Determination of rainfed wheat agriculture potential through assimilation of remote sensing data with SWAT model case study: ZarrinehRoud Basin, Iran

The increasing population growth in the world results in, amongst other things, the use of additional resources such as food needs. Therefore, considering the limited water resources for food production, the optimal management of water resources to achieve the sustainable development and production of food and agricultural products is very important. According to a study by the International Water Management Institute (IWMI), in 1950, 12 countries with a population of approximately 20 million were facing water shortages and in 1960, 26 countries with a population of approximately 300 million were facing similar water issues. Forecasts show that by 2050, approximately 65 countries with a population of more than seven billion people will face severe water shortages. Iran is also one of the arid and semi-arid regions of the world and the agricultural sector accounts for about 94% of total water consumption. Given the 6.8-fold growth of the country's population in less than 80 years, the reduction of groundwater reserves in most aquifers and the declining water levels in rivers and lakes such as Urmia Lake, the issue of fresh water resources shortage and water resources management seems crucial (Kaviani et al. 2011; Rostami & Raeini-Sarjaz 2016).

Meanwhile, 80% of the world's agricultural lands is cultivated as rainfed, which provides about 60% of human food needs and is a means of livelihood for farmers (Rockström et al. 2003). Although the importance of rainfed agriculture varies from place to place, it still produces the largest share of food for poor people in developing countries (Wani et al. 2003). In sub-Saharan Africa, 95% of arable lands are rainfed, in Latin America about 90%, in South Asia about 60%, and in North Africa and Near East about 75% of the land is rainfed (FAOSTAT 2005). According to statistics published by the Agriculture Statistics (2019), the total area of irrigated lands is 40% and rainfed lands account for 60% of the total area under annual cultivation (which is approximately equal to 15.5 million hectares).

Rainfed agriculture is an appropriate alternative to irrigated agriculture due to its compatibility with nature and long-term sustainability (Biradar et al. 2009). Therefore, recognizing and grading the potential of lands for rainfed agriculture can help in the optimal use of water resources. Some of the most widely used methods to potentialize lands for rainfed agriculture can be divided into four general categories: 1. Climatic methods, 2. Agricultural methods, 3. Hydrological methods, and 4. Hybrid methods. One of the hybrid methods that has been mostly used by researchers and, due to its nature, has obtained accurate and reliable results is the use of plant growth simulation models at the basin scale. The SWAT simulation model is a catchment scale and a time series model that applies for modeling large and complex basins. The SWAT model simulates plant growth, fertilizer and nutrients, land management as well as the hydrological cycle in daily time steps and is known as a proper tool for evaluating water resources in basins with different scales and characteristics. Numerous studies have evaluated and proven the capability of the SWAT model to long-term simulations in large agricultural basins (Borah & Bera 2003; Gassman et al. 2007; Luan et al. 2018; Uniyal et al. 2019).

Faramarzi et al. (2010) investigated the crop yield of wheat and its agricultural water productivity in Iran using the SWAT model. In this study, the effects of some policies such as improving operations of soil moisture conservation and optimizing the fertilizers use on improving crop yield and agricultural water productivity were investigated. They also reported that the yield of rainfed wheat was greater than that of the irrigated wheat. The rainfed wheat also consumed less water, thereby increasing the water productivity for this crop. Using this model, Jeimar et al. (2011) simulated corn potential evapotranspiration variations as well as crop yield based on different scenarios of the available water amount in the future in Peru. Also, by modeling the different stages of growth of the corn plant, they calculated the marginal profit of consumed water for this crop based on the function of plant water production. Mehmood et al. (2017) used the SWAT model to evaluate the productivity of agricultural water in rainfed wheat farming in Pakistan. For this purpose, first, the crop yield of rainfed wheat was extracted using the SWAT model and then the amount of agricultural water productivity was estimated using the relevant relationships. The results of this study confirmed the ability of the SWAT model to simulate the crop yield of rainfed wheat. Magbalot-Fernandez et al. (2019) used the SWAT model to estimate and evaluate crop yields in Bio City, Texas, USA under climate change conditions. Bauwe et al. (2019) evaluated the performance of the SWAT model in simulating crop yields in small agricultural basins in the northeast of Germany. They used flow rate observational data to calibrate the model and crop yield observational data to validate the model. The results of this study showed that the SWAT model has a high ability to simulate the annual crop yield, although the model results were not acceptable for simulating the crop yield on monthly and daily scales. Other similar studies have been performed to simulate crop yields in US agricultural plains using the SWAT model with a greater focus on corn and soybean crops (e.g., Srinivasan et al. 2010; Nair et al. 2011; Guo et al. 2018). Other studies have been conducted in European watersheds on most crops using the SWAT model on different scales from catchment and even sub-basin scale up to continental scale (Abbaspour et al. 2015; Maier & Dietrich 2016; Malagó et al. 2017; Gabiri et al. 2019).

