Assessing the Variation in Maize Water Footprint Under Different Tillage Practices: A Case Study from Jilin Province, China

Abstract

Studying the impact of different tillage practices on crop water consumption can help us identify optimal tillage practice choices. The traditional tillage (TT) and conservation tillage (CT) methods are the dominant practices in Jilin Province, China. Few studies have explored the differences in crop water consumption between TT and CT. To address this knowledge gap, this study utilized maize as its research object and employed the water footprint (WF) as the indicator to assess crop water consumption under TT and CT. This study aimed to investigate when differences in water consumption between TT and CT appear and whether the differences are significant. The results of this study demonstrated that the total WF under CT (339.65 m³ t⁻¹) was less than that under TT (378.19 m³ t⁻¹), and the spatial difference was distinct. The total WF exhibited a clear change under different CT durations. At the initial stage of CT implementation, the total WF decreased slightly compared to that under TT. With an increase in CT duration, the total WF was significantly reduced. The findings of this study demonstrate that CT is an effective measure to ensure sustainable crop production and that it could lead policymakers to choose CT to reduce water consumption.

1. Introduction

A large amount of water resources are consumed in the process of crop production [1]. Therefore, accounting for water consumption is crucial. Irrigation water usually received the most attention in the past, and soil water consumption was often not considered, meaning only physical water was considered. In order to reflect the total water consumption during the crop growth period, the crop water footprint (WF) has been introduced as an indicator; this not only includes the irrigation water used in crop production (called the blue water footprint (WF_blue)) but also the volume of precipitation consumed in crop production (called the green water footprint (WF_green)) [2]. The WF is a widely applied indicator for measuring water consumption during the process of crop production, and its study under different tillage practices should be a point of focus in future research.

Usually, there are several tillage practices in a certain region; traditional tillage (TT) and conservation tillage (CT) are the dominant ones, the water consumption of which may vary. The TT method includes ploughing and ridging without residue coverage, and the CT method, as a special tillage practice, includes reduced tillage or no-tillage techniques that leave at least 30% crop residue on the soil surface. According to studies of these tillage methods, TT easily leads to higher evaporation, a lower amount of soil water, and continuous pressure on agricultural water resources [3], while CT can affect surface hydrological characteristics relative to TT, resulting in a decrease in runoff and evaporation and an increase in infiltration, which alters the agricultural water use [4]. The effects of water consumption under TT and CT are different in theory, and the literature is also unclear on when these differences appear and whether they are significant. Answering these questions clearly would contribute to a better understanding of the characteristics of water consumption under TT and CT.

Currently, studies on the effect of TT and CT on WF have been conducted at the site scale. In one study, the use of reduced-tillage and no-tillage techniques at the site scale led to a decline in the WF compared to TT [5]. The CT method increased the soil moisture and enhanced the supply of nutrients and water [6,7], also altering the WF. Although these site-scale studies revealed the WF-based differences between TT and CT, the literature is lacking appropriate large-scale research on the widespread promotion of CT. Therefore, research on a larger spatial scale is essential to understand the spatiotemporal effects of different tillage practices on crop water consumption.

Remote sensing technology is fast, accurate, and can handle high-spatial-resolution data, which means it can be used efficiently to conduct large-scale research and capture land surface information quickly in regional areas [8]. Remote sensing technology can monitor and identify TT and CT by assessing crop residue cover based on the normalized difference tillage index (NDTI) [9] and achieve higher prediction accuracy [10,11]. For example, in a study where the researchers used the NDTI to estimate maize residue cover (MRC) in Lishu County, China, this method was demonstrated to be highly effective, with there being a determination coefficient (r²) of 0.78 and root mean square error (RMSE) of 0.676% between the predicted and measured MRC [12]. Similarly, utilizing the NDTI in conjunction with remote sensing datasets led to the efficient identification of an 8.41% increase in CT and a 14.45% decrease in TT in central Ohio [13], also leading to the generation of residue maps covering the 150,000 km² study area across South Dakota, North Dakota, and Minnesota, with an overall accuracy reaching 70% [14]. Although remote sensing data has been used in WF assessments [15,16], the evapotranspiration and crop yield values reported in prior studies were still based on TT, ignoring the implementation of CT. Therefore, the results of WF calculation are not very accurate due to the mixed use of TT and CT. Hence, it is necessary to evaluate the WF under TT and CT accurately and explore the changes in the WF under different CT durations and the differences in the WF between TT and CT using remote sensing technology.

