Comparative study of evapotranspiration from the SWAT model and MODIS-derived remote-sensing data in two climatic zones in Egypt

The hydrological cycle includes precipitation, surface runoff, evapotranspiration (ET), and surface infiltration, with the ET mechanism returning about 60% of precipitation to the atmosphere (Parajuli et al. 2018). ET stands for the transfer of water vapor from soil evaporation and plant transpiration to the atmosphere (Nguyen & Kappas 2015; Ghiat et al. 2021) and is thus a crucial component of the water cycle and water balance in agriculture (Wanniarachchi & Sarukkalige 2022). Precipitation and ET exchange energy and moisture between the soil and the atmosphere, making them an essential part of the water cycle (Martel et al. 2018; Dimitriadou & Nikolakopoulos 2021). It has been shown that ET is much more difficult to measure than precipitation or river discharge (Tobin & Bennett 2017).

Accurate estimation of ET is crucial for environmental sustainability, crop water-demand assessment, irrigation system planning, and effective water resource management and agricultural watersheds (Tabari et al. 2013; Parajuli et al. 2022). Alongside ET observations by ground-based devices, there are several estimation methods including coupled-process models, remote sensing, water-balance-based methods, meteorologically based regression methods, empirical formulae, and machine learning (Liu et al. 2019; Yin et al. 2020; Wang et al. 2021; Wanniarachchi & Sarukkalige 2022). Due to the lack of surface observations for ET in various countries, especially in developing countries, different authorities and researchers use these estimation methods. Talib et al. (2021) used machine-learning algorithms such as the random forest (RF) method to predict daily ET for corn, soybeans, and potatoes for farmland in the Midwestern USA from 2003 to 2019. Granata (2019) investigated ET models such as regression tree, RF, and support vector regression in central Florida. Yamaç & Todorovic (2020) used three machine-learning methods to estimate daily potato ET: k-nearest neighbor, adaptive boosting, and artificial neural networks (ANNs). Traore et al. (2010) investigated the performance of ANNs in reference evapotranspiration (ET-ref) modeling in Burkina Faso by comparing the Penman–Monteith (PM) model with the Hargreaves (HR) model. Also, Ghiat et al. (2021) discussed the main models used to estimate ET, such as the PM, Stanghellini, Priestley–Taylor (PT), and HR models.

The Soil and Water Assessment Tool (SWAT) is one of the most widely used ecohydrological models to simulate hydrological and biophysical processes under a variety of climatic and management conditions (Alemayehu et al. 2017). SWAT has been used in many places around the world to simulate hydrological processes such as runoff, ET, and vegetation dynamics such as leaf area index, crop yield, and biomass (Ma et al. 2019), and assess the impact of future climate change on the water balance based on the amount of water available for agriculture (Jung & Kim 2018). Many studies have used SWAT to calculate ET in African countries (Alemayehu et al. 2016; Aouissi et al. 2016; Poméon et al. 2018; Mengistu et al. 2019; Odusanya et al. 2019; Dile et al. 2020; Zettam et al. 2020; Akoko et al. 2021; Abdulkadir et al. 2022; Bennour et al. 2022; Muthee et al. 2023) and in many other countries around the world (Kavian et al. 2017; Chen et al. 2020; Husain et al. 2020; Ferreira et al. 2021; Lee et al. 2021; López-Ramírez et al. 2021; Raja et al. 2022; Sholagberu et al. 2022; Rane & Jayaraj 2023). The SWAT model uses PM, PT, and HR methods for ET estimation based on different climate data and actual land-surface conditions (Parajuli et al. 2022).

