Estimation of Evapotranspiration and Soil Water Content at a Regional Scale Using Remote Sensing Data

: The timely and accurate estimation of soil water content (SWC) and evapotranspiration (ET) is of great signiﬁcance in drought estimation, irrigation management, and water resources comprehensive utilization. The unsupervised classiﬁcation was used to identify the crops in the region. Based on MOD16A2 and the meteorological data, a SEBS model was used to estimate the ET in the Jiefangzha Irrigation Field from 2011 to 2015. Based on the crop water stress index (CWSI), the SWC in 2014 was retrieved and veriﬁed with the measured SWC on different underlying surfaces (sunﬂower, corn, wheat, and pepper). The results showed that: (1) The positional accuracy of maize, sunﬂower, wheat, and pepper are 0.81, 0.80, 0.90, and 0.82, respectively; (2) The annual ET from 2011 to 2015 presented well the spatial distribution of the ET within the ﬁeld; (3) The validation results of the estimated SWC on the underlying surface of wheat and sunﬂower showed a good robustness, the R 2 was 0.748 and 0.357, respectively, the RMSE was 2.61% and 2.309%, respectively, and the MAE was 2.249% and 1.975%, respectively. However, for maize and pepper with more irrigation times, the SWC estimation results, based on the CWSI were poor, indicating that the method was more sensitive to soil drought and suitable for the crop SWC estimation with less irrigation and drought tolerance. The results can provide a reference for the agricultural water resources management and the irrigation forecast at a regional scale.


Introduction
Under the influence of climate change and water pollution by human activities, the shortage of water resources has become a worldwide and challenging problem, especially in agricultural water. In China, over 60% of the total water consumption is by agriculture. As the main grain-producing areas, the irrigation districts in China can produce over 75% of the grain crop and 90% of the industrial crop by taking around 40% of the cultivated area and consuming 62% of the water supply, which is of great importance in national food security. The irrigated area in China is over four times larger, compared with its size in the 1950s and reached up to 658,700 ha by the end of 2016. However, the irrigation water use efficiency in China (~0.5) is much lower than in developed countries (~0.8), which urges the development of water-saving irrigation techniques and a high quality irrigation water management.
Soil water content (SWC) is the amount of water present in unsaturated soil layers [1]. It is an important water source for soil evaporation and vegetation transpiration. As an important soil parameter in the process of surface water circulation, it has great influences on the conversion of the sensible heat flux and latent heat flux and the coupling feedback mechanism between the surface and atmosphere systems [1]. Evapotranspiration (ET), which is the sum of water vapor dissipated into the atmosphere by soil evaporation and plant transpiration on the surface of the earth, maintains the relationship between the surface water, carbon and energy cycle. Therefore, the SWC and ET are the critical the west, and connected with the Yongji Irrigation Field in the east. The total area of the Jiefangzha Irrigation Field is 2345 km 2 , in which over 70% is farmland. It belongs to the continental arid monsoon climate of the middle temperate zone, with sufficient light, low rainfall, and a large temperature difference between day and night. The irrigation field enters the freezing stage from mid-November, and completely melts in late-May of the next year, lasting 180 days. The total water surface evaporation is 2300 mm, which is over 10 times that of the annual rainfall (151.3 mm), and the annual mean temperature is 9 • C. The food crop of the irrigation field in summer, is maize (May to October) and in spring, is wheat (April to July), and the major economic crop is sunflower (June to October), as well as a certain proportion of vegetable and fruit [14].
The Jiefangzha Irrigation Field (40°34′~41°14′ N, 106°43′~107°27′ E), which is the second largest irrigation field in the Hetao Irrigation District, was taken as the research area ( Figure 1). It is located in the west of the Hetao Irrigation District, adjacent to the Yellow River in the south, the Yinshan Mountain in the north, bordering the Ulan Buhe Desert in the west, and connected with the Yongji Irrigation Field in the east. The total area of the Jiefangzha Irrigation Field is 2345 km 2 , in which over 70% is farmland. It belongs to the continental arid monsoon climate of the middle temperate zone, with sufficient light, low rainfall, and a large temperature difference between day and night. The irrigation field enters the freezing stage from mid-November, and completely melts in late-May of the next year, lasting 180 days. The total water surface evaporation is 2300 mm, which is over 10 times that of the annual rainfall (151.3 mm), and the annual mean temperature is 9 °C. The food crop of the irrigation field in summer, is maize (May to October) and in spring, is wheat (April to July), and the major economic crop is sunflower (June to October), as well as a certain proportion of vegetable and fruit [14].

