistent with known limitations in data-driven models, underscore the importance of embedding physical constraints and diagnostic QA layers, particularly in radiatively unstable or data-sparse regions.

Recommendations include scaling fusion frameworks with built-in anomaly detection, enhancing regional cooperation on QA standards, and aligning ET research with climate and food security agendas through reproducible, physically grounded modeling strategies.