Skip to content
Frontiers
Search

Original Research

Frontiers in Agronomy ·

Volume 8 - 2026 | https://doi.org/10.3389/fagro.2026.1850435

Impact of sand-mixing and nutrient treatments on the growth of foxtail millet in moderately saline-alkali soil

  • Wei Wang1,2,3
  • Shujin Qiu2,3
  • Shubo Yang2,3
  • Ximing Duan1
  • Zhenyu Liu2,3
  • Qingliang Cui2,3*corresponding author
  1. 1College of Urban and Rural Construction, Shanxi Agricultural University, Taigu, Shanxi, China
  2. 2College of Agricultural Engineering, Shanxi Agricultural University, Taigu, Shanxi, China
  3. 3Shanxi Key Laboratory of Key Technologies and Equipment for Agricultural Machinery, Taigu, Shanxi, China
Article metrics0 viewsView details
PDFCiteMetrics

Abstract

Mixing sand into the surface soil, combined with nutrient management, may improve the hydraulic properties of saline–alkali soil and promote crop growth. However, the effects of sand mixing and nutrient management on soil water and nutrient transport and foxtail millet growth remain unclear. This study investigated the effects of sand-mixing ratio and nutrient supply on soil infiltration, water and nutrient distribution, and the agronomic performance of foxtail millet in moderately saline–alkali soil in northern Shanxi Province. One-dimensional vertical ponded-infiltration experiments were conducted using soil columns with five sand-mixing ratios (0%, 5%, 10%, 15%, and 20%) in the 0–20 cm soil layer and three infiltration-water treatments: deionized water and low- and high-concentration NPK solutions. Field experiments combined the same sand-mixing ratios with three compound fertilizer application rates of 1,050, 1,500, and 1,950 kg·ha -1. Soil infiltration characteristics, water and nutrient distributions, and the agronomic traits of foxtail millet were determined. Sand mixing and nutrient supply accelerated wetting-front advancement and enhanced soil infiltration capacity. The time required for the wetting front to reach a given depth decreased with increasing sand-mixing ratio and nutrient concentration. Sand mixing reduced soil water content in the 0–20 cm mixed layer but increased it in the underlying soil, whereas increasing nutrient concentration generally increased soil water content. Correlation analysis showed that the relationships among NO 3 - –N, NH 4 + –N, available phosphorus, and available potassium varied among nutrient treatments. In particular, the correlation between available phosphorus and available potassium became stronger with increasing nutrient concentration. In the field experiment, the agronomic traits of foxtail millet generally improved with increasing fertilizer application rate, and 10% sand mixing combined with a fertilizer application rate of 1,500 kg·ha -1 produced the greatest panicle weight. Appropriate sand mixing combined with nutrient management improved soil water and nutrient conditions and promoted foxtail millet growth. Under the experimental conditions, 10% sand mixing combined with a fertilizer application rate of 1,500 kg·ha -1 showed the best overall performance.

Keywords

1 Introduction

The Datong Basin is located in the northern part of Shanxi Province, with a total area of about 3 million acres of saline-alkali land, accounting for two-thirds of the saline-alkali land in Shanxi Province (Fan, 2023). With high concentrations of CO32- and HCO3- ions, the soil is classified as soda-type saline-alkali soil (Zheng et al., 2021). Soil salinization seriously restricts local food production and sustainable agricultural development. Due to factors such as climate change and unreasonable irrigation, the degree of soil salinization continues to rise (Eswar et al., 2021). The composition and accumulation of salt in saline-alkali soils vary in different regions (Guo and Fan, 2011). The formation and management of saline-alkali soils should be guided by the mechanisms of water and salt movement (Wang et al., 2000), and should regulate soil salt movement in a targeted manner, suppress the harm of saline and alkali to plants, and reduce the adverse impact on the ecological environment. Medium and low saline-alkali land has enormous development potential (Li et al., 2024), and adopting scientific and effective methods for its management, transformation, and utilization is of great significance for national food security.