ET data derived from remote-sensing data cannot be considered accurate for a number of reasons, including misinterpretation of models, the nature of the inputs, varying temporal and spatial resolution, and data coverage (Ferguson et al. 2010; Ayana et al. 2019; Dile et al. 2020). Therefore, ET derived from remote sensing data should be carefully examined and compared with other ET data sources before being used for water resources management (Parajuli et al. 2022). The Moderate Resolution Imaging Spectroradiometer (MODIS) includes Aqua and Terra sensors and provides continuous ET estimates with a spatial resolution of 250 m and a temporal resolution of eight days (Dash et al. 2021). The simplified surface energy balance operational method (SSEBop) developed by Senay et al. (2013) is used to calculate the global monthly ET product (Ayyad et al. 2019) based on land surface temperature from MODIS and reanalysis data (Zhuang et al. 2022). To improve our understanding of ET estimation using these different algorithms, the relative accuracy of ET data obtained using these different methods should be extensively explored (Parajuli et al. 2022).

Several studies have been conducted worldwide to compare ET estimates from SWAT models and satellite-derived data (MODIS) in different global regions and watersheds, including Australia (Dile et al. 2020), China's Lijiang River Basin (Yao & Mallik 2022), and Mississippi's Big Sunflower River Watershed (Parajuli et al. 2022), while the current study compares different methods in two regions in Egypt with different climatic and geographical conditions, which differ from other study areas where these methods have been compared. In Egypt, numerous studies have been conducted for spatiotemporal assessment of ET using different techniques, models, or satellite remote sensing over different areas (Khalil 2013; Ayyad et al. 2019; Omar et al. 2019; Fawzy et al. 2021; Anwar et al. 2022; Sobh et al. 2022; Eltarabily et al. 2023). The main objective of these studies is to analyze the spatiotemporal distribution of ET or to investigate the performance of current satellite-based ET products and to evaluate irrigation efficiency or to investigate the impact of climate change on ET in Egypt.

Thus, our study is currently the only one that uses the SWAT model application to simulate ET and compare it with MODIS-derived ET (SSEBop and MOD16A2) in the South-Delta (northern Egypt) and KomOmbo (southern Egypt) zones because they contain considerable agricultural areas and differ in their geographical and climate features. Therefore, this study aims to (i) simulate the ET using the SWAT model and generate monthly, seasonal, and annual ET time-series over the South-Delta and KomOmbo zones during the period from April 2013 to December 2020; (ii) assess and compare the ET simulated from the SWAT model and derived from MODIS (SSEBop and MOD16A2) for one-month, seasonal, and annual timescales over the two study zones; and (iii) analyze and quantify the extent of agreements or differences in the simulated ET values from the SWAT model and those derived from MODIS in the two study zones.

To avoid the cloud cover that covers large areas of the northern delta in Egypt and obscures the nature of the land beneath it, the South-Delta is chosen as a first study zone, which is sometimes exposed to cloud cover. The South-Delta zone is located between latitudes 30.01° N and 30.92° N and longitudes 30.25° E and 32.25° E with an estimated area of about 8,156.91 km², as shown in Figure 1. The second study zone is KomOmbo, which is located in Aswan Governorate in southern Egypt between latitudes 24.32° N and 24.65° N and longitudes 32.87° E and 33.45° E with an estimated area of about 925.95 km², as shown in Figure 1.

There is almost no rainfall in the KomOmbo zone throughout the year, and it does not exceed 0.5 mm during the spring and winter seasons (Figure 2(c)), where hyperarid climate conditions prevail in southern Egypt with higher temperatures compared with northern Egypt (Morsy et al. 2017; Yassen et al. 2020). Also, Figure 2(c) shows that the highest monthly values of T_max (>42 °C), T_min (>25 °C), and Rs (>28 MJ/m²) in the KomOmbo zone are found in the summer season and their annual values are 35 °C, 18 °C, and 23 MJ/m², respectively. Furthermore, the annual RH and WND in KomOmbo are 26% and 3 m/s, while the highest RH (>44%) is found in the winter season and the maximum WND (>3.5 m/s) is detected in the summer season, as shown in Figure 2(d).