In order to improve the simulations and predictions of different hydrological and agricultural models, ground-based and remote sensing data can be combined with these models using different methods. In general, these methods are divided into three categories: 1. forcing, 2. parameter estimation, and 3. state estimation. In the first method, ground observational data or remote sensing products are used as input to the models. For example, digital elevation maps, precipitation data obtained from remote sensing or ground stations and meteorological data from synoptic meteorological stations can be used as input in hydrological models and land surface models. This method is affected by the uncertainty and error of basic information (Moradkhani et al. 2006; Stisen et al. 2008; Collier 2009; Myers et al. 2021). The second method is to use ground observational data and remote sensing information in parameter estimation, also called calibration. Hydrological models usually have basic conceptual and effective parameters that are very difficult or impractical to determine directly. These parameters should be calibrated to the best local values to obtain the optimal match between the simulated and measured values. Many different studies have been done on the calibration of the SWAT model (Athira 2021; Li et al. 2021; Shah et al. 2021; Ma et al. 2022). There are different methods for analyzing uncertainty in watershed distribution models such as GLUE, ParaSol, MCMC and SUFI2 methods. SWAT CUP software package has developed in order to calibrate and perform the uncertainty analysis of the SWAT model (Abbaspour et al. 2007). The third method is to use ground observational data or remote sensing information in state estimation. The state estimation method, also known as the data assimilation (DA) method, is a process that limits model simulations through observations to improve state variable estimation. The use of a wide variety of data assimilation methods to combine observations with models has been increasing in recent years. A simple method is direct insertion, which uses an observational value to insert directly the corresponding simulated value at any given time step. This method imposes observations as the only constraint regardless of its quality. If the model has a good estimate, it would not make sense to replace it with a low-accuracy observation value. Hence, advanced data assimilation methods update model simulations through an optimal constraint based on measurements and model prediction errors. Among the various methods of data assimilation, the Standard Kalman Filter is a sequential method for linear dynamics systems that minimizes the mean squared error in state estimation. The Extended Kalman Filter is employed for nonlinear systems (Entekhabi et al. 1994; Walker & Houser 2001; Draper et al. 2009). However, this method requires very complex and large calculations and is unstable in severe nonlinear conditions (Miller et al. 1994; Reichle et al. 2002a). Evensen (1994) introduced the Ensemble Kalman Filter and showed that this method, unlike the Extended Kalman Filter method, can be used well for severe nonlinear systems, even with low computations. The EnKF method has gained high acceptance in the hydrological data assimilation and a number of previous studies have proven the performance of this method in improving hydrological predictions in various hydrological models, including the SWAT model (Reichle et al. 2002a, 2002b; Zhang et al. 2006; Clark et al. 2008; Komma et al. 2008; Xie & Zhang 2010; Han et al. 2012; Azimi et al. 2020; Lei et al. 2020; Kivi et al. 2022).

Rainfed agriculture in the Urmia Lake basin has received considerably less attention than irrigated agriculture in the context of agricultural water management. Consequently, in this paper, the aim is to investigate the rainfed agricultural potential in the ZarrineRoud sub-basin, as this is the most important sub-basin from a water resources perspective. The SWAT model was used to simulate the rainfed wheat yield at the basin. Also the model predictions were improved by combining the remote sensing derived hydrologic variables; surface soil moisture and evapotranspiration into the model, through EnKF assimilation method.