Various approaches have been proposed for the development and utilization of saline-alkali lands. However, hydraulic engineering measures are constrained by large investment and long construction cycles; chemical amelioration methods (e.g., gypsum-based amendments) face high material costs and potential soil pollution risks; and the use of halophytes for land improvement remains in its early stages (Lu et al., 2026; Qadir and Oster, 2004; Ashraf et al., 2021). Studies have shown that surface sand covering or mixing can improve soil structure, increase soil effective porosity, thereby changing the water and salt transport laws of saline alkali soil and improving its water retention and demand capacity (Wang et al., 2025; Yang et al., 2023; Li et al., 2020; Tan et al., 2018; Eck et al., 2016; Song et al., 2012). Sun et al. (2022) conducted a study on coastal saline soil areas such as the Yellow River Delta and found that a sand incorporation rate of approximately 30% yielded the optimal effect, effectively improving soil water-salt distribution, inhibiting evaporation, and promoting salt leaching. Zhou et al. (2020) conducted soil column experiments with 16 different sand admixture ratios on severely saline-alkali soils in the Hetao Irrigation District, finding that the optimal sand admixture ratio ranges from 18% to 24%. Zhang et al. (2019) studied the effect of sand mixing ratios on the improvement of moderately saline-alkali soils in the Yellow River Delta. They found that adding sand into the topsoil layer reduced soil salinity and increased crop yields, with the best improvement effect achieved by adding 15% sand. Chai et al. (2023) found through indoor pot experiments that when growing wheat on moderately saline-alkali soils in the Yellow River Delta region, mixing 15% river sand with 5 t·hm-² of organic fertilizer increased soil moisture content, reduced soil salinity, and resulted in higher wheat plant height and dry matter yield. Soil salinization can reduce soil fertility and inhibit crop growth and development. There are many studies on the single and dual-factor coupling effects of water, salt, and nitrogen on crop growth (Zhang et al., 2024; Xu et al., 2020; Yue et al., 2015; He and Tian, 2015; Xu et al., 2015), but most were conducted on primary saline-alkali soils and rarely involved the soil environment after sand amendment. As the main fertility factors of soil, the transport and distribution of nitrogen, phosphorus, and potassium are influenced by soil type, fertilization method, crop roots, and other factors, and the mechanism is not fully understood (Huang and Fan, 2018). Furthermore, the inherent patterns of nutrient transport and the coupled mechanism of water, salt and nutrients in sand-amended saline-alkali soils have been rarely reported.

The Datong Basin faces significant challenges due to freshwater resource shortages. Although rivers such as the Sanggan River and Yu River flow through the region and provide abundant sand resources, the utilization rate remains low. To explore a water-saving and efficient saline-alkali soil improvement method, we hypothesize that moderate sand-mixing combined with appropriate fertilization can improve water infiltration, optimize vertical distributions of water, salt, and nutrients, alleviate salt stress, and enhance agronomic traits of foxtail millet. An optimal sand-fertilizer combination exists for salt suppression, nutrient retention, and growth promotion, whereas excessive sand or unsuitable fertilizer may aggravate deep leaching. To test this hypothesis, one-dimensional vertical ponding infiltration experiments were conducted using local river sand as an amendment. Soil infiltration processes, water-salt distribution, and nutrient transport patterns under different treatments were examined. The vertical distribution characteristics governed by the coupled interactions of water, salt, and fertilizer were characterized, and synergistic effects among key soil indicators were analyzed. Based on the data obtained from the soil column experiments, this study aims to introduce multiple machine learning algorithms, including multiple linear regression (MLR), random forest (RF), XGBoost, and LightGBM, to quantify the influence of various factors (sand-mixing ratio, nutrient levels, soil depth, etc.) on soil water, salt, and nutrient transport. Since different algorithms have their own strengths and weaknesses in data adaptability and nonlinear relationship capture, a single algorithm cannot guarantee the fitting effect on this dataset and the reliability of factor influence quantification. Therefore, by comparing the predictive performance of the above models, the optimal algorithm will be selected. Furthermore, a foxtail millet planting experiment was carried out in Huairen City to test saline-alkali soil amelioration through sand-mixing, aiming to identify the optimal sand and fertilizer application scheme and to provide a theoretical basis for the amelioration of soda-type saline-alkali soil.

2 Materials and methods

2.1 Experimental materials

The experimental soil was taken from moderately saline-alkali farmland in Huairen City (39.82°N, 113.10°E), Datong Basin. The field experiment was conducted at the same site, and the foxtail millet variety used was Tonggu 30. To minimize sampling error, disturbed soil samples were collected from a depth of 0–20 cm using a Z-shaped sampling method. Before the experiment, the soil was air-dried, ground, and passed through a 2 mm sieve, then homogenized for use. The contents of ammonium nitrogen (NH4+-N), nitrate nitrogen (NO3--N), available phosphorus (AP), and available potassium (AK) in the soil were 48.65, 7.51, 8.50, and 88.05 mg·kg-1 respectively. River sand was selected as the amendment material, which was washed, air-dried, and passed through a 2 mm sieve. The physical properties of the processed soil were presented in Table 1.