The two study zones cover large areas, so the area average for ET data is obtained from MOD16A2 and SSEBop derived from MODIS with 1 km spatial resolution (sufficient to investigate the regional and climatic variations) during the study period (2013–2020). These ET data are freely accessible on the Climate Engine Web Application at https://app.climateengine.org/climateEngine. The ET data are available as total monthly values from the SSEBop dataset and summed over eight-day intervals from the MOD16A2 dataset, which is converted to total monthly values using Microsoft Excel. The total seasonal and annual ET values are computed from the derived monthly total obtained from MOD16A2 and SSEBop datasets for further statistical analysis.

SWAT is a physically based, semi-distributed hydrological model based on the water balance concept. SWAT simulates watershed processes such as ET, runoff, crop growth, and nutrient and sediment transport using meteorological, soil, and land cover data, as well as practical land-management strategies (Abiodun et al. 2018). The SWAT model uses the PM, PT, and HR methods to simulate ET based on different weather variables and actual land-surface conditions (Parajuli et al. 2022) as follows.

The SWAT model uses daily data for rainfall in mm, maximum temperature (T_max) in °C, minimum temperature (T_min) in °C, solar radiation (Rs) in MJ/m², RH in %, and WND in m/s as input weather parameters to compute ET. The SWAT weather data generator is used to prepare the SWAT input weather files (Bera et al. 2020), which require weather data at different sites within the study area. Thus, the required daily weather data is obtained from the NASA-POWER agroclimatology dataset (https://power.larc.nasa.gov/data-access-viewer/) at nine sites in the South-Delta zone and five sites in the KomOmbo zone (Table 1) during the study period (2013–2020). Also, the monthly value of these weather parameters is computed from the obtained daily data to analyze and interpret the climate patterns over the two study zones (Figure 2).

The Shuttle Radar Topography Mission (SRTM; Farr et al. 2007) digital elevation model (DEM) data is downloaded from the USGS (https://earthexplorer.usgs.gov/) with 1 arc-second (30 m) spatial resolution over the two study zones. This DEM data is used within the SWAT model to derive the sub-basin's topographic parameters, such as area and slope, and to extract a watershed delineation map. The elevation ranges in South-Delta from −60 to 39 m (Figure 1(a)), while it ranges from 68 to 202 m in KomOmbo, as shown in Figure 1(d).

Soil data for the two study zones were obtained through field study as well as the global soil database developed by the Food and Agriculture Organization of the United Nations (FAO), which can be accessed via the following link: https://www.fao.org/land-water/land/land-governance/land-resources-planning-toolbox/category/details/en/c/1026564/. The dominant soil-type in South-Delta is clay, clay loam, sandy loam, and loam (Figure 1(c)), while the dominant soil-type in KomOmbo is loam, sandy loam, and clay loam (Figure 1(f)).

Changes in land use land cover (LULC) are connected to global climate change, resettlement programs (Regasa et al. 2021), urbanization, and vegetation changes (Srivastava et al. 2012) and have an impact on solar radiation, wind velocity, albedo, evaporation, transpiration, and water budget (Morsy & Aboelkhair 2021). Landsat 5 and 8 satellite-images are obtained with 30 m spatial resolution from the USGS website (https://earthexplorer.usgs.gov/) to identify the LULC types in the two study zones using the supervised classification and maximum likelihood method in ENVI 5.3 software following the methodology from Morsy & Aboelkhair (2021). Accordingly, the two study zones are classified into four LULC categories: (i) agricultural lands, (ii) urban areas, (iii) water bodies, and (iv) desert areas, as shown in Figure 1(b) and 1(e). The produced LULC maps are prepared as raster images to be suitable input for the SWAT model within ArcGIS 10.8.

The monthly reference evapotranspiration (ET₀) is computed following the FAO56 PM and HR methods using the basic irrigation scheduling application (BIS; Snyder et al. 2014) on MS Excel based on the NASA-POWER monthly weather data during the study period (April 2013–December 2020). The main reason for calculating ET₀ is not to compare its values with ET from SWAT and MODIS but rather to compare and ensure that their behavior (increase/decrease) changes similarly based on interannual changes in weather parameters.