ZarrinehRoud Basin is the largest sub-basin of Urmia Lake basin, which is located in the position of 45̊ 47′–47̊ 20′ longitude and 35̊ 41′–37̊ 27′ latitude. This river provides 47% of the volume of water entering Lake Urmia. The area of this basin is more than 12,025 square kilometers and its main river has a length of 300 kilometers. The cities of Miandoab, Shahin Dej, Takab and Saqez are important urban centers of this basin. Figure 1 shows the geographical location of the ZarrinehRoud basin relative to the Urmia Lake basin and Iran. One of the factors that have a significant impact on the hydrological cycle of this basin as well as the development of downstream agriculture is the ZarrinehRoud Dam. This dam has a total water volume of 762 million cubic meters and its dominant use is the agriculture sector, domestic water consumption and environmental needs of Lake Urmia.

In this study, the remote sensing results of the surface soil moisture and actual evapotranspiration variables in the ZarrinehRoud basin, which is taken from the study of Rostami (2020), are used to assimilate the SWAT model predictions. In that study, the MODIS¹ images and the SEBAL² algorithm (Bastiaanssen et al. 1998) were used to estimate the actual evapotranspiration for both 2010 and 2014. To estimate the surface soil moisture, first, the triangular space of land surface temperature against vegetation index was extracted using MODIS images. Then the TVDI³ was calculated by analyzing this triangular space. Finally using the TVDI and the UCLA⁴ downscaling method (Kim & Hogue 2012), the surface soil moisture remote sensing products were downscaled to the spatial resolution of MODIS images (1000 m) and the time series of surface soil moisture were obtained for the study area for the years 2010 and 2014. Table 1 summarizes the information about the data, methods, and related accuracy of each resulting variable.

Daily temperature, precipitation, solar radiation, relative humidity and wind speed data were prepared from meteorological stations that were closest to the study area to simulate the desired processes (Table 2) and in cases where there was no data, variables were created by the SWAT model itself. The monthly discharge flow data of six hydrometric stations were used to calibrate and validate the model. The location of these stations is shown in Figure 2 and their characteristics are presented in Table 3. In this paper, instead of the local names of hydrometric stations, their row numbers will be used in Table 3.

In this study, a 30-meter cell size DEM map taken from the 2010 SRTM product was used to generate physiographic information of rivers, basins, and sub-basins (Figure 3). Land use information is required to implement the cultivation pattern in the basin and hydrological simulation and estimate water productivity. The land use map used in this research is presented in Figure 4. In this study, due to the unavailability of up-to-date information and maps of the soils of the region, the soil map of FAO (1974) has been used, which includes all the characteristics and physical information of the soils of the region, including soil texture and structure. Table 4 shows some properties of soil units in the region based on the SWAT model database. Figure 5 shows the soil type map used in the study.

Calibration of the SWAT model is one of the most important and time-consuming steps of modeling. In order to accelerate the sensitivity analysis, calibration and to evaluate the uncertainty of the model results, SWAT-CUP software has been developed by Abbaspour et al. (2007). There are various methods for calibration in this software, including ParaSol, MCMC, GLUE and SUFI-2. In this study, SUFI-2 method has been used. This method has higher computational speed and better performance in calibrating parameters than other mentioned methods (Yang et al. 2008).

In this equation, i is the observational data, is the data modeled using the parameter series θ, and is the average of the observational data. The value of this index can vary from −∞ to 1. When the value of this index is equal to one (NSE=1), it indicates the complete agreement of the modeling values with the observational data. When the value of the index is equal to zero (NSE=0), it indicates that the model estimates have an accuracy equal to the average of the observational data. Values below zero (NSE<0) of this index represent that the average of the observed data is a better estimator than the model.

