Comparative analysis of empirical evapotranspiration models across Northern Ethiopia’s Climatic Zones Authors: Hana Mekonnen, Ephrem Yetbarek, Chalachew Abebe Mulatu Published: 2026 ABSTRACT AbstractAccurate estimation of reference evapotranspiration (ETo) is essential for irrigation scheduling, water resources management, and climate adaptation. This study evaluated the performance of twelve empirical ETo models at Bahir Dar, Gondar, Debre Tabor, Debre Markos, Kombolcha, Mekelle and Adigrat meteorological stations representing diverse climatic and topographic conditions in northern Ethiopia. The assessed models included temperature-based, radiation-based, and mass-transfer methods. ETo estimates were compared against the FAO 56 Penman–Monteith method. The results revealed substantial spatial variability in model performance, reflecting the climatic heterogeneity of the region. The temperature-based Enku model consistently outperformed other methods at five stations, achieving IoA values of up to 0.87, NSE values exceeding 0.70, and NRMSE as low as 12.61%, particularly in humid and sub-humid highland environments. In contrast, the radiation-based Abtew model provided the most accurate estimates in the arid Mekele station (IoA = 0.87, NRMSE = 19.59%), highlighting the dominant influence of solar radiation under water-limited conditions. The Priestley–Taylor, Turc, and Makkink models generally exhibited lower accuracy and larger biases across most stations. The findings demonstrate that no single empirical model is universally applicable across northern Ethiopia and emphasize the importance of climate-specific model selection, local calibration, and validation.Keywords: Ethiopia, Evapotranspiration, FAO-56 PM, Empirical models, Model comparisonSubject terms: Climate sciences, Environmental sciences, Hydrology Introduction Empirical evapotranspiration (ET) models are simplified mathematical equations designed to estimate the amount of water lost from soil and vegetation through the processes of evaporation and transpiration. These models achieve this by establishing relationships between ET and readily measurable weather parameters, such as air temperature, solar radiation, humidity, wind speed, and vapor pressure deficit1–3. Accurate estimation of ETo is particularly important in developing countries where agriculture remains the dominant economic sector and increasing pressure on water resources threatens food security and sustainable development4,5. Reliable ETo estimates support crop water requirement determination, irrigation system design, water allocation planning, and drought monitoring, thereby contributing to improved agricultural productivity and water-use efficiency6,7. The FAO-56 Penman–Monteith (FAO-56 PM) method is widely recognized as the international standard for estimating reference evapotranspiration because of its strong physical basis and robust performance across diverse climatic conditions2. However, its application requires complete and reliable measurements of solar radiation, air temperature, relative humidity, and wind speed. Such comprehensive meteorological datasets are often unavailable in many developing countries, including Ethiopia, where weather stations frequently contain missing records, short observation periods, or incomplete measurements8,9. Consequently, numerous empirical ETo models have been developed that require fewer meteorological inputs and can be grouped into temperature-based, radiation-based, and mass-transfer approaches7,10. Over recent decades, a wide range of empirical ETo models have been evaluated worldwide, with results consistently demonstrating that model performance is strongly climate dependent. Temperature-based models such as Hargreaves–Samani, Thornthwaite, and Blaney–Criddle are widely applied because of their minimal data requirements, particularly in data-scarce regions. However, their accuracy often varies substantially across climatic zones11,12. Radiation-based methods, including Priestley–Taylor, Makkink, Turc, and Abtew, have shown satisfactory performance in environments where available energy is the dominant control on evapotranspiration, while mass-transfer approaches perform better where aerodynamic factors exert stronger influence10,13. Recent studies have further confirmed that no single empirical model consistently outperforms others across all climatic conditions, emphasizing the need for site-specific evaluation and calibration14–16. Recent advances in ETo model evaluation have moved beyond traditional single-metric assessments toward comprehensive multi-criteria frameworks that simultaneously consider accuracy, bias, agreement, and uncertainty. Studies conducted in the Sahel region and other climate-sensitive environments have demonstrated that model rankings can vary considerably depending on climatic conditions and evaluation criteria, highlighting the importance of robust decision-support frameworks for model selection17,18. Similarly, recent investigations in West Africa, East Africa, the Middle East, South Asia, and Latin America have reported substantial spatial variability in model performance due to differences in humidity, solar radiation, cloud cover, and wind regimes13,19–21. These studies collectively underscore the necessity of climate-specific model selection and rigorous regional validation before operational implementation. Ethiopia presents a particularly challenging environment for ETo estimation because of its highly variable topography and climate. Elevation ranges from below sea level in the Afar Depression to more than 4,600 m in the Ethiopian Highlands, creating pronounced gradients in temperature, rainfall, humidity, solar radiation, and wind regimes over relatively short distances22,23. Consequently, evapotranspiration processes vary substantially across the country’s humid highlands, transitional mid-altitude zones, and arid lowlands. Previous Ethiopian studies have reported contrasting performance among empirical ETo models, with some methods performing well in specific climatic regions while exhibiting poor accuracy elsewhere24–27. These findings suggest that a single empirical model cannot be universally recommended for nationwide application and reinforce the need for regional model validation. Despite growing research interest in Ethiopia and other sub-Saharan African countries, several important knowledge gaps remain. First, most previous studies have focused on a limited number of stations, individual climatic zones, or a relatively small subset of empirical models, restricting the transferability of their findings. Second, few studies have simultaneously compared temperature-based, radiation-based, and mass-transfer models across multiple Ethiopian climatic zones using a consistent evaluation framework. Third, while recent Sahelian and international studies have increasingly adopted