In-Situ Measurement
The meteorological data used in the surface dataset consists of the daily precipitation, the daily average air temperature, the maximum air temperature and minimum air temperature, the daily average relative humidity, the daily average wind speed, and the daily sunshine duration. These data were obtained from 10 national meteorological stations distributed in or outside the Jiefangzha Irrigation Field ( Table 1). The spatial interpolation used in the study was the angular distance weighting method, which can result in a relatively smooth spatial distribution. The spatial distribution of the meteorological input was thus generated from the plot-scale data collected by the meteorological stations.

In-Situ Measurement
The meteorological data used in the surface dataset consists of the daily precipitation, the daily average air temperature, the maximum air temperature and minimum air temperature, the daily average relative humidity, the daily average wind speed, and the daily sunshine duration. These data were obtained from 10 national meteorological stations distributed in or outside the Jiefangzha Irrigation Field ( Table 1). The spatial interpolation used in the study was the angular distance weighting method, which can result in a relatively smooth spatial distribution. The spatial distribution of the meteorological input was thus generated from the plot-scale data collected by the meteorological stations.
The SWC measurements were set at 10 places in the typical croplands of wheat, maize and sunflowers throughout the irrigation fields. By using the CTMS-On line system, developed by the China Institute of Water Resources and Hydropower Research, the continuous SWC and soil temperature were automatically observed and collected. The SWC profile has three sensors that reach a depth of 10 cm, 20 cm, and 40 cm, from 1 May 2014 to 24 September 2014, which covered the major growing period of the typical crops. The observation interval was one h. By eliminating the abnormal values, the daily average SWC was thus calculated from the hourly measurements ( Figure 2). It shows that the moisture in the field with the sunflower cultivation is, for some time periods, lower at the depth of 20 cm, compared to the moisture at the depth of 10 cm. The sunflower has a very developed root system with a strong ability to absorb water and nutrients. It is widely and deeply distributed in the soil, and about 60% of the roots are distributed in the 0-40 cm soil layer. Most of this is at a depth about 20 cm. From mid-August to late-September, the sunflower is in the filling stage and maturity stage, and the root water absorption is high, which causes this phenomenon. The SWC measurements were set at 10 places in the typical croplands of wheat, maize and sunflowers throughout the irrigation fields. By using the CTMS-On line system, developed by the China Institute of Water Resources and Hydropower Research, the continuous SWC and soil temperature were automatically observed and collected. The SWC profile has three sensors that reach a depth of 10 cm, 20 cm, and 40 cm, from 1 May 2014 to 24 September 2014, which covered the major growing period of the typical crops. The observation interval was one h. By eliminating the abnormal values, the daily average SWC was thus calculated from the hourly measurements ( Figure 2). It shows that the moisture in the field with the sunflower cultivation is, for some time periods, lower at the depth of 20 cm, compared to the moisture at the depth of 10 cm. The sunflower has a very developed root system with a strong ability to absorb water and nutrients. It is widely and deeply distributed in the soil, and about 60% of the roots are distributed in the 0-40 cm soil layer. Most of this is at a depth about 20 cm. From mid-August to late-September, the sunflower is in the filling stage and maturity stage, and the root water absorption is high, which causes this phenomenon.

Remote Sensing Data
The remote sensing data used in our research were Landsat-7 ETM+ and the Moderate Resolution Imaging Spectroradiometer (MODIS) land surface product on board the Terra and Aqua satellites. Landsat-7 was launched by the National Aeronautics and Space Administration (NASA) on 15 April 1999, and carries an enhanced thematic imaging sensor (ETM+) that retains the original multispectral characteristics of the Landsat-5 thematic imaging sensor (TM). However, its spatial resolution in the thermal infrared bands is increased from 60 m to 120 m. The Landsat-7 ETM+ also adds a panchromatic bands with a spatial resolution of 15 m. In view of the advantages of the sensor in space and the spectral resolution, Landsat-7 has become one of the most commonly used remote sensing data sources, widely used in agricultural surveying, hydrological forecasting, and environmental monitoring [15]. The MOD16A2 products reflect the comprehensive effects of the global soil moisture evaporation, vegetation transpiration, vegetation canopy interception, and water evaporation, including ET, the potential evapotranspiration (ET p ), the latent heat flux (LE), and the potential latent heat flux (LE p ). The MODIS data provide two observations daily, at an over-passing time of approximately 10:30 am and 1:30 pm local time. Both data were re-projected into the Universal Transverse Mercator (UTM) map projection system. The images with a heavy cloud cover were removed according to the quality flags in the datasets [16].