In hydrology, data assimilation methods are used to improve model predictions by combining observations with incomplete information from the hydrological processes expressed in hydrological models (Walker & Houser 2005). In this study, the EnKF method has been used to assimilate the remote sensing information of the surface soil moisture as well as the actual evapotranspiration in the SWAT model. After calibrating the model using the measured discharge information and estimating the optimal parameters, the remote sensing data is assimilated to estimate the state variables (surface soil moisture and actual evapotranspiration) using several modules coded in FORTRAN language for the EnKF method algorithm and the source code of the SWAT model. In Evensen (1994) and Han et al. (2012) the algorithm of EnKF method is given.

The SWAT model estimates the crop yield using all input data used in calibration and data assimilation indirectly, in addition to other climatic and agronomic variables directly included in its theoretical relationships. Therefore, to comprehensively (hydrological, climatic and agronomic) study the potential of rainfed wheat agriculture in the study area, first the average crop yield of rainfed wheat were estimated using the calibrated and assimilated SWAT model. Then, using the range of values obtained from the simulation and the statistical values in the study area, the classification map of rainfed wheat yield was produced. In the next step, using the existing DEM map of the study area, the classification map of land slope was created. By combining and overlapping these two classified maps (crop yield map and land slope map) using the weighted overlap method, the final zoning map of rainfed wheat agricultural potential in the study area was obtained.

Before calibrating and validating the model, a sensitivity analysis (using SWAT CUP software) concerning the parameters that had the greatest impact on the discharge flow of the basin was performed. Table 5 shows the results of this analysis. In this table, the impact degree of each of the parameters involved in the simulation is determined by their rank, as well as their p-value and t-stat value. Each parameter with a higher absolute value of t-stat and a p-value close to zero has a greater effect on the discharge flow. Table 5 shows that the curve number parameter (CN2) with the highest t-stat value and also the lowest p-value had the greatest effect on the discharge flow values of the basin. After CN2, GW-DELAY is in second place. These two parameters affect the surface runoff and the rate of groundwater participation in the total outflow from the basin, respectively.

After performing the sensitivity analysis step, the model was calibrated and validated using the monthly data of six hydrometric stations. In this study, two indices, the coefficient of determination (R) and Nash-Sutcliffe coefficient (NSE) were used to evaluate the model. Table 6 represents the results of this evaluation. The final range of model parameters is shown in Table 5. By examining different methods of sensitivity analysis, Li et al. (2021) found parameters for calibration with daily and monthly streamflow data that were similar to those of the present study (Table 5). Similarly, the results of the SWAT model calibration in that study provided acceptable NSE values, which are consistent with those in this study (Table 6). In all statistical analysis tables, the symbols * and ** in each correlation coefficient indicate a significant correlation between the two variables at a level of 1 and 5%, respectively.

Figure 4 shows the graphs of the average surface soil moisture for consecutive days from 2010 to 2014 (Julian days when remote sensing data were estimated on those days). In these graphs, TRUE is related to observational data (remote sensing data presented in Table 1), MODEL is related to the results of SWAT model simulations calibrated by the discharge flow of basin under SUFI2 algorithm, and EnKF is related to the combination of observational data and the SWAT calibrated model using the ENKF method. In addition, Table 7 represents the results of statistical evaluation in the form of RMSE, R, MBE, MAE and MARE indices.

As can be seen, the ENKF graph is located between the MODEL and TRUE graphs in both years, because the nature of the data assimilation method, the EnKF, is the estimation of the state variable (in this case surface soil moisture) using observations and model simulations and considering the amount of error for both of them and finally minimizing the covariance of the error resulting from both data sets. The table shows that the values of all statistical indicators improved significantly in both 2010 and 2014. However, the improvement of model simulations in 2010 is greater. Similar results and evaluations about improving the different hydrological models, especially the SWAT model simulations with the assimilation of surface soil moisture data and EnKF algorithm, has been reported in various studies including Crow & Kustas (2008), Han et al. (2012), Xie & Zhang (2010), Sun et al. (2015) and Laiolo et al. (2016). As an example, Azimi et al. (2020) also assimilated different remote sensing soil moisture products with the SWAT model and found that all products improved the accuracy of SWAT discharge simulations, both in terms of error and Nash-Sutcliffe efficiency index.