integrated multi-criteria evaluation approaches, similar frameworks remain largely absent in Ethiopian ETo research. Consequently, there is still limited guidance regarding the most suitable empirical ETo models for Ethiopia’s diverse agro-ecological zones, particularly under data-limited conditions. The present study addresses these gaps by conducting a comprehensive evaluation of twelve widely used empirical ETo models across seven meteorological stations representing humid, transitional, and arid climatic conditions in northern Ethiopia. Those models are temperature-based (Enku and Melesse27, Blaney–Criddle28, Hargreaves–Samani29, Thornthwaite30), Radiation-based (Abtew31, Makkink32, Priestley–Taylor33, Turc34), and Mass Transfer-based (Dalton35, Penman36, WMO37, Trabert38), models. Unlike previous Ethiopian studies that evaluated only a few models or focused on individual climatic regions, this study simultaneously assesses temperature-based, radiation-based, and mass-transfer approaches using the FAO-56 PM method as the reference standard. Furthermore, a Multi-Criteria Group Decision-Making (MCGDM) framework is employed to integrate multiple statistical performance indicators and establish climate-specific model rankings. The novelty of this work lies in its comprehensive multi-climatic assessment, inclusion of a locally developed Ethiopian model (Enku), and application of a structured decision-support framework for model selection. Specifically, it is intended to: (i) evaluate the performance of twelve empirical ETo models against the FAO 56-PM reference method, (ii) investigate the influence of climatic variability on model behavior across humid, transitional, and arid environments, and (iii) rank competing models using a multi-criteria decision-making framework and provide climate-specific recommendations for ETo estimation and water resource management in Ethiopia and other data-scarce regions with comparable environmental conditions. Materials and methods Study area description This study was carried out at seven sites in northern Ethiopia, a region marked by complex topography and a tropical highland climate shaped by its equatorial location and altitudinal gradients39. The area experiences Ethiopia’s characteristic tri-modal rainfall pattern: Kiremt (June–September), the main rainy season, contributes 50–70% of the annual precipitation40; Bega (October–January), a dry season, is associated with mild temperatures and occasional rainfall41; and Belg (February–May), a secondary rainy season, provides 20–30% of annual rainfall but is generally less reliable in northern parts of the country42. The study area lies primarily within Ethiopia’s dry sub-humid to semi-arid climatic zones, as defined by the country’s five-tier classification system43. While generally suitable for rain-fed agriculture, the region experiences pronounced microclimatic variability due to its rugged topography5. These conditions increase vulnerability to climate extremes, with historical records indicating recurring droughts every 5–7 years and episodic flash floods, both of which threaten food security44. Dataset For this study, seven meteorological stations located at Bahir Dar, Debre Markos, Gondar, Debre Tabor, Kombolcha, Mekelle, and Adigrat were selected to represent the major climatic and agro-ecological conditions of northern Ethiopia (Fig. 1; Table 1). Monthly meteorological data, including maximum and minimum air temperature, mean relative humidity, sunshine duration, and wind speed measured at 2 m height, were obtained from the Ethiopian Meteorology Institute for the period 1998–2023. These variables were used to estimate reference evapotranspiration (ETo) using the FAO-56 PM method and the selected empirical models. Map of study area. Table 1Description of the seven meteorological stations.SNStationClimate zoneAnnual rainfall (mm)Dominant characteristics1Bahir DarHumid> 1200Lake-influenced, low PET2GondarSub-humid~ 1000Highland climate, moderate humidity3Debre TaborHumid> 1200High elevation, humid, dense vegetation4Debre MarkosHumid1200–1300Cool temperatures, reduced PET5KombolchaSub-humid–Semi-arid800–1000Transitional climate, high PET6MekelleArid–Semi-arid~ 600Dry climate, high PET, sparse vegetation7AdigratSemi-arid650–700Dry climate, moderate rainfall Description of the seven meteorological stations. Prior to analysis, the dataset was subjected to quality-control procedures to identify missing records and physically inconsistent observations. Missing data were filled using standard hydrometeorological techniques. When missing observations constituted less than 10% of the available record and neighboring stations exhibited similar climatic characteristics, the simple arithmetic mean method was applied. For stations with more than 10% missing data, the normal ratio method was used following the recommendations of Chow et al.45 and Adilah & Hannani46. After gap filling, the dataset was reviewed to ensure consistency and completeness before ETo calculations were performed. It should be noted that the analysis was conducted using monthly aggregated meteorological data. While monthly data are appropriate for long-term ETo assessment and regional model comparison, temporal aggregation may smooth short-term weather variability and extreme events. Therefore, the findings of this study primarily reflect model performance for monthly ETo estimation and should be interpreted accordingly when considering applications requiring daily time-step analyses. Methods for calculating reference evapotranspiration The FAO-56 PM Equation2. was employed as the benchmark for estimating monthly reference evapotranspiration (ETo) across the diverse climatic zones of northern Ethiopia. Twelve widely used empirical ETo models, encompassing temperature-based, radiation-based, and mass-transfer approaches, were applied to generate modeled ETo values. Performance of these models was evaluated by comparing their estimates against the FAO-56 PM reference values. FAO-56 Penman–Monteith Method In this study, the FAO-56 PM equation, recommended by the Food and Agriculture Organization, was used to estimate monthly reference evapotranspiration (ETo). As the standard method for ETo estimation, FAO-56 PM is a combination-based model that integrates multiple climatic variables, including air temperature, wind speed, solar radiation, dew point temperature, and atmospheric pressure (Eq. 1). The method estimates ETo for a hypothetical reference grass surface characterized by a height of 0.12 m, a fixed surface resistance of 70 s/m, and an albedo of 0.23. where ETo reference evapotranspiration [mm day− 1], Rn net radiation at the crop surface [MJ m− 2 day− 1], G soil heat flux density [MJ m− 2 day− 1], T air temperature at 2 m height [°C], u2 wind speed at 2 m height [m s− 1], es saturation vapour pressure [kPa], ea actual vapour pressure [kPa], es - ea saturation vapour pressure deficit [kPa], D slope vapour pressure curve [kPa °C− 1], and g psychrometric constant [kPa °C− 1]. The FAO-56 PM ETo values were computed from observed meteorological data and are regarded as a reliable representation of local climatic conditions. Owing to its strong physical foundation and proven accuracy across a wide range of climates, the FAO-56 PM method is widely recognized as the standard for validating empirical or simplified ETo models2. Estimation of ETo using empirical models This study selected the most widely used 12 empirical ETo estimation models which were developed in diverse climatic conditions to conduct the analysis (Table 2). The twelve empirical ETo models evaluated in this study were selected to represent the principal categories of evapotranspiration estimation methods, namely temperature-based, radiation-based, and mass transfer-based approaches. The selection was guided by three considerations: (i) the widespread use and scientific acceptance of the models in evapotranspiration research, (ii) their suitability for data-scarce regions due to relatively low meteorological data requirements, and (iii) their previous application in tropical, subtropical, semi-arid, and highland environments with climatic characteristics comparable to those of Ethiopia. The inclusion of models requiring different combinations of meteorological variables enabled a comprehensive assessment of their applicability under varying levels of data availability. Furthermore, the Enku model was included because it was developed specifically for Ethiopian conditions and has shown promising performance in previous studies. This combination of internationally recognized and locally developed models provides a robust basis for identifying suitable alternatives to the FAO-56 PM method in Ethiopian climatic settings. Table 2List of 12 selected ETo empirical estimation methods.CategoryModelInput VariablesEquationReferencesTemperature-basedBlaney–CriddleTETo = p(0.46T + 8.13)Blaney–Criddle (1950)28ThornthwaiteTmax, Tmin, LETo = PET(N/12)(dm/30)Thornthwaite (1948)30Hargreaves–SamaniT, Tmax, Tmin, RaETo = 0.0023(Tmax − Tmin)0.5(T + 17.8)RaHargreaves and Samani (1985)29EnkuTETo = (Tmax/k)nEnku and Melesse (2014)27Radiation-basedMakkinkT, RsETo = 0.61Δ/(Δ + γ)Makkink (1957)32Priestley–TaylorT, Rs, RHETo = αΔ/(Δ + γ)Priestley and Taylor (1972)33TurcT, RsETo = 0.013[T/(T + 15)](23.88Rs + 50)/λTurc (1961)34AbtewT, RsETo = 0.53(Rs/λ)Abtew (1996)31Mass Transfer-basedDaltonT, RH, uETo = (3.648 + 0.7223u)(es − ea)Dalton (1802)35PenmanT, RH, uETo = 2.625 + 0.000479u(es − ea)Penman (1948)36WMORH, uETo = 1.298 + 0.934u(es − ea)WMO (1966)37TrabertT, RH, uETo = 3.075√u(es − ea)Trabert (1896)38where ETo is the reference evapotranspiration (mm/day), T, Tmax, Tmin are mean, maximum, and minimum air temperature (°C), Td is the daily dew point air temperature in (°C), TD is the difference between the Tmax and Tmin (°C), Rs is the total solar radiation (MJ/m2 /day), Ra is extraterrestrial radiation (MJ/m2/day), Rn is net solar radiation (MJ m− 2/ day), u is wind speed at 2 m (m/s), RH is relative humidity (%), and ea and es represent actual and saturation vapor pressure (kPa), Rs is solar radiation in (MJ/m2 /day), α is a constant (1.26), L is the latitude at each point (degree), z is the elevation (m), λ represents latent heat of evaporation (MJ/Kg), Δ is the slope of saturation vapor pressure-temperature curve (kPa /°C), and γ is the Psychometric constant (KPa/°C). k is computed as k = 48*Tmm − 330 for combined wet and dry conditions or k = 73*Tmm − 1015 for the dry phase, and k = 36*Tmm − 63 for the rain phase, where Tmm (⸰C) is the long term daily mean maximum temperature for the seasons under consideration (Enku and Melesse, 2014), and latent heat of vaporization, 2.45, MJ/kg. List of 12 selected ETo empirical estimation methods. where ETo is the reference evapotranspiration (mm/day), T, Tmax, Tmin are mean, maximum, and minimum air temperature (°C), Td is the daily dew point air temperature in (°C), TD is the difference between the Tmax and Tmin (°C), Rs is the total solar radiation (MJ/m2 /day), Ra is extraterrestrial radiation (MJ/m2/day), Rn is net solar radiation (MJ m− 2/ day), u is wind speed at 2 m (m/s), RH is relative humidity (%), and ea and es represent actual and saturation vapor pressure (kPa), Rs is solar radiation in (MJ/m2 /day), α is a constant (1.26), L is the latitude at each point (degree), z is the elevation (m), λ represents latent heat of evaporation (MJ/Kg), Δ is the slope of saturation vapor pressure-temperature curve (kPa /°C), and γ is the Psychometric constant (KPa/°C). k is computed as k = 48*Tmm − 330 for combined wet and dry conditions or k = 73*Tmm − 1015 for the dry phase, and k = 36*Tmm − 63 for the rain phase, where Tmm (⸰C) is the long term daily mean maximum temperature for the seasons under consideration (Enku and Melesse, 2014), and latent heat of vaporization, 2.45, MJ/kg. Model performance evaluation Model performance was evaluated using six widely applied statistical indicators: coefficient of determination (R²), Nash–Sutcliffe Efficiency (NSE), Root Mean Square Error (RMSE), Normalized RMSE (NRMSE), Mean Absolute Error (MAE), and Index of Agreement (IoA). Because each statistical indicator evaluates a different aspect of model performance, conflicting results may occur among metrics. For example, the coefficient of determination (R²) quantifies the strength of the linear relationship between estimated and reference ETo values but is insensitive to systematic bias and error magnitude. Consequently, a model may achieve a high R² while simultaneously exhibiting large RMSE, NRMSE, or MAE values if it consistently overestimates or underestimates ETo. In contrast, the Index of Agreement (IoA) and Nash–Sutcliffe Efficiency (NSE) evaluate the degree of agreement between estimated and observed values and are more sensitive to prediction errors. Therefore, a high R² should not be interpreted as evidence of superior model performance in isolation. When discrepancies occur among performance metrics, models with consistently favorable agreement and error statistics are generally preferred for practical applications because they provide more reliable estimates of ETo magnitude. This consideration motivated the adoption of the MCGDM framework, which combines multiple complementary indicators to obtain a balanced assessment of model performance. All metrics were calculated using Eqs. (2)-(7). where PM is the FAO-56 PM ETo values, is the mean FAO-56 PM ETo, and E is ETo estimations by candidate empirical models. Multi-Criteria Group Decision-Making (MCGDM) Ranking Because individual performance indicators often provide conflicting assessments of model performance, a Multi-Criteria Group Decision-Making (MCGDM) approach was employed to obtain an overall ranking of