Table 1. Physical properties of soil and sand.

TextureParticle size (mm) and mass fraction (%)Bulk densitySalt content
2–0.50.5–0.020.02–0.002<0.002(g·cm-3)(g·kg-1)
Sandy clay loam0.1742.0043.7214.111.42.51
Sand67.6125.815.610.971.7No detectable

2.2 Test method

The indoor experimental setup is shown in Figure 1. The experiment consisted of a plexiglass Mariotte bottle and a soil column. Both were cylindrical plexiglass columns: the Mariotte bottle was 50 cm in height with an inner diameter of 8 cm, while the soil column was 80 cm height with an inner diameter of 10 cm and was fitted with four 1cm diameter drainage holes at the base. During the experiment, the Mariotte bottle was connected to the water inlet at the top of the soil column via a silicone tube. The bottle mouth was sealed with a rubber stopper, and an air inlet tube was inserted through the stopper, with its lower end extending into the bottle. The water outlet was fixed 2 cm above the soil surface, thereby achieving a constant water head of 2 cm. During water supply, when the water level inside the bottle dropped to the lower end of the air inlet tube, outside air entered the bottle and formed bubbles, which restored the internal air pressure, thus automatically maintaining the constant water head at the outlet. In addition, the top of the soil column was open to the atmosphere, allowing free escape of air from within the soil column during water infiltration, preventing air entrapment or pressure fluctuations that could affect the water supply from the Mariotte bottle. Along the side of the soil column, circular sampling ports with an inner diameter of 2 cm were installed at 5 cm intervals, starting 15 cm from the top. These ports, sealed with rubber stoppers during the experiment, were reserved for soil sampling.

Setup of the soil column experiment.
Figure 1. Setup of the soil column experiment.

The experiment was conducted with five sand-mixing ratios (volume ratio): S0 (0%), S1 (5%), S2 (10%), S3 (15%), S4 (20%); and three infiltration-water treatments: F1 (deionized water), F2 (low NPK, 5 g·L-1 NH4NO3 and 2.5 g·L-1 KH2PO4 prepared with deionized water), F3 (high NPK, 10 g·L-1 NH4NO3 and 5 g·L-1 KH2PO4 prepared with deionized water). To suppress the boundary effect along the inner wall of the organic glass column, the wall was first roughened with 80-grit sandpaper to create a uniform frosted surface. A thin layer of Vaseline (approximately 0.5 mm thick) was then uniformly applied over the abraded surface. Filter paper was placed at the bottom of the soil column, followed by a 5 cm-thick layer of gravel (3–6 mm in diameter) to enhance air permeability and facilitate water infiltration. The soil was packed into the column at a bulk density of 1.4 g·cm-³. To ensure uniformity, it was added in 5 cm layers, with the surface of each layer scarified before the next was added. A total depth of 45 cm of untreated soil was placed first. The top 20 cm consisted of a soil-sand mixture prepared according to the designated sand-mixing ratio. A 2 cm ponding depth was maintained above the soil surface, resulting in a total soil depth of 65 cm. Each treatment was replicated three times. After packing, the columns were allowed to equilibrate for 24 hours before testing.

The experiment was conducted from May to July 2024, in the Hydraulics Laboratory of Shanxi Agricultural University. The soil column surface was marked with a scale to allow observation of the wetting front position. The test was terminated when the wetting front reached the 50 cm depth.

The field experiment was conducted from May to October 2024 in moderately saline-alkali farmland in Huairen City. This region had a temperate continental climate. During the experimental period, the average temperature was 19.42 °C, and annual precipitation reached 370.20 mm, representing a deviation of less than 10% from the multi-year average (approximately 350 mm), which is within the normal range of inter-annual variability. Precipitation is unevenly distributed within the year and concentrated in July and August. The soil type and basic properties were consistent with those presented in Table 1. Five sand-mixing ratios were designed, consistent with those used in the indoor soil column experiment. The application rates of compound fertilizer (N-P2O5-K2O, 15-15-15, i.e., 15 % nitrogen, 15 % phosphorus pentoxide, and 15 % potassium oxide by mass) were set as F1 (1050 kg·hm-²), F2 (1500 kg·hm-²), and F3 (1950 kg·hm-²) (Gao et al., 2025; Feng et al., 2020). Each treatment was replicated three times, resulting in a total of 45 plots, each with an area of 2 m × 5 m. Before tillage, the sieved river sand was evenly spread on the soil surface, followed by rotary tillage to a depth of 20 cm. All treatments were managed consistently with standard field management practices.