Jilin Province is a main grain-producing region in China and it is drought-prone [17]. In order to maintain an adequate and reliable water supply and promote sustainable agricultural development, CT has been vigorously promoted in Jilin Province [18], and it is the most extensively used practice for maize farming in the region. This study quantitatively evaluated the maize WF under TT and CT at the pixel level via incorporating multi-source data and determined the differences in WF between TT and CT in Jilin Province from 2010 to 2020. The main objectives of this study were to reveal the trends in WF under different CT durations, to demonstrate whether the WF reduces at the beginning of CT implementation and whether the WF continues to decline with the continuous implementation of CT, and establish what the magnitude of this decline is. The findings of this study could not only contribute to a better understanding of the importance of CT in effective water resource utilization but also provide a basis for policymakers to promote CT in water-scarce and grain-producing areas.

2. Materials and Methods

2.1. Study Area

Jilin Province is located in the middle of northeastern China (40°52′–46°18′ N, 121°38′–131°19′ E) (Figure 1A) and is one of three golden maize belts in the world. Black soil is widely distributed throughout the region, and the soil texture is suitable for planting maize. The maize planting area accounts for about 70% of the crop planting area in Jilin Province. The area’s heat and precipitation are affected by the East Asian continental monsoon climate. From 2010 to 2020, the average values of sunshine hours, temperature, and precipitation during the maize growing season in Jilin Province were 1121.92 h, 19.82 °C, and 517.99 mm, respectively. According to natural and geographical factors, Jilin Province is usually divided into three parts: east, middle, and west.

2.2. Data Sources

The data adopted in this study included remote sensing data, meteorological data, statistical data, and other types of data (Table S1). The remote sensing data were obtained from the Landsat, Shuttle Radar Topography Mission (SRTM), and moderate resolution imaging spectroradiometer (MODIS) datasets. Landsat and SRTM data were acquired at https://earthengine.google.com (accessed on 20 November 2024). The adopted MODIS datasets were from MOD16A2 and MOD14A1, which detail the dynamics of 8-day ET and fire data, respectively. All MODIS data can be downloaded at https://ladsweb.modaps.eosdis.nasa.gov (accessed on 20 November 2024).

Thirty-seven meteorological stations located in and around Jilin Province were selected for this study (Figure 1B). Daily meteorological data from 2010 to 2020 were acquired from the China National Surface Weather Station in the China Meteorological Sharing Network (http://data.cma.cn, accessed on 20 November 2024). These data included sunshine hours, precipitation, temperature, wind speed, and relative humidity values. The meteorological data were all produced using the inverse distance weighted (IDW) interpolation method with a spatial resolution of 500 m in ArcGIS.

The chemical fertilizer (CF) and agricultural machinery (AM) we used were the same as those listed in the statistical yearbook in order to compare the differences in influencing factors between CT and TT. To obtain the spatial grid distribution for maize CF and AM, a principal component analysis was conducted, and the rasterization method based on grid GDP data was used [19,20,21,22,23] (Supplementary S1).

2.3. Methods

The research framework included three parts (Figure 2): (1) extracting the maize planting area based on the Google Earth Engine, (2) identifying the CT, and (3) calculating the WF under TT and CT.