the evaluated ETo models. The method integrates multiple statistical indicators into a single performance score while accounting for both accuracy and agreement measures. The six statistical indicators were classified into two groups. The coefficient of determination (R²), Nash–Sutcliffe Efficiency (NSE), and Index of Agreement (IoA) were treated as benefit criteria, where larger values indicate better model performance. Conversely, RMSE, NRMSE, and MAE were treated as cost criteria, where smaller values indicate superior performance. To eliminate differences in scale among the performance indicators, all criteria were normalized before aggregation47. For benefit criteria, normalization was performed using Eq. (8): where ri, j is the normalized value of criterion j for model I, xi, j is the value of criterion j for model i and max(xj) is the maximum value among all models for criterion j. For cost criteria, normalization was performed using Eq. (9): where min(xj) represents the minimum value of criterion j among all models. Following normalization, weights were assigned to each criterion. In this study, equal weights were initially adopted to avoid subjective prioritization among performance measures. The overall performance score of each model was then calculated using Eq. (10): where Si is the overall score of model i and wj is the weight assigned to criterion j. Models were ranked according to their final aggregated scores, with the highest score indicating the best overall performance. This approach provides a balanced assessment by simultaneously considering goodness-of-fit, predictive accuracy, and error magnitude, thereby reducing the influence of any single performance metric on the final ranking. Results and discussion Overall model performance The comparative evaluation of twelve ETo estimation models across seven Ethiopian stations revealed significant variations in performance, largely driven by local climatic conditions and model structures. The analysis, based on statistical metrics including the R2, NSE, IoA, RMSE, NRMSE, and MAE, identified clear differences in model reliability among stations. These findings are consistent with studies conducted in tropical and subtropical highland environments, which have shown that the performance of empirical ETo models is highly dependent on local climatic factors such as temperature, humidity, wind speed, and solar radiation availability. Similar conclusions have been reported by Sentelhas et al.11, Djaman et al.48, and Tabari et al.13, who found that no single empirical model performs consistently across diverse climatic conditions and emphasized the importance of site-specific validation prior to model application. Among the twelve models evaluated, the Enku model emerged as the strongest performer, demonstrating superior accuracy at five of the seven study stations: Bahir Dar, Gondar, Debre Tabor, Debre Markos, and Adigrat. Its peak performance was observed in Bahir Dar, where it achieved an IoA of 0.87, an NRMSE of 12.61%, and an RMSE of 0.49 mm/day as shown in Table 3. This strong performance is consistent with previous studies indicating that air temperature is a major determinant of ETo variability in humid and sub-humid highland environments. Furthermore, the correlation analysis conducted in this study showed a strong positive relationship between mean air temperature and FAO-56 PM ETo, supporting the effectiveness of temperature-based approaches in these climates. The Enku model’s minimal data requirement, relying solely on temperature observations, further enhances its applicability in data-scarce regions. However, while Enku performed well in transitional zones such as Debre Markos, it’s slightly lower IoA (0.72) compared with the WMO and Trabert models suggests that incorporating additional climatic variables, such as humidity and wind speed, may improve estimation accuracy in stations characterized by greater climatic variability. Table 3Statistical summary twelve models as compared to FAO-56 and seven climatic stations.MethodR2NSERMSENRMSEMAEIoAR2NSERMSENRMSEMAEIoABahir DarGondarBC0.24− 1.781.0426.770.910.520.57− 0.170.7720.120.910.52TW0.64− 4.991.5239.301.280.620.61− 2.741.3843.591.280.78Har0.62− 2.451.1629.821.10.580.400.270.6114.871.100.93Enku0.650.380.4912.610.370.870.540.120.671.170.370.99Mak0.07− 1.550.9925.640.820.510.10− 5.391.8199.220.820.08PRI0.00− 15.572.5365.352.330.250.01− 4.061.6191.872.330.00Turc0.64− 10.122.0753.542.010.340.62− 10.982.47146.850.080.60Abtew0.13− 4.971.5239.231.450.410.550.100.6884.661.450.08Dalton0.6− 6.641.7244.371.380.520.70− 19.063.288.321.380.04PEN0.49− 2.931.2331.831.040.60.59− 1.311.0969.491.040.24WMO0.71− 6.630.7144.341.620.470.79− 1.751.1983.171.620.09TARB0.67− 3.691.3534.751.190.620.79− 5.041.7685.801.190.06Debre TaborDebre MarkosBC0.33− 0.700.8724.530.750.520.610.080.7018.420.560.59TW0.42− 0.460.8022.720.640.750.72− 0.720.9625.140.770.80Har0.33− 0.590.8423.720.720.620.71− 37.074.50118.284.450.25Enku0.34− 0.270.7521.250.580.720.690.390.6617.280.550.84Mak0.20− 0.270.7521.250.640.660.47− 2.561.3836.171.120.59PRI0.02− 17.202.8480.332.700.240.17− 16.223.0379.542.250.34Turc0.04− 6.321.8050.931.650.320.75− 9.092.3260.912.260.37Abtew0.35− 3.371.3939.361.290.440.75− 0.380.8622.510.750.75Dalton0.24− 5.131.6546.601.410.480.80− 4.161.6643.551.140.68PEN0.32− 3.831.4341.361.240.490.78− 3.031.4638.481.320.62WMO0.01− 7.101.9053.581.660.300.82− 3.461.5440.491.400.61TARB0.21− 4.451.5543.951.330.440.82− 1.841.2332.281.040.76KomobolchaMekelleBC0.35− 2.040.9826.160.910.500.05− 0.031.2225.170.940.38TW0.18− 13.380.9926.450.790.570.20− 1.001.6934.971.400.56Har0.32− 1.200.8322.230.720.590.07− 12.234.3590.043.950.32Enku0.25− 0.670.7319.380.590.670.21− 1.111.7435.931.420.52Mak0.08− 4.461.3235.081.130.400.07− 0.341.3928.701.120.54PRI0.01− 14.092.1958.311.850.250.04− 3.202.4550.742.040.42Turc0.31− 24.852.8676.312.820.230.46− 6.503.2867.793.110.42Abtew0.27− 5.201.4037.391.320.400.400.370.9519.590.770.87Dalton0.37− 5.431.4338.051.180.550.41− 7.013.3870.021.870.50PEN0.34− 9.101.7947.701.560.400.52− 2.152.1343.971.890.59WMO0.38− 13.381.7956.922.020.330.77− 0.821.6233.421.370.80TARB0.32− 6.911.5842.211.360.430.76− 2.212.1444.341.700.74AdigratBC0.24− 1.781.0426.770.900.52TW0.64− 4.991.5239.301.280.62Har0.62− 2.451.1629.821.100.58Enku0.650.380.4912.610.370.87Mak0.07− 1.550.9925.640.820.51PRI0.00− 15.572.5365.352.330.25Turc0.36− 15.572.0753.542.010.34Abtew0.13− 10.121.5265.351.450.49Dalton0.60− 6.641.7239.231.380.58PEN0.49− 2.931.2344.371.040.60WMO0.71− 6.631.7244.331.620.47TARB0.67− 3.691.3534.751.190.62 Statistical summary twelve models as compared to FAO-56 and seven climatic stations. The Hargreaves model demonstrated notable success in Gondar, achieving a positive NSE of 0.27 and an NRMSE of 14.87%. This performance aligns with the model’s design for regions with moderate humidity and limited data inputs49. In