2.3 Measured parameters

The elapsed time and the position of the wetting front were recorded. Upon completion of the experiment, soil samples were immediately collected at ten depths upward from the wetting front via the sampling ports. Soil water content was determined by the oven-drying method. Soil NH4+–N and NO3-–N were extracted with KCl, whereas available phosphorus (AP) was extracted with NaHCO3 in the presence of activated carbon. Their concentrations were determined using a SmartChem 450 fully automated chemical analyzer. Available potassium (AK) was extracted with ammonium acetate and determined using a flame photometer; the electrical conductivity of the soil extract was measured with a DDS-307A conductivity meter.

At the maturity stage, three foxtail millet plants were randomly selected from each plot. Plant height was measured with a tape measure, and stem diameter with a vernier caliper. The panicles were detached, placed into self-sealing bags, and brought back to the laboratory, where panicle weight was measured using an electronic balance.

2.4 Simulation of water, salt, and nutrient distribution in soil profiles

The sand-mixing ratio, infiltration-water treatment, and soil depth were used as input features, while soil water content, salt content, NH4+–N, NO3-–N, available phosphorus (AP), and available potassium (AK) were used as target variables. To characterize the potential linear and nonlinear relationships between the input features and target variables and to compare the predictive performance of different types of models, multiple linear regression (MLR) was used as the linear baseline model, while random forest (RF), extreme gradient boosting (XGBoost), and light gradient boosting machine (LightGBM) were used as ensemble-learning models. Each of the four models was fitted separately to each target variable. MLR establishes predictive relationships through linear combinations of the input features and is readily interpretable. RF uses bootstrap samples to construct multiple decision trees and obtains the final prediction by averaging their outputs, thereby reducing model variance and improving prediction stability. XGBoost and LightGBM are both based on gradient-boosted decision trees and iteratively add decision trees to progressively minimize the objective function, thereby reducing prediction errors. XGBoost incorporates regularization terms into its objective function to control model complexity and uses first- and second-order gradient information together with an efficient split-finding algorithm to improve optimization efficiency. LightGBM employs histogram-based feature binning to reduce computational and memory requirements and uses a leaf-wise tree-growth strategy to enhance model-fitting capability (Trinco et al., 2025; Chen and Guestrin, 2016; Ma et al., 2025).

2.5 Data analysis

Data were collated using Microsoft Excel 2020, and figures were generated using Origin 2019. Subsequent statistical analysis using SPSS 20.0 was performed on these data to determine significant differences (p< 0.05) in soil water content, salt content, and nutrient content. Machine-learning analyses were performed in Python, primarily using the scikit-learn, XGBoost, and LightGBM packages. For each target variable, the modeling dataset was randomly divided into training and test sets at a ratio of 7:3, with 70% of the data used for model training and the remaining 30% used as the test set to evaluate predictive performance. A random seed of 42 was used for data partitioning and for the RF, XGBoost, and LightGBM models to ensure reproducibility. The models were trained using predefined hyperparameters. For RF, the number of decision trees was set to 100 and the maximum tree depth to 10. For both XGBoost and LightGBM, the number of boosting trees was set to 100, the maximum tree depth to 6, and the learning rate to 0.1. MLR was implemented using the default settings of scikit-learn, and all unspecified hyperparameters of the other models were set to their respective package defaults.

Three metrics were used to evaluate model performance: the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE). These metrics were calculated using Equations 1–3, respectively.

(1) R2=1−∑i=1n(yi−y^i)2∑i=1n(yi−y¯i)2

(2) RMSE=∑i=1n(yi−y^i)2n

(3) MAE=1n∑n=1n|yi−y^i|

Where: yi is the measured value of the i-th sample; ŷi is the predicted value of the i-th sample; y¯i is the arithmetic mean of the measured values; n is the total number of samples.