2.3.1. Extracting the Maize Planting Area

A total of 29,376 sample points were selected through field surveys and high-resolution image interpretation, of which 70% were used as the training set, and the remaining 30% were used as the validation set. Based on Landsat images from Google Earth Engine (https://earthengine.google.com, accessed on 20 August 2024), the first seven 30 m resolution bands were selected as the basic spectral data in the Landsat classification experiment, and vegetation indices—the normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), normalized difference water index (NDWI), and normalized difference built-up index (NDBI)—and topographic variables, such as digital elevation model (DEM) and slope data from the SRTMGL1v003 dataset (https://lpdaac.usgs.gov/, accessed on 20 August 2024), were also selected as the feature recognition datasets for crop classification and identification. A random forest model was used to classify the crops, and the total number of random forests was set to 100. The number of optimal segmentation nodes was set to four. The above operations were all run via the Javascript API on the Google Earth Engine. The accuracy of the identification results was assessed using Producer Accuracy, User Accuracy, Overall Accuracy, and Kappa Coefficient (Supplementary S2), and these four indices all exceeded 0.8. Moreover, the r² and RMSE between the estimated maize planting area and recorded in the Statistical Yearbook of Jilin Province were 0.70 and 5.19 × 10⁵ hm², respectively (Supplementary S3, Figure S1). These results suggest that the maize planting area is both reliable and accurate.

2.3.2. Extracting the CT Area

MRC was used to identify the CT area, which was greater than 30% [24]. An MRC estimation model was constructed by exploiting the relationship between the field MRC and the NDTI calculated via the use of Landsat images. The NDTI has been proved to be a stable and high-precision tillage index for evaluating crop residue [10]. To avoid the detrimental effects of multi-collinearity, partial least squares regression was applied, helping to establish a model between MRC and NDTI [25]. Partial least squares regression is a robust data analysis technique with high accuracy [10]. The field measurement and MRC survey data were collected at 485 fields in Jilin Province from late-April to mid-May from 2010 to 2020. Overall, 197 of these fields used TT practices, while 288 used CT practices. Two-thirds of the fields were randomly selected to construct the MRC estimation model, and the remaining third was used to validate the model. In addition, straw burning was as a supplementary condition for CT identification. If a fire point was detected in an area, it was be excluded from the CT classification. Therefore, fire data was incorporated to identify CT. The area of CT and TT was resampled to a spatial resolution of 500 m using ArcGIS.

2.3.3. Estimating the Maize WF

The growth period of maize is from May 1 to September 30, and it can be divided into six key nodes: sowing, seedling, jointing, tasseling, milky, and mature [26]. For TT, because of the spring drought, the maize needs to be irrigated at the sowing–seedling stage, creating a microenvironment with sufficient soil water to ensure germination and seedling establishment; the maize then receives no irrigation water at any other stage [27,28]. For CT, because of the crop residue on the surface of soil, the maize does not need to be irrigated. Hence, the value of WF_blue is 0 in the CT area. The calculation of the WF values of maize production at the pixel level can be carried out as follows:

$$W{F}_{total}=W{F}_{blue}+W{F}_{green}$$

(1)

$$W{F}_{blue}=10\times {\displaystyle \frac{E{T}_{blue}}{Y}}=10\times {\displaystyle \frac{max\left(0,E{T}_{a1}-P_{e1}\right)}{Y}}$$

(2)

$$W{F}_{green}=10\times {\displaystyle \frac{E{T}_{green}}{Y}}=10\times {\displaystyle \frac{min\left(E{T}_{a1},P_{e1}\right)+E{T}_{a2}}{Y}}$$

(3)

where WF_total is the total WF of maize production (m³ t⁻¹); WF_blue is the blue WF (m³ t⁻¹), which represents the irrigation water during the sowing–seedling growth stage; and WF_green is the green WF (m³ t⁻¹), which represents the volume of the precipitation consumed in maize production. Y is the maize yield (t hm⁻²), ET_green and ET_blue are the green and blue water evapotranspiration (mm), ET_a1 and P_e1 are the actual evapotranspiration and effective precipitation (mm) during the sowing–seedling growth stage, and ET_a2 is the actual evapotranspiration (mm) for the jointing, tasseling, milky, and mature growth stages. Based on the daily precipitation, the effective precipitation was counted via the U.S. Department of Agriculture Soil Conservation method [29]. The actual evapotranspiration under CT and TT came from the MOD16A2 product, distinguishing based on the spatial distribution of CT and TT, and the maize yield was estimated using remote sensing technology.