contrast, the model performed very poorly in Debre Markos (NSE = -37.07). Verification of the calculations and input data indicated that this extreme negative NSE was not caused by computational errors but rather by substantial seasonal bias, particularly the consistent overestimation of ETo during humid and cloudy periods. The Hargreaves equation is fundamentally temperature-driven and does not explicitly incorporate the effects of humidity, wind speed, or solar radiation variability. Consequently, its performance may deteriorate in environments where these atmospheric factors play a significant role in regulating evapotranspiration6,11. Similar instability has been reported in tropical and humid highland environments where temperature-based methods tend to overestimate ETo under high cloud cover11. These findings suggest that while Hargreaves may be suitable for some sub-humid zones such as Gondar, local calibration is necessary before application in areas characterized by pronounced seasonal humidity variations. The Abtew model exhibited its best performance in Mekelle, achieving the highest Index of Agreement (IoA = 0.87) among the evaluated models and an acceptable NRMSE of 19.59%. This performance is likely associated with the predominance of solar radiation as a controlling factor of evapotranspiration in the arid conditions of Mekelle. Because the Abtew model is primarily radiation-based, it appears well suited to environments where evapotranspiration is strongly governed by available energy. Similar findings have been reported by McMahon et al.50, who noted that radiation-based methods often perform well in arid and semi-arid climates. In contrast, the model’s performance declined substantially in humid stations, such as Bahir Dar (R² < 0.20), suggesting reduced applicability in climates where additional atmospheric variables, including humidity and aerodynamic effects, exert greater influence on evapotranspiration processes. On the other hand, the Priestley–Taylor (PRI) model exhibited poor performance across most study stations, with NSE values below − 15 in six of the seven locations and very large prediction errors (e.g., NRMSE = 118.28% in Debre Markos). These results suggest that the model may be less suitable for the diverse climatic conditions represented in the study area. The observed inaccuracies may be related to the model’s simplified formulation and its limited representation of aerodynamic influences compared with more physically based approaches. Similar limitations have been reported in previous studies, where the Priestley–Taylor model showed reduced accuracy under conditions characterized by substantial variability in humidity, wind speed, and atmospheric demand7. Therefore, the poor performance observed in this study is consistent with previously documented challenges associated with applying the model outside the climatic conditions for which it was originally developed. The Turc model also performed poorly, with NRMSE values exceeding 50% at all stations. The largest errors were observed in Kombolcha (NRMSE = 76.31%) and Mekelle (NRMSE = 67.79%). This poor performance may be related to the model’s original development under humid climatic conditions, where evapotranspiration is largely controlled by available energy rather than aerodynamic factors. Because the Turc equation relies primarily on temperature and solar radiation and does not explicitly represent wind speed or atmospheric moisture demand, its applicability may be limited in environments exhibiting strong climatic variability, such as the humid highlands and arid zones included in this study. In contrast, the Enku model provided the most accurate estimates among the temperature-based methods in Kombolcha, suggesting that locally adapted formulations may better capture regional evapotranspiration dynamics. The Makkink model also exhibited low explanatory power, particularly in Bahir Dar and Adigrat (R² < 0.20). The model is based on the assumption that evapotranspiration can be estimated primarily from solar radiation through a simplified energy-balance relationship. However, because it does not explicitly account for aerodynamic effects, humidity variations, or vapor pressure deficit, its predictive capability may decrease in climates where these factors exert substantial influence on evapotranspiration2,6,49. In the humid highland environments of Bahir Dar and Adigrat, frequent cloud cover, variable atmospheric moisture conditions, and seasonal weather patterns may weaken the direct relationship between incoming radiation and reference evapotranspiration. Similar limitations of radiation-based approaches have been reported in previous studies, particularly when applied outside the climatic conditions for which they were originally developed11,48,51. Sensitivity analysis A sensitivity analysis based on correlation coefficients between FAO-56 PM reference evapotranspiration (ETo) and the principal meteorological variables (maximum temperature, minimum temperature, relative humidity, wind speed, and solar radiation) was conducted for each study station as shown in Table 4. The sensitivity analysis revealed substantial spatial variability in the dominant climatic controls of ETo. Temperature emerged as the most influential variable in most stations. Maximum temperature exhibited very strong positive correlations with ETo in Bahir Dar (r = 0.94), Debre Markos (r = 0.99), Adigrat (r = 0.95), Kombolcha (r = 0.88), and Gondar, indicating that thermal conditions are the primary drivers of evapotranspiration in these environments. Table 4Correlation coefficient between evapotranspiration and main parameters.SNStationTmaxTminRelative humidityWind speedSolar radiation1Bahir Dar0.940.330.680.910.622Gondar0.790.750.800.890.633Debre Tabor0.960.340.620.780.674Deber Markos0.990.360.740.880.675Kombolcha0.720.880.600.680.406Mekelle0.780.270.600.890.887Adigrat0.950.480.350.700.70 Correlation coefficient between evapotranspiration and main parameters. These findings provide quantitative support for the strong performance of the temperature-based Enku and Hargreaves models in the humid and transitional stations. The sensitivity analysis indicates that ETo variability in these locations is largely explained by temperature fluctuations, which is consistent with the formulation of these models. Conversely, the relatively poor performance of some radiation-based and mass-transfer models in these stations may be related to their dependence on variables that exert weaker control on ETo under these climatic conditions. The analysis also showed that relative humidity consistently exhibited moderate to strong negative correlations with ETo, ranging from approximately − 0.62 in Bahir Dar to -0.78 in Debre Tabor and Debre Markos. This finding supports the observed deterioration of several simplified models in humid environments, where