3 Results

3.1 Effects of sand-mixing ratios and infiltration-water treatments on soil water infiltration

The temporal variation of the wetting front under different sand-mixing ratios and infiltration-water treatments is shown in Figure 2. During infiltration, the wetting front advanced rapidly initially and then decelerated gradually, and this pattern was observed across all treatments. The higher the sand-mixing ratio, the shorter the time required to reach the same wetting front depth. Taking a wetting front depth of 50 cm as the benchmark, the S0 treatment took approximately 650 min under the deionized water treatment (F1) and 390 min under the high-concentration NPK treatment (F3), while the 20% sand-mixing treatment (S4) reduced the infiltration time by 26% and 44%, respectively. Consequently, sand-mixing effectively reduced the total infiltration time and increased the wetting front velocity. At the same sand-mixing ratio, the deionized water treatment (F1) required the longest time to reach the same wetting front depth. Compared with F1, NPK fertilizer treatments (F2, F3) significantly accelerated infiltration, and the infiltration rate further increased with rising fertilizer concentration within the NPK groups. Under non-amended conditions, the high-concentration NPK treatment (F3) shortened infiltration time by roughly 40% relative to the deionized water control (F1). At the 20% sand-mixing ratio, both the low and high NPK treatments took a similar amount of time (around 210 min) to reach a wetting front depth of 50 cm.

Effects of sand-mixing and infiltration-water treatments on the wetting front. **(a–c)** represent infiltration-water treatments F1, F2, and F3, respectively.
Figure 2. Effects of sand-mixing and infiltration-water treatments on the wetting front. (a–c) represent infiltration-water treatments F1, F2, and F3, respectively.

3.2 Effects of sand-mixing ratios and infiltration-water treatments on soil water content

The results of analysis of variance (ANOVA) for soil moisture content at different depths under various sand-mixing ratios and infiltration-water treatments are presented in Table 2. Sand-mixing ratio, nutrient treatment, and soil depth all exhibited significant effects on soil moisture content.

Table 2. ANOVA results for the effects of sand-mixing ratio, infiltration-water treatments, and soil depth on soil moisture content.

Source of varianceFreedomMean square errorF valueP value
Sand-mixing ratio (S)458.539121.063<0.001
Infiltration-water treatment (F)2191.460395.958<0.001
Depth (D)9229.403474.429<0.001
S × F82.3074.770<0.001
S × D3611.67224.139<0.001
F × D180.6801.4060.127
Error3000.484
R2 = 0.934

The variation curves of soil moisture content with depth under different treatments are shown in Figures 3a-c. Figure 3a illustrates the variation of moisture content with depth across soil layers under sand-mixing treatments, while Figure 3b presents the change in soil moisture content with depth under different nutrient levels. As shown in Figure 3a, the surface soil exhibited the highest moisture content. Within the 0–20 cm layer, moisture content decreased with increasing depth. The sand-mixing ratio reduced soil moisture in this layer, demonstrating a clear inverse relationship in which moisture content declined with higher sand-mixing ratios. This indicates that sand-mixing effectively lowered the soil water content in the 0–20 cm layer. At 20 cm depth, soil moisture followed the order of S0 > S1 ≈ S2 > S3 > S4. From 25 cm downward, all sand-mixing treatments consistently maintained higher moisture content than the S0 control. The moisture content in the S2 and S3 treatments was significantly higher than that in S0 (P< 0.05), whereas the values for S1 and S4 showed no statistically significant difference from the control. In summary, surface sand-mixing created a contrasting moisture profile: it reduced moisture within the mixed layer but enhanced it in the underlying soil. Sand-mixing ratios of 10%–15% proved most effective in improving moisture retention in the subsoil. As shown in Figure 3b, soil moisture content increased with higher nutrient levels across all soil layers, with the F3 concentration significantly elevating water content throughout the profile. While moisture content exhibited an overall declining trend with depth, a notable exception occurred at around 25 cm, where the presence of the sand layer resulted in a local moisture peak. Subsequently, moisture content resumed its decline, reaching its minimum observed value at 50 cm depth.

Soil moisture content across soil layers under different treatments. Different lowercase letters indicate significant differences among treatments (p< 0.05). **(a)** represents the sand-mixing treatments, and **(b)** represents the infiltration-water treatments.
Figure 3. Soil moisture content across soil layers under different treatments. Different lowercase letters indicate significant differences among treatments (p< 0.05). (a) represents the sand-mixing treatments, and (b) represents the infiltration-water treatments.