The crop yield has a strong relationship with the net primary productivity (NPP) [30], and the yield estimation based on this relationship has been widely applied [31,32]. Hence, the yield of maize at the pixel level is calculated as follows:

$$Y={\displaystyle \frac{NPP\times \alpha \times HI}{1000}}$$

(4)

$$HI={\displaystyle \frac{y_s}{y_s+1}}$$

(5)

where NPP is the net primary productivity (gC m⁻² year⁻¹), α is the conversion ratio between carbon content and dry matter, HI is the harvest index [31], and y_s is the grain-to-stalk yield ratio. The NPP estimation was based on the Carnegie–Ames–Stanford approach (CASA) model.

2.3.4. Lindeman, Merenda, and Gold (LMG) Method

To identify the main influencing factors causing the variation in the maize WF under TT and CT, the Lindeman-Merenda-Gold (LMG) method based on multivariate linear models was used [33], ran in the ‘relaimpo’ R package (version 4.1.0). This method can effectively eliminate the order effect between each regression variable, overcome multi-collinearity problems, and accurately measure the relative importance of each influencing factor [28,34]. Therefore, the LMG method was used to quantify the degree of influence of WF under TT and CT in this study, along with other factors of climate and agricultural management.

2.3.5. Verification Analysis

The r² and RMSE were used to verify the degree of explanation and accuracy of the models in this study. The calculation formulas are as follows:

$$r^2=1-{\displaystyle \frac{{\displaystyle {\sum }_{i=1}^N{\left(y_i-{\widehat{y}}_i\right)}^2}}{{\displaystyle {\sum }_{i=1}^N{\left(y_i-{\overline{y}}_i\right)}^2}}}$$

(6)

$$RMSE=\sqrt{{\displaystyle \frac{1}{N}}{{\displaystyle {\sum }_{i=1}^N\left(y_i-{\widehat{y}}_i\right)}}^2}$$

(7)

where N is the number of samples, y_i is the field measurement value in sample i, and $\widehat{y_i}$ is the model prediction value in sample i. In general, models with good performance generally have higher r² and lower RMSE values. Additionally, t-test results with p < 0.05 were reported as statistically significant.

3. Results

3.1. Mapping the Spatial Distribution of CT and TT

The MRC model is the key to distinguish TT and CT; our MRC model was constructed based on the MRC values measured in the field and the NDTI calculated via the remote sensing data. The fitted line was y = 1836.30x³ − 117.29x² + 118.55x + 38.95 (y represents the MRC; x represents the NDTI), and the r² value for the NDTI and MRC was 0.69 (Figure 3A), which indicated a good relationship between the NDTI and MRC. To further validate the MRC model, the predicted values were compared with the measured MRC values. The results showed that the predicted MRC increased with an increase in the measured MRC, and the fitted line was y = 0.77x + 11.22 (y represents the predicted MRC; x represents the measured MRC); the RMSE was 3.15% (Figure 3B), indicating that the application of this model in this study was effective.

After eliminating abnormal values, along with the survey data and fire data, the spatial distribution of the CT area in Jilin Province from 2010 to 2020 was calculated. To evaluate the accuracy of CT area estimation, the relationships between the estimated values and the reported values were explored. The estimated area of CT showed very high consistency with the data recorded in the China Agricultural Machinery Industry Yearbook pertaining to the period from 2010 to 2020 (r² = 0.89, RMSE = 1.59 × 10⁵ hm²) (Figure 3C). Additionally, in a previous investigation of land management in Jilin Province, an adoption rate of 12% of CT was reported for cropland under maize production in 2015 [35], similar to our finding (13.03%). Moreover, a comparison of this study with other studies that used the same definition of CT revealed good spatial agreement [10,36].