Figure 5 shows the graphs of the average value of the actual evapotranspiration variable at the basin area on consecutive days of the study period (estimated remote sensing data time series) in 2010 and 2014. TRUE, MODEL and ENKF symbols in this figure are similar to those mentioned in Figure 4. Table 8 shows the results of statistical evaluation in the form of the mentioned indicators. In this figure and table values, it can be seen that the model simulations have significantly improved under the influence of assimilation of observational data by EnKF method. The value of r index in 2010 increased from 0.84 to 0.95 and in 2014 increased from 0.73 to 0.95 and the value of RMSE index decreased from 1.48 mm/day in 2010 to 0.64 mm/day and in 2014 decreased from 0.9 to 0.47 mm/day. Therefore, it is concluded that the application of the data assimilation method (EnKF method) in improving the accuracy of simulations of SWAT model in both variables of surface soil moisture and actual evapotranspiration is quite significant. Also, similar results regarding the effect of using different data assimilation methods such as EnKF on improving the accuracy of simulations of different hydrological models including SWAT in estimating actual evapotranspiration has been reported in various studies such as Liu et al. (2010) and Wu et al. (2012). Similarly, Shah et al. (2021) and Paul et al. (2021) demonstrated an improvement in the SWAT model ET and streamflow simulation when combined with remote sensing derived ET and LAI data.

Table 9 shows the results of evaluation of simulated values of basin discharge in two cases: 1. calibrated model with observational discharge values of hydrometric stations, and 2. assimilated model with remote sensing data (surface soil moisture and actual evapotranspiration) under EnKF algorithm in 2010 and 2014. The results of the first case are the same values presented in the calibration and validation section of the SWAT model, which are listed in this table to analyze the results. As can be seen, in 2010 the rate of improvement of the R² coefficient index due to the assimilation of observations with SWAT model from a minimum of 4% at the station 4 is up to a maximum of 15% at station 1. The rate of improvement of NSE index in 2010 from a minimum of 3% in station 3 is up to a maximum of 19% in station 1. In 2014, the R² coefficient showed an improvement from 1% at station 4 to 2% at station 1. The NSE index in 2014 showed an improvement from 6% at stations 4 and 5 up to 25% at station 1. The high improvement of the evaluation indicators at station 1 is related to the poor results of the initial simulations of the SWAT calibrated model. In general, the assimilation of remote sensing data improves the simulation of the discharge flow values of the basin, even in the calibrated model, by an average of 5–10%.

By assimilation of the remote sensing data (surface soil moisture and actual evapotranspiration) with the SWAT calibrated model (with observational discharge flow data), the values of some parameters such as HVSTI, BLAI, FRGRW1, FRGRW2, LAIMX1, LAIMX2, which have a basic role in the simulation of crop yield in the SWAT model, were obtained. Table 10 presents the final values of these parameters.

Figure 6 shows the zoning map of the average yield of simulated rainfed wheat for the years 2010–2014 in the ZarrinehRoud basin. This figure shows that the yield of rainfed wheat in the northern regions is higher than the southern regions of the basin, and this value is equal to 2.2 tons per hectare in sub-basin 3 on average, which has the highest amount compared to other sub-basins. In addition, sub-basin 9 has the lowest yield with an average yield value of 1.17 tons per hectare. The average simulated yield value of rainfed wheat in the study area is approximately equal to 1.5 tons per hectare, which is in good agreement with the statistics of the Agriculture Statistics (2010, 2014). According to these statistics, the average yield value of rainfed wheat in the plains of Kurdistan province is 1.26 tons per hectare and in West Azerbaijan province is equal to 1.7 tons per hectare, and the average value for the whole study area for the years 2010–2014 is approximately 1.57 tons per hectare.

Figure 7 shows the average values of actual evapotranspiration simulated over the entire growing season of rainfed wheat crop from 2010 to 2014 for each sub-basin. According to this figure, in general, the amount of actual evapotranspiration in the northern sub-basins (including sub-basins 1, 2 and 5) is higher than the southern areas of the basin (including sub-basins 9, 10 and 11). Therefore, the yield value zones of rainfed wheat correspond to the simulated actual evapotranspiration zones; the yield values of rainfed wheat and its actual evapotranspiration in the northern sub-basins are higher than the southern sub-basins. The highest amount of actual evapotranspiration of rainfed wheat is in sub-basin 1, equal to 280.7 mm, and the lowest amount related to sub-basin 4 is equal to 201 mm.