Data Fusion Method
In order to take advantage of both the finer resolution remote sensing data (Landsat-7 ETM+) and the coarse data with a shorter recurrence interval (MODIS), the ESTARFM, developed by Zhu et al. [17], was used in our research. The main idea of the ESTARFM is to make use of the correlation to blend the multi-source data and to minimize the system biases. The relationship between the coarse-resolution reflectance and the fine-resolution reflectance can be expressed as: where F and C denote the fine-resolution reflectance and coarse-resolution reflectance, respectively; w is the search window size; t 0 and t p are reference time and prediction time, respectively; (x w/2 , y w/2 ) is a given pixel location for both the fine-resolution and the coarse resolution images; N is the size of the similar pixels, including the central prediction pixel; (x i , y i ) is the location of the ith similar pixel; W i is the weight of the ith similar pixel; V i is the conversion coefficient of the ith similar pixel. The final predicted fine-resolution reflectance at the prediction time t p is calculated as: where F(x w/2 , y w/2 , t p , B) denotes the final predicted fine-resolution reflectance; F m (x w/2 , y w/2 , t p , B) and F n (x w/2 , y w/2 , t p , B) are the fine-resolution reflectance at t m and t n ; T m and T n are the temporal weight at t m and t n , respectively.

ISODATA Cluster Algorithm and Spectral Matching
The Landsat-7 ETM+ images and the MODIS normalized difference vegetation index (NDVI) data were re-merged to produce a mega data tube, which was then classified using the ISODATA cluster algorithm. In the spectral matching, the algorithms need the definition of some criteria for measuring the similarity and closeness of the pixels.
The spectral similarity value (SSV) is a combined measure of the correlation similarity and the Euclidian distance, which can be formulated as: where d e is the Euclidian distance; n is the length of the NDVI time series; X and Y denote the NDVI time series; r is Pearson's correlation coefficient, and ranges from −1 to 1. The images of the Jiefangzha Irrigation Field were divided into 50 clusters by the ISODATA algorithm, and generated the NDVI variation curve of each cluster. The statistical characteristic curves of the NDVI in the croplands were compared with the realistic NDVI curves through a survey (Figure 3), and then recognized and analyzed using the spectrum similarity.
where de is the Euclidian distance; n is the length of the NDVI time series; X and Y denote the NDVI time series; r is Pearson's correlation coefficient, and ranges from −1 to 1.
The images of the Jiefangzha Irrigation Field were divided into 50 clusters by the ISODATA algorithm, and generated the NDVI variation curve of each cluster. The statistical characteristic curves of the NDVI in the croplands were compared with the realistic NDVI curves through a survey (Figure 3), and then recognized and analyzed using the spectrum similarity.

SEBS Model
In 2002, Su [18] proposed a single-layer model, based on the principle of energy balance, namely the SEBS model, which has a high estimation accuracy and is widely used in drought monitoring and water resources evaluation [19]. The model assumes that the sensible heat flux of each pixel is between the dry limit and the wet limit of the pixel, which avoids the error caused by the uncertainty of the spatial interpolation of the meteorological data, to a certain extent. The SEBS model can be expressed as follows [20]: where, Rn is the net radiation flux, W/m 2 ; G is the soil heat flux, W/m 2 ; H is the sensible heat flux, W/m 2 ; λE is the latent heat flux, W/m 2 . Rn is the difference between the total radiation from the surface up and down [21]. The calculation formula is as follows:

SEBS Model
In 2002, Su [18] proposed a single-layer model, based on the principle of energy balance, namely the SEBS model, which has a high estimation accuracy and is widely used in drought monitoring and water resources evaluation [19]. The model assumes that the sensible heat flux of each pixel is between the dry limit and the wet limit of the pixel, which avoids the error caused by the uncertainty of the spatial interpolation of the meteorological data, to a certain extent. The SEBS model can be expressed as follows [20]: where, R n is the net radiation flux, W/m 2 ; G is the soil heat flux, W/m 2 ; H is the sensible heat flux, W/m 2 ; λE is the latent heat flux, W/m 2 . R n is the difference between the total radiation from the surface up and down [21]. The calculation formula is as follows: where, α is the albedo, that is, the ratio of the flux of the solar radiation reflected from the ground to the flux of the incoming solar radiation. It is determined by the surface features and the solar altitude. R s is the downward shortwave radiation, W/m 2 ; R ↓ L is the downwelling longwave radiation, W/m 2 ; R ↑ L is the upwelling longwave radiation, W/m 2 ; ε is the surface emissivity.
G is the heat exchange, per unit area of soil in a unit time, and is the ratio of the heat stored in the soil and vegetation, due to conduction [21]. The empirical statistical formula is as follows: where T s is the land surface temperature, K; NDVI is the normalized difference vegetation index. H refers to the turbulent heat exchange between the atmosphere and the underlying surface caused by temperature changes [21]. It is obtained from the surface temperature and the meteorological parameters, and is proportional to the temperature difference. The calculation formula can be expressed as follows: where ρ is the air density and its value is 1.29 kg/m 3 ; C p is the specific heat of air at a constant pressure and its value is 1004 J (kg · K) −1 ; T a is the reference height temperature, K. γ a is the aerodynamic resistance, m/s.

CWSI Method
Jackson et al. [22] proposed the crop water stress index (CWSI), based on the canopy temperature, according to the energy balance theory, formulated as: where ET a denotes the actual evapotranspiration, which could be calculated from the MOD16A2; ET p denotes the potential evapotranspiration, which could be calculated by the Penman-Monteith method [23]. The CWSI reflects the ratio between the plant transpiration and the possible maximum evapotranspiration, which could be considered as the evaluation index of the water condition at the root layer. Under the circumstances of water stress, the ET a is the planted ET under the condition of a sufficient water supply (i.e., the ET p ), multiplied by the modified soil moisture coefficient (K s ).
ET a = K s ET p The K s can be calculated by the power function method as follows [24]: where θ is the average SWC, cm 3 ·cm −3 ; θ wp is the wilting water content, cm 3 ·cm −3 ; θ j is the field capacity, cm 3 ·cm −3 ; c and d are the empirical constant, which varies from crop types and soil conditions. According to Equations (11) and (12): The SWC could be calculated as follows:

Statistical Metrics
To fully evaluate the performance of the estimated SWC, three metrics were used, including the determination coefficients (R 2 ), the mean absolute error (MAE), and the root mean square error (RMSE). The R 2 indicates the fit degree of the model and the larger the R 2 , the better the fitting effect and the stronger robustness of the model. The RMSE indicates the prediction accuracy of the model and the smaller the RMSE, the higher the prediction accuracy of the model. The MAE is the average of the absolute value of the deviation between all individually measured values and the estimated values. Compared with the mean error, the MAE will not be offset by the positive and negative because the deviation is absolute. Therefore, the MAE can better reflect the actual situation of the predicted error and the smaller the MAE, the better the fitting effect of the model.
The metrics are calculated as follows: where n is the number of the validation value during the day; M i and S i are the measured and estimated SWC, respectively.

Planting Structure
As shown in Figure  diction accuracy of the model. The MAE is the average of the absolute value of the deviation between all individually measured values and the estimated values. Compared with the mean error, the MAE will not be offset by the positive and negative because the deviation is absolute. Therefore, the MAE can better reflect the actual situation of the predicted error and the smaller the MAE, the better the fitting effect of the model.
The metrics are calculated as follows:  (18) where n is the number of the validation value during the day; Mi and Si are the measured and estimated SWC, respectively.

Planting Structure
As shown in Figure 4   The groundwater level at the Jiefangzha Irrigation Field in March 2015, using the Kriging interpolation method [25], is shown in Figure 5. The southwest and part of the northeast areas of the study area have relatively shallower groundwater levels. Combined with the planting structure, sunflower had a higher intensity in the region with a shallower groundwater level; while wheat and maize tended to grow in the region with a deeper groundwater level, i.e., middle and southeast parts of the irrigation field. This characteristic planting structure can give suggestions for the irrigation water distribution to the management sectors.