The spatial distribution of the CT area in Jilin Province from 2010 to 2020 was mapped (Figure 4). In 2012, a small number of maize planting areas were implemented via CT, and the implementation area accounted for approximately 0.42% of the maize planting area in Jilin Province. The CT area increased significantly after 2015, and the area under CT accounted for approximately 25.91% of the maize planting area in 2020. Notably, the larger CT area was distributed mainly in the middle region, accounting for approximately 43.48% of the maize planting area by 2020. CT requires a high degree of mechanization, and most straw-returning machines need to be operated in areas with plain land. The middle region of Jilin Province is suitable for implementing CT, with its flat terrain and high levels of mechanization. Furthermore, most farmers in the middle region have come to realize that implementing CT can increase yield and, therefore, ensure they generate a greater income. Therefore, we found that the middle region of Jilin Province was the area wherein CT was most actively implemented.

3.2. Spatiotemporal Variations in WF Values Under TT and CT

3.2.1. Validation in Maize Yield

To obtain the spatial distribution of WF, maize yield at the pixel level was estimated based on remote sensing data. The estimated maize yield in this study was validated against the actual yield at the survey sample points. The results showed that the r² was 0.81, and the RMSE between the 370 survey sample points and the estimated yield was 0.62 t hm⁻² (Figure 5), indicating that the estimated yield in this study was in good agreement with the actual yield data. Meanwhile, the spatial distribution of the estimated yield was consistent with the findings reported in [37]. Therefore, the comparison results indicate that the methods used in this study can be considered reliable.

3.2.2. The Temporal Variation in the WF Values

The WF of maize at the province level was not obviously different from 2010 to 2020 (Figure 6). Due to the special irrigation schedule, the average WF_blue was always low, only amounting to 2.73 m³ t⁻¹ under TT. The variation in WF_blue was primarily influenced by yield and climate parameters, and the lowest average WF_blue value was observed in the years with ample precipitation (Figure S2). The average WF_green and WF_total values were 370.74 m³ t⁻¹ and 373.47 m³ t⁻¹, respectively. Regarding the WF values under TT, WF_blue was 2.73 m³ t⁻¹, WF_green was 375.46 m³ t⁻¹, and WF_total was 378.19 m³ t⁻¹. While for the WF values under CT, WF_blue was always 0 m³ t⁻¹, and WF_green was 339.65 m³ t⁻¹, which was WF_total. Under CT, WF_total was not only smaller than that of the average but also smaller than that under TT. Therefore, CT not only changed the structure of crop water use but also reduced the amount of crop water consumption and improved water use efficiency.

3.2.3. The Spatial Variation in the WF_total Values

The spatial distribution of WF_total showed dynamic changes from 2010 to 2020. In general, WF_total showed a spatial distribution characteristic of being lower in the middle region and higher in the east and west (Figure 7). The WF_total values in the middle region were mainly concentrated in the range of 200 to 400 m³ t⁻¹, while in the eastern and western regions, the WF_total values were often greater than 400 m³ t⁻¹. Because the middle region is one of the world’s golden maize belts, the better growing environment for maize leads to a higher yield compared to other regions, resulting in a lower WF_total value. For the TT areas in the east and west, the spatial distribution of WF_total did not clearly change. For the CT areas, the WF_total values were mainly concentrated in the range of 200 to 400 m³ t⁻¹ from 2012 to 2014, with no significant spatial variation. After 2015, the spatial variation in WF_total exhibited an obvious fluctuation. The WF_total under CT in the south of the middle region was usually lower than that in other regions. This may be due to the fact that the south of the middle region was the first area to implement CT and therefore had been practicing CT for the longest duration. Although WF_total under CT was lower than that of TT, it changed in response to local climate change. For example, in 2020, WF_total under CT was larger than those in other years because there were three typhoons from August to September in Jilin Province, which severely damaged maize yield, resulting in an overall higher WF_total value. However, WF_total in the CT areas was still less than that in the TT areas because there was clear lodging resistance under CT.