R is used to investigate the strength, closeness, and direction of the correlation between the simulated ET from SWAT and the derived ET from MODIS. R has the range between 0, which implies no correlation, and ±1, which indicates a perfect direct/indirect correlation. Also, R is used to assess the temporal (monthly, seasonal, and annual) correlation between the ET (SWAT, BIS, and MODIS) and the selected weather variables (rainfall, T_max, T_min, Rs, RH, and WND) during the study period over the two study zones. The strength of R between ET and weather variables is classified according to Schober et al. (2018) into four categories that are: (i) very strong correlation (R ≥ ±0.9); (ii) strong correlation (±0.7 ≤ R < ±0.9); (iii) moderate correlation (±0.4 ≤ R < ±0.7); and (iv) weak correlation (R < ±0.4).

The coefficient of determination (R²) is the square of the coefficient (R) and ranges between 0 and 1. R² is used to find out how the differences in the derived ET from MODIS can be explained by (fitted on the regression line) the simulated ET from SWAT in the two study zones.

Finally, the box-and-whisker plot statistical chart in MS Excel is used for further analysis of monthly, seasonal, and annual ET and their spatiotemporal variability across the study period from both SWAT and MODIS over the two study zones. If the median value of ET is close to its mean value in the box diagram, this means that the data are distributed symmetrically, while the data are asymmetrically distributed and contain extreme values if its median is much higher or lower than its mean (Jenifer & Jha 2021). The long box (large interquartile range) indicates more spread of the middle 50% of ET data, while the short box (small interquartile range) refers to less spread of the middle 50% of ET data (Mabrouk et al. 2022).

In this section, the monthly, seasonal, and annual ET over the two study zones in Egypt will be analyzed and discussed based on the simulation from the SWAT model, estimated from the BIS application, and retrieved from MODIS (SSEBop and MOD16A2).

The maximum, minimum, mean, and CV% values of monthly ET from SWAT (PM, HR, and PT) and MODIS (SSEBop and MOD16A2) over South-Delta and KomOmbo are demonstrated in Table 2. The monthly maximum value of the simulated ET from SWAT methods over South-Delta and KomOmbo is lower than that derived from SSEBop by about −67 and −93 mm, respectively, while it is higher than that derived from MOD16A2 by about 43 and −5 mm, respectively. The same results are detected for the monthly mean value of ET, where the SWAT values are less than that from SSEBop by about −35 mm in the two study zones and more than that from MOD16A2 by 13 mm in South-Delta and 4 mm in KomOmbo. The monthly minimum value of ET simulated from SWAT is nearly equal to that derived from SSEBop with a difference around zero in South-Delta and 2 mm in KomOmbo, while it is higher than that derived from MOD16A2 by about 3 mm in South-Delta and 10 mm in KomOmbo. Moreover, the variation of monthly ET from SWAT (CV = 44%) is lower than that of SSEBop (CV = 56.78%) and higher than that of MOD16A2 (CV = 34.54%) in South-Delta, while it is lower than that of both MOD16A2 (CV = 76.49%) and SSEBop (CV = 60.04%) in KomOmbo. This result reveals that the simulated ET from SWAT has intermediate values between SSEBop and MOD16A2, which indicates that the SWAT model may achieve acceptable results for ET simulation in the two study zones but still requires further evaluation against observed ET data.

The SWAT (PM, HR, and PT) model and MODIS-derived (SSEBop and MOD16A2) are different ET simulation and estimation tools that use different inputs. SWAT is based on daily data for T_max, T_min, Rs, RH, and WND along with actual land-surface conditions (soil, land use, and slope) to simulate ET, while MODIS is based only on remotely sensed vegetation cover and climate data to estimate ET within each revisit period (Khan et al. 2018; Parajuli et al. 2022). Thus, none of them can be considered as an actual ET observation method to evaluate the estimated ET from the other methods.