atmospheric moisture conditions substantially influence evapotranspiration processes. It also helps explain why models that do not adequately account for humidity effects, such as Priestley–Taylor and Turc, frequently produced large estimation errors. Wind speed exhibited variable influence among stations. Strong positive correlations were observed in Bahir Dar (r = 0.91) and Debre Markos (r = 0.74), whereas weaker relationships were found in Adigrat (r = 0.35). These results help explain why mass-transfer models occasionally performed well in some transitional stations but lacked consistent performance across all climatic zones. The findings suggest that aerodynamic controls are important in certain locations but do not uniformly dominate ETo variability throughout northern Ethiopia. Solar radiation exhibited particularly strong correlations with ETo in the arid stations. In Mekelle, solar radiation showed the strongest relationship with ETo (r = 0.88), exceeding the influence of most other climatic variables. Similar positive relationships were observed in Adigrat (r = 0.70), Debre Tabor (r = 0.67), and Debre Markos (r = 0.67). These results provide direct quantitative support for the superior performance of the radiation-based Abtew model in Mekelle and confirm that available energy is a major control on evapotranspiration in arid environments. Regional performance patterns The regional performance patterns focus spatial variation in model performance across different geographical areas. Table 1 shows climatic zones classification and its justification to understand the model’s robustness and limitations across diverse geographical and climatic conditions. Humid highlands The humid highland stations (Bahir Dar, Debre Tabor, and Debre Markos) provided an opportunity to examine how local climatic controls influence ETo model behavior in environments characterized by relatively low vapor pressure deficits, moderate temperatures, frequent cloud cover, and distinct wet and dry seasons. Under such conditions, evapotranspiration is often energy-limited rather than aerodynamically controlled, meaning that variations in available energy play a greater role than variations in wind speed or atmospheric moisture demand2,52. The strong performance of the Enku model across these stations suggests that air temperature effectively integrates the combined influence of radiation and atmospheric energy availability in humid highland environments. This finding indicates that temperature can serve as a practical surrogate for more data-intensive meteorological variables where observational networks are limited. Similar observations have been reported in tropical highland regions where temperature-based approaches provided reliable ETo estimates despite their simplified formulations13,48. Although the three stations belong to the same broad climatic zone, notable differences in model behavior were observed. Debre Markos exhibited greater divergence among performance metrics than Bahir Dar and Debre Tabor. For example, the WMO and Trabert models produced relatively high coefficients of determination, indicating their ability to reproduce temporal variability in ETo, whereas the Enku model achieved superior agreement and lower prediction errors. This contrast suggests that different climatic drivers may dominate different aspects of evapotranspiration dynamics at the station. The location of Debre Markos within a transitional climatic setting may expose it to stronger seasonal shifts in humidity and wind conditions than the other humid highland stations, increasing the complexity of ETo estimation. A similar pattern was observed for the WMO model in Bahir Dar, where a relatively high R² was accompanied by lower agreement statistics. This result highlights the limitations of relying on a single performance metric. While R² measures the strength of the linear relationship between observed and estimated values, it does not evaluate systematic bias or overall predictive accuracy53. The discrepancy therefore indicates that the model was able to capture temporal fluctuations in ETo but was less successful in reproducing their magnitude. Such behavior is consistent with previous findings that aerodynamic-based methods may have limited applicability in humid environments where wind-driven processes contribute less to evapotranspiration variability than energy-related factors52. Transitional zones The transitional climate stations (Kombolcha and Gondar) exhibited more complex model responses than either the humid highlands or arid environments. These locations are characterized by intermediate hydroclimatic conditions where neither energy availability nor aerodynamic forcing consistently dominates evapotranspiration throughout the year. Instead, seasonal fluctuations in temperature, humidity, wind speed, and cloud cover create shifting controls on ETo, making model performance less predictable and more sensitive to formulation structure2. In Kombolcha, the divergence among performance metrics suggests that different models captured different components of the evapotranspiration process. While the Enku model achieved the highest agreement statistics, several mass-transfer models produced comparatively stronger linear relationships with the FAO-56 PM estimates. This contrast indicates that temperature-based approaches were more effective in reproducing the magnitude of ETo values, whereas mass-transfer models were better able to reflect temporal fluctuations associated with changes in atmospheric moisture demand and wind conditions. Such mixed behavior is expected in transitional climates, where the relative importance of energy-driven and aerodynamic processes varies seasonally and cannot be adequately represented by a single dominant climatic variable. Gondar displayed a more balanced climatic regime, with moderate humidity and relatively stable wind conditions. Under these circumstances, temperature remains strongly linked to evapotranspiration variability, while aerodynamic influences are sufficient to affect ETo without becoming the primary control. The comparatively good performance of the Hargreaves model suggests that simple temperature-based formulations can remain effective where climatic conditions are neither strongly energy-limited nor strongly water-limited. Similar findings have been reported in sub-humid and transitional environments, where temperature-based methods often provide reasonable estimates despite their simplified structure4,6. The results from both stations emphasize the inherent complexity of transitional climatic zones. Unlike humid or arid regions, where dominant evapotranspiration controls are relatively well defined, transitional environments are influenced by multiple interacting atmospheric drivers whose importance changes throughout the year. Consequently, model performance may vary substantially