3.3. Difference in WF Under Various CT Durations

The changes in maize WF_total from 2012 to 2020 indicated that the WF_total under CT was less than that under TT. While the CT duration was different, the effect of CT duration on maize WF needed to be clarified. For this paper, we explored all CT pixels from 2012 to 2020 and identified the changes in WF_total at different tillage ages. It was found that the WF_total under CT changed significantly with a duration of 1 to 9 years, fluctuated greatly with a duration of 1 to 4 years, and exhibited a slightly downward trend with a duration of 5 to 6 years, also experiencing a significant and sustained decrease with duration a of 7 to 9 years (Figure 8). The implementation of CT has greatly changed the maize growing environment. In the early stage of CT adoption, straw decomposition mainly took place, and the level of nutrient release was low and uneven. This, coupled with disturbances caused by pests and weeds, meant that the WF_total showed an irregular fluctuating trend within 4 years after implementation of CT. WF_total showed a clear decreasing trend and a smaller range of variation after the implementation of CT exceeded 6 years. This is mainly because the soil organic matter content continues to increase under the CT method. Under CT, the soil organic carbon sequestration rate at a depth of 0–20 cm in Northeast China was 0.953 mg C hm⁻² year⁻¹, much higher than China’s national average (at the same depth) of 0.157–0.390 mg C hm⁻² year⁻¹ [38,39]. Thus, CT is likely to be an effective management strategy for sequestering soil organic carbon and restoring soil fertility, and it also alleviates issues such as weeds and pests to a certain extent, leading to improved and more stable maize yields. With the continuous advancement of CT practice, its impact on the WF_total of maize production gradually emerged. Generally, it can be inferred that an increase in CT duration could further reduce WF_total values and thus increase water use efficiency.

4. Discussion

4.1. Effects of CT on Maize WF

Yield and ET_a are the determinants of WF_total, but how the two variables affect the WF_total of maize production under CT is not clear. The changes in ET_a and yield under CT and TT from 2012 to 2020 in Jilin Province are shown in Figure 9. Compared with TT, CT has a higher yield per unit area and a smaller difference in ET_a, resulting in a lower WF_total for maize production under CT, suggesting that CT reduced the WF_total primarily by increasing maize yield. This is mainly because the practice of CT can establish a reasonable ploughing layer in 0–15 cm of soil and improve soil properties and water-holding capacity. The maize yield per unit area increased under CT [40]; thus, the WF of maize production was decreased. A study based on remote sensing data also found larger yield values under CT compared to TT, which could also result in a decreasing WF under CT [41]. On the other hand, CT increased soil moisture and enhanced water availability [42,43], resulting in no irrigation water (blue water) being used for maize growth, which could also reduce the WF_total of maize production to a certain extent.

An analysis of the differences in influencing factors would provide a better understanding of WF changes under CT and TT. The factors of climate and agricultural management were the main influencing factors of WF [44,45]. In this study, we selected sunshine hours, precipitation, relative humidity, temperature, and wind speed as the climate factors, choosing CF and AM as the factors of agricultural management, and used the Lindeman, Merenda, and Gold method to quantitatively evaluate the relative importance of these factors and their changes. Regardless of whether CT or TT was used, the climate factors were the dominant factors affecting WF in Jilin Province. However, 28.29% pixels were more vulnerable to agricultural management, where the concentrated distribution area of CT was located in the south of the middle region (Figure 10). These pixels were mainly characterized by a longer duration of CT, along with a higher degree of agricultural mechanization. It could be inferred that the role of agricultural management has increased, and the effects of climate factors on maize WF could be reduced in CT areas. Several studies also indicated that climate factors were a major factor affecting the WF and that CT could mitigate the impact of climate factors on the WF [28,44].