Table 3 shows the values of the statistical metrics (MBE, MBE%, R, and R²) between the simulated monthly ET (mm) from SWAT and that derived from MODIS over the South-Delta and KomOmbo zones. It is found, as previously discussed from Table 2, that the SWAT is estimates lower ET than the SSEBop with MBE of about −35 mm (−41%) in South-Delta and −35 mm (−66%) in KomOmbo, while it estimates higher ET than the MOD16A2 with MBE of about 13 mm (34%) in South-Delta and 4 mm (28%) in KomOmbo. The highest R values reveal a high correlation between ET simulated from SWAT (PM, HR, and PT) and derived from MODIS (SSEBop and MOD16A2) in the two study zones. The correlation (R) between the simulated ET from SWAT with those derived from SSEBop and MOD16A2 exceeds 0.85 and 0.75, respectively, in South-Delta and is higher than 0.87 and 0.81, respectively, in KomOmbo. Also, R² values indicate that more than 70% (55%) of the derived ET from MODIS can be explained by (fitted on the regression line) the simulated ET from SWAT in the South-Delta (KomOmbo).

The mean values of the monthly ET in South-Delta over the study period (2013–2020) range from the highest value in July from both SWAT and MODIS-derived to the lowest value in November, December, and May from SWAT, SSEBop, and MOD16A2, respectively. In KomOmbo, the highest mean of monthly ET is detected in July from SSEBop, May from MOD16A2, and June from SWAT, while the lowest mean ET is found in December from SSEBop and SWAT, and January from MOD16A2. The maximum and minimum monthly ET values in South-Delta from both SWAT and MODIS-derived occur in the same months as the highest and lowest mean, respectively. The same pattern is observed in KomOmbo except that the maximum monthly ET from SSEBop occurs in June, and the minimum monthly ET from MOD16A2 and SWAT-PM occurs in December and January, respectively. The shortest boxes of monthly ET in South-Delta are found in March from SSEBop, November from MOD16A2, and December from SWAT with relatively small variations (CV = 3.28%–4.42%), while the most-extended boxes are detected in April from SSEBop, and July from MOD16A2 and SWAT with relatively high variations (CV = 7.96%–8.93%). In KomOmbo, the shortest boxes of monthly ET are found in October from SSEBop, January from MOD16A2, June from SWAT-PM, November from SWAT-HR, and August from SWAT-PT with relatively small variations (CV = 8.50%, 41.38%, 3.88%, 1.92%, and 0.94%, respectively). Also, the most-extended boxes of monthly ET in KomOmbo are detected in July from SSEBop, April from MOD16A2, March from SWAT-PM, March from SWAT-HR, and May from SWAT-PT with relatively high variations (CV = 8.42%, 46.58%, 11.07%, 6.88%, and 5.03%, respectively). Furthermore, the SSEBop produced the highest monthly ET with high variations as compared with SWAT and MOD16A2, because each of them depends on different input parameters (Khan et al. 2018). Also, the monthly ET derived from MODIS and simulated from SWAT during summer months is often higher than in other months in the two study zones. This is consistent with the findings of Parajuli et al. (2022), which showed that the ET was highest in June, July, and August and lowest in other months. Generally, the monthly ET is higher in South-Delta than in KomOmbo (decreases southward), except in May from MOD16A2.

Tables 5 and 6 present the descriptive statistics of seasonal and annual ET values, respectively, from SWAT and MODIS over the two study zones during the study period. The seasonal and annual ET values derived from SSEBop are higher with high variation than those simulated from SWAT (PM, HR, and PT) and derived from MOD16A2 in South-Delta and KomOmbo. The seasonal (annual) maximum ET from SWAT is lower than that derived from SSEBop by about −220 (380) mm, but it is higher than that derived from MOD16A2 by about 98 (180) mm in South-Delta. Also, the maximum seasonal ET from SWAT in KomOmbo is lower than that from SSEBop and MOD16A2 by about 243 and 10 mm, respectively, while the maximum annual ET from SWAT is lower than that from SSEBop (−430 mm) and higher than that from MOD16A2 (40 mm). The minimum and mean values of seasonal and annual ET follow the same situations as their maxima, except for some values in KomOmbo, where the seasonal minimum ET from SWAT is slightly lower than that from SSEBop (∼2 mm) and the seasonal minimum and mean ET from SWAT are higher than that from MOD16A2 by about 35 and 11 mm, respectively. Furthermore, the variation of seasonal and annual ET from SWAT in South-Delta is higher than that from SSEBop and MOD16A2, except for the seasonal from SSEBop, while the seasonal and annual variations of ET from SWAT in KomOmbo are lower than those from SSEBop and MOD16A2. The results of seasonal (Table 5) and annual (Table 6) ET reveal that the simulated ET from SWAT has intermediate values between SSEBop and MOD16A2 in the two study zones, which indicates that the SWAT model may achieve reliable results but still requires a further evaluation.