depending on the evaluation metric used and the specific climatic conditions prevailing during the study period. These findings highlight the importance of multi-metric model evaluation and local validation when selecting ETo estimation methods for climatically heterogeneous regions. Arid to semi-arid The arid and semi-arid stations exhibited evapotranspiration dynamics that differed markedly from those observed in the humid highlands and transitional zones. In these environments, low atmospheric humidity, high solar radiation, and elevated vapor pressure deficits increase the importance of available energy as the primary driver of evapotranspiration. Consequently, models that explicitly incorporate radiation tend to perform better than those developed primarily for humid or energy-limited conditions1,2. The strong performance of radiation-sensitive approaches in Mekelle suggests that evapotranspiration in dry environments is closely linked to variations in incoming solar energy. Under conditions of limited atmospheric moisture, a larger proportion of available energy is converted into latent heat flux, strengthening the relationship between solar radiation and ETo. This behavior is consistent with findings from arid and semi-arid regions worldwide, where radiation-based methods often provide reliable estimates because they directly represent the dominant physical controls on evapotranspiration31,54. Adigrat, although classified within the drier climatic group, displayed somewhat different characteristics. The relatively strong performance of temperature-based approaches suggests that air temperature effectively integrated the influence of radiation and atmospheric demand at this location. This result indicates that even within broadly similar climatic zones, local topographic influences, elevation differences, and seasonal weather patterns can modify the relative importance of evapotranspiration drivers. Such variability highlights the need to consider local climatic characteristics rather than relying solely on broad climate classifications when selecting ETo estimation methods. Compared with the humid and transitional zones, the arid stations exhibited a clearer relationship between climatic drivers and model behavior. In humid environments, interactions among radiation, humidity, cloud cover, and wind often complicate model performance, whereas in arid regions the dominant role of available energy produces more predictable responses. Nevertheless, the contrasting performance of models among stations demonstrates that even within dry climates, local calibration and validation remain essential. These findings reinforce the importance of matching model structure to the dominant evapotranspiration controls operating at each location, particularly in regions characterized by strong spatial climatic variability and complex topography. This station-specific evaluation demonstrates that Ethiopia’s pronounced topographic gradients and climatic heterogeneity create distinct microclimates that strongly influence empirical ETo model performance. The observed spatial variability agrees with previous studies showing that ETo models are highly climate-dependent and often require regional calibration for reliable application. For instance, Mekoya and Workneh26 found that the performance ranking of 17 PET/ETo methods varied considerably among different climatic regions of Ethiopia, while Almorox & Grieser55 showed that the performance of the Hargreaves–Samani model varies substantially among climatic zones and improves with local calibration. These findings underscore the need for climate-specific model selection and validation across Ethiopia’s diverse environments. Given the limited availability of complete meteorological datasets, carefully validated empirical models provide a practical alternative for improving ETo estimation, irrigation planning, and water resource management. Future research should focus on regional calibration of the best-performing models and assess their robustness under changing climatic conditions. Model ranking Ranking in each station The ranking of ETo models across the seven meteorological stations in northern Ethiopia revealed distinct patterns of model suitability closely linked to local climatic conditions (Table 5). The temperature-based Enku model consistently emerged as the top performer in most stations, including Bahir Dar, Debre Tabor, Kombolcha, and Adigrat, demonstrating its strong capability to capture evapotranspiration dynamics in humid highland and transitional climates. Hargreaves (Har) performed best in Gondar and frequently ranked among the top three models in other stations, reflecting its effectiveness under moderate climatic conditions and limited data availability. The radiation-based Abtew model achieved the highest ranking in the arid Mekelle station, emphasizing the dominant role of solar radiation in controlling evapotranspiration under water-limited conditions. Similar climate-dependent behavior of temperature- and radiation-based models has been reported in several regional and international studies, which found that model performance is strongly influenced by the dominant meteorological drivers of a given environment48. Table 5Ranking ETO models in each station.RankBahir DarGondarDebre TaborDebre MarkosKomobolchaMekelleAdigrat1EnkuHarEnkuBCEnkuAbtewEnku2MakEnkuTWEnkuHarWMOHar3BCBCMakAbtewBCMakMak4HarPENHarTARBDaltonBCBC5PENAbtewBCTWTWTWTARB6WMOWMOAbtewMakMakTARBPEN7TARBTurcPENPENAbtewPENTW8TWTWTARBWMOTARBEnkuDalton9AbtewMakDaltonDaltonPENTurcWMO10DaltonTARBTurcTurcWMODaltonAbtew11TurcPRIWMOHarPRIPRITurc12PRIDaltonPRIPRITurcHarPRI Ranking ETO models in each station. Other models, including Blaney–Criddle (BC), Makkink (Mak), and WMO, occasionally ranked among the top three performers in specific stations but showed less consistent performance across climatic zones. In contrast, Turc, Priestley–Taylor (PRI), and Dalton consistently ranked among the lowest-performing models across most stations, with Dalton frequently occupying the last position. Similar shortcomings of these models have been documented in environments characterized by strong climatic variability, where simplified assumptions regarding radiation, humidity, or aerodynamic controls may not adequately represent local evapotranspiration processes4,8,11. The ranking results further highlight the strong climate dependency of empirical ETo models. The dominance of the Enku model in humid highland and transitional stations, together with the higher performance of the Abtew model in the arid Mekelle station, indicates that the relative importance of temperature and solar radiation varies considerably across northern Ethiopia’s climatic zones. This finding agrees with previous studies showing that no single empirical model performs consistently across contrasting climates and that local validation