Spatial Distribution of the ET
The daily ET was calculated from the MOD16A2 data on-board the Terra satellite. The results from one selected day (13 August 2014), are shown in Figure 6, in which the cloud contaminated pixels were eliminated (Figure 6a), and thus interpolated by upscaling from the continuous ET estimation from the same pixel. The Figure 6b shows that after the interpolation, the ET in the missing pixels, is smaller than the pixels exposed to clear day, which was reasonable because under the cloud contaminated pixels, the available energy (i.e., net radiation minus the soil heat flux) was smaller, due to a lack of sunshine.
The groundwater level at the Jiefangzha Irrigation Field in March 2015, using the Kriging interpolation method [25], is shown in Figure 5. The southwest and part of the northeast areas of the study area have relatively shallower groundwater levels. Combined with the planting structure, sunflower had a higher intensity in the region with a shallower groundwater level; while wheat and maize tended to grow in the region with a deeper groundwater level, i.e., middle and southeast parts of the irrigation field. This characteristic planting structure can give suggestions for the irrigation water distribution to the management sectors.

Spatial Distribution of the ET
The daily ET was calculated from the MOD16A2 data on-board the Terra satellite. The results from one selected day (13 August 2014), are shown in Figure 6, in which the cloud contaminated pixels were eliminated (Figure 6a), and thus interpolated by upscaling from the continuous ET estimation from the same pixel. The Figure 6b shows that after the interpolation, the ET in the missing pixels, is smaller than the pixels exposed to clear day, which was reasonable because under the cloud contaminated pixels, the available energy (i.e., net radiation minus the soil heat flux) was smaller, due to a lack of sunshine.
The five-year continuous regional estimation of the ET was calculated over the entire Jiefangzha Irrigation Field. Figure 7 shows the spatial distribution of the annual ET total from 2011 to 2015. The spatial mean ET, during the crop growing season (April to October) of five years over the croplands, were 530 mm, 604 mm, 563 mm, 530 mm, and 551 mm, respectively. The precipitation during the same period from 2011 to 2015, were 45.7 mm, 207.1 mm, 102.8 mm, 141.8 mm, and 111.6 mm, respectively. Despite the annual rainfall, it varied a lot in the irrigation field, due to the sufficient irrigation from the Yellow River, the ET in the study area remained at a stable level. The annual ET in 2011 was slightly lower than that in 2012, because of the severe drought in 2011, with the annual rainfall only 1/4 of that in 2012.
The spatial distribution of the annual ET in the five years, had a different pattern because of the change in the planting structure. In 2015, the ET of wheat, maize, sunflower and the interplanting during the major growing seasons (April to October) were 493 mm, 603 mm, 497 mm, and 709 mm, respectively.  The five-year continuous regional estimation of the ET was calculated over the entire Jiefangzha Irrigation Field. Figure 7 shows the spatial distribution of the annual ET total from 2011 to 2015. The spatial mean ET, during the crop growing season (April to October) of five years over the croplands, were 530 mm, 604 mm, 563 mm, 530 mm, and 551 mm, respectively. The precipitation during the same period from 2011 to 2015, were 45.7 mm, 207.1 mm, 102.8 mm, 141.8 mm, and 111.6 mm, respectively. Despite the annual rainfall, it varied a lot in the irrigation field, due to the sufficient irrigation from the Yellow River, the ET in the study area remained at a stable level. The annual ET in 2011 was slightly lower than that in 2012, because of the severe drought in 2011, with the annual rainfall only 1/4 of that in 2012.
The spatial distribution of the annual ET in the five years, had a different pattern because of the change in the planting structure. In 2015, the ET of wheat, maize, sunflower and the interplanting during the major growing seasons (April to October) were 493 mm, 603 mm, 497 mm, and 709 mm, respectively.