Figure 8 shows the average amount of total rainfall during the wheat growing season in each of the sub-basins. According to Figure 8, the total amount of rainfall during the growing season of wheat varies from 281.7 to 424.7 mm, with the highest amount related to sub-basin 10 and the lowest amount related to sub-basins 3, 4 and 6. It should be noted that in modeling the crop yield by the SWAT model, three factors play a key role, which are: 1. temperature stress, 2. water stress and 3. fertilizer stress. In other words, in the SWAT model, hydrological factors (soil moisture), climatic factors (including temperature and other climatic parameters) and plant and agricultural factors (including fertilizers and parameters mentioned in Table 10) are all effective in estimating crop yield. Therefore, the lower yield of rainfed wheat in areas with more rainfall can be due to sandy soil (less water holding capacity) or due to temperature stress. Also, the high yield of rainfed wheat crop in northern sub-basins can be due to lower temperature stress or other favorable conditions aforementioned.

Figure 9 shows the spatial variation of the average soil moisture during the wheat growing season from 2010 to 2014 in the study area. The average amount of soil moisture in the northern sub-basins is relatively higher than the southern sub-basins. According to the soil map, this is due to the type of soil texture and can be due to the general slope of this area, and despite the lower average annual rainfall in these sub-basins, the amount of stored soil moisture is higher. Therefore, despite the low average annual rainfall in northern sub-basins, the high average yield and actual evapotranspiration of rainfed wheat indicates that there are better physical characteristics of the soil to store soil moisture. These characteristics led to less water stress and lower temperature stress in the northern sub-basins compared to the southern sub-basins.

Table 11 shows the yield classification values of rainfed wheat in the study region. This classification has been done according to the range of yield values obtained from the simulations and the average yield values of rainfed wheat in the study area, which is presented in the historical data of the Agriculture Statistics (2010, 2014). Yield of rainfed wheat is classified into four classes: ‘Poor’, ‘Moderate’, ‘Good’ and ‘Very Good’. Using the values of this table as well as the zoning map of rainfed wheat yield (Figure 6), Figure 10 shows the classification map of the yield of rainfed wheat crop in the study area.

Using the DEM map, Figure 11 shows the classification map of land slope into four ranges of less than 10, 10–15, 15–30 and above 30%, which represent the ‘Very Good’, ‘Good’, ‘Moderate’ and ‘Poor’ classes, respectively. This classification is according to the range of land slopes recommended in various studies for rainfed wheat cultivation including Vaezi et al. (2019). According to this figure, the northern and central parts of the region have lower land slopes than the southern and western regions, which are generally highlands.

Finally, by weighing each of the different classes of both variables, land slope percentage and yield value of rainfed wheat, a map of the potential of rainfed wheat agriculture in the study area is produced. Table 12 shows the weight percentages assigned to each of the classification classes. It is not feasible to commercially produce wheat in land slopes greater than 30%, therefore by assigning a weight percentage of 100% to this class of the land slope variable, all lands with a slope higher than 30%, regardless of each yield class value, wheat potential production is ‘Inappropriate’. Considering that in the classes of ‘Very Good’, ‘Good’ and ‘Moderate’ of the land slope variable, the effect of the variable of rainfed wheat yield by including various factors (hydrological, climatic and agricultural) is much greater. Therefore, the weight percentage assigned to the yield of rainfed wheat is 70% and the weight percentage assigned to the slope variable is 30%.