CT can withstand natural disasters and maintain a high production level, as fully demonstrated in Jilin Province. In 2020, there were three typhoons concentrated in late August to early September in Jilin Province (Figure 11A), which coincided with the critical maize growth stages of tasseling and milking. During these stages, the maize was particularly susceptible to heavy rainfall associated with typhoons, resulting in soil loss around the roots and plant lodging, seriously affecting maize yield [46]. Simultaneously, the WF_total of maize in Jilin Province experienced a significant abnormal increase in 2020, whereas the WF_total under CT was still less than that under TT because of the lodging resistance. The maize yields were significantly different under CT and TT within 100 km of the typhoon track path. The maize yield under CT was 8.11 t hm⁻², significantly higher than that under TT (7.54 t hm⁻²) (Figure 11B). Therefore, CT is an effective measure for mitigating the adverse effects of natural disasters on WF and ensuring the sustainability of maize production.

4.2. Changes in WF Under Different CT Durations

The WF of maize production was not stable in the early stages of CT implementation, and after a certain amount of time, the WF showed a continuous decline (Figure 8). The time nodes of this WF decline were different, including both a slight decrease and a significant decrease. Therefore, the mechanism of the impact of different CT durations on WF needs to be understood. CT can minimize physical disturbance to the soil, avoiding the damage to the soil structure caused by TT [47]. Simultaneously, practices like straw mulching increase the soil organic matter content, promote the formation and stability of soil aggregates, and enhance the soil water-holding capacity [48]. An improved soil structure reduces the water evaporation and deep percolation, thereby increasing the efficiency of precipitation use. Over time, the continuous increase in the soil organic matter and microbial activity further enhances the soil structure, water-holding capacity, and water supply stability, ensuring stable water availability during the critical maize growth stages and improving yield [49,50]. However, these effects of CT encompass cumulative processes which are not obvious in the early stages of implementing CT; therefore, the WF fluctuates in the initial stage and decreases slightly in the later stage. Only the WF under long-term CT is significantly reduced, because the soil permeability and stability continue to improve, surface runoff and soil erosion are further reduced, the precipitation use efficiency is significantly enhanced, and the yield increases significantly with the prolonged implementation of CT [51]. Therefore, with increasing CT duration, these combined effects lead to a notable improvement in maize water use efficiency, increasing the dry matter production per unit of water and thereby reducing the water resource consumption per unit of yield, significantly lowering the WF of maize.

4.3. Limitations and Uncertainties

There are some uncertainties in this study. Soil moisture can affect NDTI values and, therefore, affect the accuracy of CT identification, and this study only reduced the disturbance of soil moisture by selecting data from continuous sunny days, which may have affected our accuracy in identifying CT to some extent. Although the harvest index was regarded as unchanged at the county level [52], some errors in yield estimation may be experienced when using the harvest index at the county scale. Moreover, changes in the length of the maize growing season were not considered in this study, which also could have caused errors in the estimation of ET_a and yield. In future, optical remote sensing data should be combined with radar data to reduce the effect of soil moisture on CT identification, the harvest index should be spatially differentiated with machine learning models using multi-source remote sensing data, and AquaCrop models should be used to simulate the crop growing season.

5. Conclusions

Studying the variation in WF under different cultivation systems can help clarify the difference in crop water consumption between CT and TT and lead to a better understanding of the impact of CT on WF. The results of this study showed that the maize WF_total under CT was not only less than that under TT but also indicative of a clear change under different CT durations. Compared to the TT practice, WF_total for maize decreased slightly at the initial stages of CT implementation and decreased significantly after 6 years of CT implementation, indicating that the longer the CT duration, the lower the maize WF_total. Moreover, this practice can reduce the effects of climate factors on maize WF, mitigate the adverse effects of natural disasters, and improve crop water use efficiency. Overall, these results help to improve our understanding of how maize WF values change under CT and provide a basis for exploring sustainable crop production and food security under this practice.