Spatial Distribution of the SWC
The empirical parameters c and d in Equation (15), were calculated using different crop types, including wheat, maize, sunflower, etc. Figure 8 shows the scatter plot of the estimated SWC using the remote sensing data on the clear days compared with the in-situ measurements by the CTMS-On line system at a 10 cm layer, during the growing season (April to October) in 2014. The SWC profiles measured by the CTMS-On line system were already validated by the drying and weighting method sampled twice a month near the instruments. The determination of coefficient of sunflower, maize, wheat and pepper

Spatial Distribution of the SWC
The empirical parameters c and d in Equation (15), were calculated using different crop types, including wheat, maize, sunflower, etc. Figure 8 shows the scatter plot of the estimated SWC using the remote sensing data on the clear days compared with the in-situ measurements by the CTMS-On line system at a 10 cm layer, during the growing season (April to October) in 2014. The SWC profiles measured by the CTMS-On line system were already validated by the drying and weighting method sampled twice a month near the instruments. The determination of coefficient of sunflower, maize, wheat and pepper were 0.82, 0.87, 0.73, and 0.85, respectively, which demonstrated that the SWC profiles measured by the CTMS-On line system were reliable.
The results showed that the estimated SWC, by the remote sensing data, was more discrete. Thus, some croplands tended to suffer from water scarcity, such as wheat and sunflower. Figure 8a (sunflower) illustrates a weak relationship between the measured and the estimated SWC (R 2 = 0.357), and the wheat simulation results fit well with the observation. However, some sufficient irrigation croplands, such as pepper, the measured SWC was maintained at a relatively high level, therefore the remote sensing estimation not perform well.
Based on the calibration results of every single type of cropland, the spatial distribution of the SWC of the whole irrigation field, in 2014, was calculated. The planting structure distribution with a spatial resolution of 30 m is shown in Figure 4. The cropland of the Jiefangzha Irrigation Field was divided into four parts, i.e., wheat, maize, sunflower, and pepper, and then combined together.
The results showed that the estimated SWC, by the remote sensing data, was more discrete. Thus, some croplands tended to suffer from water scarcity, such as wheat and sunflower. Figure 8a (sunflower) illustrates a weak relationship between the measured and the estimated SWC (R 2 = 0.357), and the wheat simulation results fit well with the observation. However, some sufficient irrigation croplands, such as pepper, the measured SWC was maintained at a relatively high level, therefore the remote sensing estimation not perform well. Based on the calibration results of every single type of cropland, the spatial distribution of the SWC of the whole irrigation field, in 2014, was calculated. The planting structure distribution with a spatial resolution of 30 m is shown in Figure 4. The cropland of the Jiefangzha Irrigation Field was divided into four parts, i.e., wheat, maize, sunflower, and pepper, and then combined together. Figure 9 shows the spatial distribution of the SWC in four typical days, during the growing season, and the white pixels indicate non-cropland. The results show that at the early stage of growing season, the SWC at the surface layer is relatively small, which means that before the irrigation, during the early stage, the Jiefangzha Irrigation Field suffered, to some extent, of water scarcity. However, during July and August, the SWC of the surface layer remained at a high level because of the sufficient precipitation and irrigation.  Figure 9 shows the spatial distribution of the SWC in four typical days, during the growing season, and the white pixels indicate non-cropland. The results show that at the early stage of growing season, the SWC at the surface layer is relatively small, which means that before the irrigation, during the early stage, the Jiefangzha Irrigation Field suffered, to some extent, of water scarcity. However, during July and August, the SWC of the surface layer remained at a high level because of the sufficient precipitation and irrigation.

Discussion
In our study, the method of the ISODATA cluster algorithm and spectral matching was used to classify the crops in the Jiefangzha Irrigation Field of the Hetao Irrigation District, and the spatial distribution of the planting structure was obtained. The ISODATA can automatically adjust the number and center of the categories in the process of clustering, so that the clustering results can be closer to the objective and the real clustering results. It is the basic data for the agricultural water resources management. According to the classification results in 2014 and 2015, the total planting area of wheat and sunflower in the field was more than 50%, which is basically consistent with the statistical data. Different from the statistical data, the planting structure extracted by remote sensing had