Now, according to the values in Table 12, Figures 10–12 show the final map of the potential of rainfed wheat agriculture in the study area. Figure 12 shows that, as expected from the maps of rainfed wheat yield and the land slope, areas with ‘Very Appropriate’ and ‘Appropriate’ potential for rainfed wheat agriculture are mainly in the northern and western areas of the region and areas with more ‘Moderate’ and ‘Inappropriate’ potential are located in the southern and east areas of the region. In general, as shown in Table 13, 10.5% of the lands in the study area with an approximate area of 125,000 hectares in terms of hydrological, climatic and agricultural conditions and land slope have a ‘Very Appropriate’ potential for rainfed wheat agriculture. In addition, 25.4% of the lands in the study area with an area of approximately 300,000 hectares in terms of these conditions have an ‘Appropriate’ potential for rainfed wheat agriculture. In contrast, areas with ‘Moderate’ and ‘Inappropriate’ potential occupy 64.1% of the region with an area of 770,000 hectares, and due to the relatively weak and unsuitable hydrological, climatic, agricultural and land slope conditions in these areas, rainfed wheat agriculture is not recommended for these areas.

Considering the importance of rainfed agriculture in arid and semi-arid climates and the critical situation of water resources in the Urmia Lake basin, in this study, the potential of rainfed agriculture in Zarrinehroud basin was studied. Furthermore, this study focused on the wheat plant because it is the most dominant and crucial crop for rainfed agriculture in the study area. The results of this study showed that the SWAT model has a relatively good performance in runoff simulation using calibration of measured flow information, so that the correlation coefficient was obtained from 0.69 to 0.84 in the calibration period and from 0.64 to 0.86 for the validation period. However, the assimilation of remote sensing data with the calibrated SWAT model showed that the model simulations for both surface soil moisture and actual evapotranspiration variables were significantly improved, indicating the high ability of the EnKF method and remote sensing data to improve the model simulation accuracy. In addition, evaluation of the SWAT-EnKF hybrid model shows that the combination of remote sensing observations of soil surface moisture and actual evapotranspiration improves the simulation of the flow rate by an average of 5–10% compared to the calibrated model. The high simulation capability of the SWAT model and the variety of its outputs made it possible to estimate the variables and parameters required to simulate the yield of rainfed wheat crop in the basin by considering all hydrological, climatic and agronomic components. By estimating the yield of rainfed wheat in the study area, zoning maps of rainfed wheat yield were extracted which shows that in the northern sub-basins the yield of rainfed wheat is higher than the southern sub-basins. The final map of rainfed wheat agricultural potential in four classes ‘Very Appropriate’, ‘Appropriate’, ‘Moderate’ and ‘Inappropriate’ was drawn by overlapping the classified maps of rainfed wheat yield and land slope. Areas with Very Appropriate and Appropriate potential for rainfed wheat agriculture are mainly located in the northern and western regions of the region and areas with Moderate and Inappropriate potential are located in the southern and eastern regions of the region. The results showed that a total of 35.9% of the lands of Zarrinehroud basin have ‘Very Appropriate’ and ‘Appropriate’ potential for rainfed wheat agriculture, which can be considered as an alternative to irrigated agriculture in the region.

This study provides a valuable solution for managers and decision-makers of water resources and agriculture in the basin as well as an effective tool and guide for researchers looking to improve basin simulations using assimilation of remote sensing resulted variables and data, like soil moisture and evapotranspiration. In order to complete the results of this study, it is suggested that similar studies be conducted for other common and suitable crops in rainfed agriculture in the region. According to this study, if there are sufficient data and ground observations to validate the remote sensing resulted variables, a similar study could be conducted to analyze and compare the probable effects of climate change in recent years.

No potential conflict of interest was reported by the authors.

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

The authors confirm contributions to the paper as follows: study conception and design: Amin Rostami, Mahmoud Raieni-Sarjaz; data collection: Amin Rostami, Jafar Chabokpour and Sumit Kumar; analysis and interpretation of results: Amin Rostami, Mahmoud Raieni-Sarjaz, Jafar Chabokpour, Hazi Md Azamathulla; draft manuscript preparation: Amin Rostami, Jafar Chabokpour, Hazi Md Azamathulla and Sumit Kumar. All authors reviewed the results and approved the final version of the manuscript.

Not applicable.

Not applicable.

Not applicable.

Not applicable.

Data cannot be made publicly available; readers should contact the corresponding author for details.