The MBE, MBE%, R, and R² statistical metrics are also performed to assess and compare the differences between the simulated seasonal and annual ET from SWAT and those derived from SSEBop and MOD16A2, as shown in Table 7. SWAT estimates lower seasonal and annual ET values than SSEBop and MOD16A2 in the two study zones. The MBE (MBE%) of the seasonal ET between SWAT and SSEBop is about −105 mm (−41%) in South-Delta and −106 mm (−66%) in KomOmbo, while SWAT estimates higher seasonal ET than MOD16A2 with the MBE (MBE%) of about 38 mm (34%) in South-Delta and 11 mm (28%) in KomOmbo. For the annual ET, SWAT estimates lower ET than SSEBop with MBE (MBE%) of −404 mm (−41%) and −400 mm (−66%) in South-Delta and KomOmbo, respectively, while it estimates higher ET than MOD16A2 with MBE (MBE%) of 153 mm (34%) and 50 mm (28%) in South-Delta and KomOmbo, respectively. Since the annual ET values are the sum of the seasonal values, which are also the sum of the monthly values, it is found that the monthly, seasonal, and annual ET from SWAT (PM, HR, and PT) and from SSEBop and MOD16A2 have the same MBE% in each study zone. The highest correlation (R) values for seasonal ET (77%–99%) reveal a high correlation between ET simulated from SWAT (PM, HR, and PT) and derived from MODIS (SSEBop and MOD16A2) in the two study zones. The R for annual ET between SWAT (PM, HR, and PT) and MODIS (SSEBop and MOD16A2) is positive and relatively small, which may be due to the seven-year sample size (2014–2020), which in turn reduced the R² values. The R² values for the seasonal ET indicate that about 99% of the derived ET from SSEBop and more than 77% of the derived ET from MOD16A2 can be explained by (fitted on the regression line) the simulated ET from SWAT in South-Delta; similarly, about 91% of the derived ET from SSEBop and more than 92% of the derived ET from MOD16A2 can be explained by the simulated ET from SWAT in KomOmbo.

The highest seasonal mean of ET from SWAT and MODIS-derived occurred in the summer in the two study zones, while the lowest mean value is found in the winter from SWAT and MODIS-derived in the two study zones except for the South-Delta from MOD16A2, which is in the autumn (Figure 7 and Table 8). The results of the highest and lowest seasonal values of ET are consistent with the results of the monthly ET analysis. The maximum and minimum seasonal ET values in the two study zones from both SWAT and MODIS-derived occurred in the same seasons that have the highest and lowest mean, respectively. The shortest boxes of seasonal ET are found in autumn from SWAT in the two study zones, from MOD16A2 in South-Delta, and from SSEBop in KomOmbo, while they are found in South-Delta from SSEBop in spring and in KomOmbo from MOD16A2 in winter. The longest boxes are found in spring from SWAT and MOD16A2 and in summer from SSEBop in the two study zones.