Discussion
In our study, the method of the ISODATA cluster algorithm and spectral matching was used to classify the crops in the Jiefangzha Irrigation Field of the Hetao Irrigation District, and the spatial distribution of the planting structure was obtained. The ISODATA can automatically adjust the number and center of the categories in the process of clustering, so that the clustering results can be closer to the objective and the real clustering results.
It is the basic data for the agricultural water resources management. According to the classification results in 2014 and 2015, the total planting area of wheat and sunflower in the field was more than 50%, which is basically consistent with the statistical data. Different from the statistical data, the planting structure extracted by remote sensing had spatial distribution information, which can provide the data support for the subsequent diagnosis of the nitrogen concentration and tracing of the non-point source pollution [26]. At the same time, we also noticed that, in addition to the classification method used in our study, many scholars have proposed several classification methods, such as the multisource remote sensing classification [27][28][29], the object-oriented classification [30][31][32], the machine learning classification [33][34][35], and so on. These methods have been well applied in specific areas. Sun et al. [36] used the 1D-CNN machine learning classification method to identify and extract crops in the Weishan Irrigated District, Shandong Province, and achieved a satisfactory accuracy and provided good basic data for the study of the crop water consumption and agricultural water resource management. Belgiu et al. [37] used the pixel-based and object-oriented methods to map farmland in Romania with an accuracy ranging from 78.05% to 96.19%. The deficiency of our study is that there is no comparative analysis of the various classification methods, which may be a key study in the future.
The SEBS model was used to estimate the ET in the Jiefangzha Irrigation Field, from 2011 to 2015, which can provide a reference for the development of the crop irrigation systems. At present, there are many algorithms for the ET estimation. Cheng et al. [10] used the SEBAL model to produce the ET products in China, from 2001 to 2018, which provides reference for improving remote sensing algorithms. Grosso et al. [38] used Landsat-8 images to estimate the ET using the SEBAL model on a maize field in Italy and compared it with the Food and Agriculture Organization (FAO) method. The results showed that the estimated ET was in a good agreement with the ET calculated by the FAO method.
In this research, the farmland was divided into wheat, corn, sunflower, pepper, and other different underlying surfaces. The SWC of the crop growth period in the Jiefangzha Irrigation Field, from April to October 2014, was estimated. The purpose is to provide reference for the irrigation of different crops. Many scholars [7,9,39,40] have carried out remote sensing inversion of farmland soil moisture but have not classified the land features. At the same time, the validation results of the SWC in the maize farmland were not satisfactory, which may be related to the small amount of validation data (just 24). However, when some scholars [39] conducted the SWC estimation using the remote sensing method, more than 50 SWC verification points were obtained, and good verification results were obtained. Therefore, in future studies, if the conditions permit, more measured data should be collected for the model verification. Having obtained the spatial distribution information of the soil moisture, the next step would be to select the typical crops and typical farmlands in the studied region, to carry out the experimental research on the crop wilting moisture as the threshold of the lower water content. Then, the irrigation area is extracted according to the threshold value, and the irrigation water demand of each month is calculated and compared with the actual irrigation water consumption. Then, the irrigation forecast and dispatch need to be considered. The spatial resolution of the regional SWC, obtained in this study, is 250 m. This may not be enough to meet the needs of a more refined agricultural water management. However, it also has the ability to access the water demands of large regions through remote sensing. In the future, the combination of optical and microwave remote sensing can be used to obtain the daily SWC with a spatial resolution of 20 m. This can provide more valuable data for the fine management of the agricultural water resources at a regional scale.

Conclusions
In this research, Landsat-7 ETM+ and MODIS images were used to estimate the spatial distribution of the crop planting structure, the ET, and the SWC in the Jiefangzha Irrigation Field. The study applied the ESTAFM to generate synthetic Landsat-7 ETM+ images with MODIS and constructed the high spatial and temporal NDVI data sets. By using the ISODATA cluster algorithm, the planting structure of the Jiefangzha Irrigation Field was effectively extracted in 2014 and 2015. The positional accuracy of maize, sunflower, wheat, and interplanting are 0.81, 0.80, 0.90, and 0.82, respectively. Based on the SEBS model, the daily ET was estimated, using the long-term MODIS images. The annual ET from 2011 to 2015, well presented the spatial distribution of the ET within the study area. According to the estimated ET and CWSI, the surface layer of the SWC was calibrated and applied to the Jiefangzha Irrigation Field. The results were validated, using four in-situ measurements of the soil water profile instruments at different croplands. The validation results showed that the estimated SWC, by remote sensing on the underlying surface of wheat and sunflower, showed a good robustness, the R 2 was 0.748 and 0.357, respectively, the RMSE was 2.61% and 2.309%, respectively, and the MAE was 2.249% and 1.975%, respectively. However, for maize and pepper with more irrigation times, the SWC estimation results based on the CWSI were poor, indicating that the method was more sensitive to soil drought and suitable for the crop SWC estimation with less irrigation and drought tolerance.