is essential before model application26,55. Overall, the observed ranking patterns are consistent with the station-specific analyses presented above in the regional performance patterns section and further confirm the spatial variability of evapotranspiration controls across the study area. Sensitivity analysis of the model ranking To assess the robustness of the MCGDM ranking, a sensitivity analysis was conducted by assigning greater importance to error-related performance metrics (RMSE, NRMSE, and MAE) while reducing the weights assigned to agreement-based indicators (R², NSE, and IoA). Under the error-focused weighting scheme, RMSE was assigned a weight of 23%, NRMSE and MAE each received 22%, and R², NSE, and IoA were each assigned 11% as shown in Table 6. The resulting rankings were then compared with those obtained using equal weights. Table 6Ranking ETO models after sensitivity analysis.RankBahir DarGondarDebre MarkosDebre TaborKombolchaMekelleAdigrat1EnkuEnkuEnkuEnkuEnkuAbtewEnku2HarHarBCTWHarWMOHar3WMOBCAbtewMakBCBCBC4BCAbtewTWHarTWMakTARB5PENTurcTARBBCDaltonTARBMak6TARBPENPENAbtewAbtewTWPEN7MakWMODaltonPENTARBEnkuTW8TWTWMakTARBMakPENDalton9DaltonTARBWMODaltonPENPRIWMO10AbtewMakTurcTurcWMODaltonAbtew11TurcDaltonPRIWMOPRITurcTurc12PRIPRIHarPRITurcHarPRI Ranking ETO models after sensitivity analysis. The sensitivity analysis revealed that the highest-ranked models were generally stable across weighting schemes. The Enku model retained its first-place ranking in Bahir Dar, Debre Markos, Kombolcha, and Adigrat, and improved from second to first place in Gondar. Similarly, the Abtew model remained the highest-ranked model in Mekelle under both weighting scenarios. These results indicate that the higher performance of Enku in humid and transitional climates and Abtew in the arid environment is not an artifact of the selected weighting scheme but reflects their consistently strong performance across multiple evaluation criteria. Several models exhibited moderate ranking changes when error measures were given greater emphasis. In Bahir Dar, the Hargreaves model improved from fourth to second place, while Makkink declined from second to seventh. This suggests that although Makkink performed relatively well according to agreement-based statistics, its prediction errors were larger than those of Hargreaves. A similar pattern was observed in Gondar, where Enku overtook Hargreaves as the highest-ranked model, indicating that Enku produced smaller overall errors despite comparable agreement statistics. The most notable ranking shifts occurred in Debre Markos and Debre Tabor. In Debre Markos, the Blaney–Criddle model improved from first to second place while Enku rose from second to first. More importantly, the Hargreaves model dropped from eleventh to twelfth place, confirming that its previously poor performance became even more evident when error measures were prioritized. In Debre Tabor, Thornthwaite improved from second to first place while Enku remained highly ranked, suggesting that both models captured ETo dynamics reasonably well, but Thornthwaite generated lower overall prediction errors. Among the mass-transfer models, WMO generally became more competitive under the error-focused scenario. For example, it improved from sixth to second place in Mekelle and from sixth to third place in Bahir Dar. This indicates that although WMO did not always achieve the highest agreement statistics, it often produced comparatively small estimation errors. Conversely, Dalton consistently remained among the lower-ranked models, confirming its limited suitability across the study region regardless of the weighting scheme. The lower-ranked models were also relatively stable. Priestley–Taylor (PRI) remained the lowest-ranked model in most stations, while Turc consistently occupied one of the bottom positions. The persistence of these poor rankings under both weighting schemes suggests that their inferior performance is systematic rather than a consequence of the selected evaluation criteria. Conclusion This study evaluated twelve empirical reference evapotranspiration (ETo) models across seven climatically diverse stations in northern Ethiopia using the FAO-56 PM method as the reference standard. The results demonstrate that empirical ETo model performance is highly dependent on local climatic conditions and that no single model can be reliably applied across all agro-climatic zones. Among the evaluated methods, the temperature-based Enku model showed the most consistent performance, particularly in humid and sub-humid highland environments, while the radiation-based Abtew model was more suitable for arid conditions such as Mekele. Conversely, the Priestley–Taylor, Turc, and Makkink models generally exhibited poor performance and are not recommended without local calibration. These findings highlight the importance of climate-specific model selection for improving ETo estimation in data-scarce regions. Several limitations should be acknowledged. The analysis relied on the FAO-56 PM method as a benchmark rather than direct lysimeter measurements, which may introduce reference-related uncertainties. In addition, the study was based on available station records and temporal aggregation of meteorological data, potentially masking short-term climatic variability. The performance of empirical models may also be influenced by data quality, missing observations, and local microclimatic conditions that were not explicitly considered. Future research should focus on validating these models against field-measured evapotranspiration data, assessing their performance under changing climatic conditions, and developing locally calibrated or hybrid approaches. Such efforts will improve the reliability of ETo estimation and support more effective irrigation planning, water resource management, and climate adaptation strategies in Ethiopia. Author contributions **Hana Mekonnen** : data collection, methodology, modeling validation, and writing the draft. **Ephrem Yetbarek** : supervision, conceptualization, writing, review, and editing. **Chalachew A. Mulatu** : review and editing. Funding This research received no external funding. Data availability The meteorological data used in this study were obtained from the Ethiopian Meteorological Institute (EMI) under data-use permission. Data sets used are available from the corresponding author on reasonable request. Declarations Competing interests The authors declare no competing interests. Ethical statement The authors state that the research was conducted according to ethical standards. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Data Availability Statement The meteorological data used in this study were obtained from the Ethiopian Meteorological Institute (EMI) under data-use permission. Data sets used are available from the corresponding author on reasonable request.