The highest variation of the seasonal ET in South-Delta is found in spring from SWAT and MOD16A2 (CV = 7%–8%) and in autumn from SSEBop (CV = 8.5%), while in KomOmbo the highest seasonal variation is found in winter from SWAT (CV = 4%–11%) and MODIS-derived (13%–54%). The lowest seasonal ET variation is found in autumn from SWAT (CV = 1%–3%) and MOD16A2 (CV = 4%–13%) in the two study zones, while it is found from SSEBop (CV = 2%–5%) during spring in South-Delta and during summer in KomOmbo. The highest seasonal ET is produced from SSEBop with high variations as compared with SWAT and MOD16A2. Also, the seasonal ET from SWAT and MODIS-derived in the two study zones is higher in the summer season than the other seasons, which coincides with Parajuli et al. (2022), and it is higher in South-Delta than in KomOmbo (decreases southward).

Moreover, the maximum and minimum annual ET values (Table 9) in South-Delta from SWAT are found in 2015 and 2019, respectively, with an annual mean of about 600 mm, box-length interquartile range (IQR) of about 40 mm, and variation (CV) of about 5%. The maximum annual ET value in South-Delta from SSEBop and MOD16A2 is found in 2014 and 2020, respectively, and their minimum annual ET is found in 2018, while their mean annual ET (IQR) are 1,004 and 446 (34 and 15) mm, respectively, with a variation of about 3%. For KomOmbo, the maximum annual ET from SSEBop and SWAT-PM is found in 2018, from MOD16A2 and SWAT-PT is found in 2014, and from SWAT-HR is found in 2015. The minimum annual ET from MOD16A2 and SWAT-HR occurred in 2016, from SSEBop in 2014, from SWAT-PM in 2019, and from SWAT-PT in 2015. The annual mean ET values from SSEBop, MOD16A2, SWAT-PM, SWAT-HR, and SWAT-PT are 630, 175, 243, 216, and 190 mm, respectively, while their IQRs (variations) for annual ET are 33 (4%), 14 (7%), 1.4 (3%), 3 (1.2%), and 4 (3%) mm, respectively.

In the South-Delta region, the monthly and seasonal ET simulated by SWAT (PM, HR, and PT) shows a moderate to strong correlation, the ET estimated by BIS (PM and HR) and the ET derived by SSEBop show a strong to very strong correlation, and the ET derived by MOD16A2 shows a weak correlation with the weather parameters. This implies that the estimated ET from BIS has the highest correlation, followed by SSEBop and SWAT, while the derived ET from MOD16A2 has the lowest correlation. In KomOmbo, the monthly and seasonal ET from SWAT and MODIS-derived mostly have very strong correlation with the weather variables, except a weak negative correlation with rainfall. For the annual correlation in the two study zones, the ET mostly has a varied (positive/negative) weak correlation with the weather variables, and it mostly has a positive weak correlation with rainfall that matches the results of Liu et al. (2022). Also, the annual ET from BIS sometimes has a moderate to strong correlation with the weather variables followed by SWAT and sometimes (in South-Delta) by MOD16A2. This weak correlation for annual ET values with weather parameters may be due to the small number of years (seven) from 2014 to 2020.

This study encountered a few limitations, which are clearly represented in the lack of surface observations of ET in the different climatic zones of Egypt. As a result, the simulated ET values from SWAT and derived from MODIS are subject to some uncertainties and thus still need to be further evaluated against the observed ET values. In addition, the spatial distribution of ET could not be easily obtained from the SWAT model compared with that obtained from the MODIS-derived model.

The study assessed ET in two different climatic zones in Egypt using three remote-sensing-based techniques. It was found that the SWAT model provided ET values that often fell between MODIS and SSEBop. Although the SWAT model had a lower coefficient of variation, indicating consistency, further validation against observed ET data is required. ET values were higher in South-Delta than in KomOmbo, suggesting a north–south gradient in ET in Egypt.

In addition, the study showed that ET correlates differently with meteorological parameters depending on the region and time scale (monthly, seasonal, and annual). The correlation patterns varied between the two zones, with generally stronger correlations found in South-Delta. Finally, our future work will focus on evaluating the accuracy of the SWAT model and MODIS in areas around Egypt where surface ET monitoring data is available to determine which method is best for estimating ET.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.