3.3 Effects of sand-mixing ratios and infiltration-water treatments on soil salt content

The variation of salt content with depth in different sand-mixing treatments under various infiltration-water treatments is shown in Figures 4a-c. As shown in Figure 4a, under the deionized water control treatment (F1), within the 0–40 cm layer, soil salt content was below the initial levels and increased with soil depth. All surface sand-mixing treatments resulted in lower salt content compared with the non-sand-mixed control (S0), demonstrating the efficacy of sand-mixing in mitigating soil salinity. Between 40 cm and 50 cm, salt content increased sharply. At 50 cm depth, soil salt content followed the order of S0 > S1 > S2 > S3 > S4, revealing a clear inverse relationship where higher sand-mixing led to lower soil salt content.

Soil salt content across soil layers under different treatments. **(a–c)** represent infiltration-water treatments F1, F2, and F3, respectively.
Figure 4. Soil salt content across soil layers under different treatments. (a–c) represent infiltration-water treatments F1, F2, and F3, respectively.

The average soil desalination rates (Tian et al., 2024) under different treatments are summarized in Table 3. Within the 0–40 cm layer, all sand-mixing treatments achieved higher average desalination rates than S0. Specifically, compared to S0, the desalination rates for S1, S2, S3, and S4 were higher by 10.31%, 14.39%, 37.90%, and 43.04%, respectively. However, due to the sharp increase in salt content within the 40–50 cm layer, the average desalination rate for the full 0–50 cm profile was lower than that of the 0–40 cm subsection.

Table 3. Average desalination rates under different sand-mixing treatments.

Soil layer (cm)S0S1S2S3S4
0–40 cm12.2522.5626.6450.1555.29
0–50 cm011.9916.6939.0644.62

As shown in Figure 4b, under the low-concentration NPK fertilizer solution treatment (F2), soil salt content was elevated in all layers relative to the initial level, with the sole exception being the S0F2 treatment at 20 cm depth, where it remained below the initial level. The S2 treatment exhibited the highest total salt content in the 0–15 cm layer. Across the 0–20 cm depth, total salt content generally decreased with soil depth, reaching a minimum around 20 cm. Further down (20–40 cm), salt content varied without a clear trend, while from 40 to 50 cm, it increased again with soil depth.

As shown in Figure 4c, under the high-concentration NPK fertilizer solution treatment (F3), salt content trends in the 0–15 cm layer differed among sand-mixing treatments. Below 15 cm, however, trends became more consistent across treatments. For the S0 treatment, salt content increased from 0 to 15 cm, decreased from 15 to 25 cm, and stabilized thereafter. For the S2, S3, and S4 treatments, salt content decreased from 0 to 10 cm, peaked within the 10–30 cm interval before declining, and then showed an overall increasing trend from 30 to 50 cm. At the 50 cm depth, both fertilizer treatments exhibited the same salt content order: S4 > S3 > S2 > S1 > S0, indicating that higher sand-mixing ratios consistently led to greater salt accumulation at deeper depths, regardless of fertilizer concentration.

3.4 Effects of sand-mixing ratio and infiltration-water treatments on nutrient dynamics

3.4.1 NH4+-N and NO3--N

The vertical distribution of NH4+–N content under different treatments is shown in Figure 5. Under the F1 treatment (Figure 5a), NH4+-N content remained relatively stable with depth. In the surface soil, the NH4+-N content decreased with increasing sand-mixing ratio. Between 20 and 25 cm depth, NH4+-N content increased in all sand-mixing treatments. Deeper in the profile, S0 consistently exhibited higher NH4+-N levels than the sand-mixing treatments. Under the F2 treatment (Figure 5b), soil NH4+-N content decreased with increasing depth in the 0–20 cm layer, and the highest content was recorded in the S0 treatment. NH4+-N content declined progressively with increasing sand-mixing ratio. From 25 to 50 cm depth, sand-mixing had minimal effect on NH4+-N content. Under the F3 treatment (Figure 5c), the overall trend of NH4+-N was similar to that observed under F2. However, at 25 cm depth, NH4+-N levels under all treatments were higher than those under the corresponding F2 treatments. This suggests that the elevated fertilizer concentration enhanced the vertical mobility of ammonium nitrogen, facilitating its leaching to greater depths.

Variation curves of NH<sub>4</sub><sup>+</sup>-N with soil depth under different treatments. **(a–c)** represent infiltration-water treatments F1, F2, and F3, respectively.
Figure 5. Variation curves of NH4+-N with soil depth under different treatments. (a–c) represent infiltration-water treatments F1, F2, and F3, respectively.

The vertical distribution of NO3-–N content under different treatments is shown in Figure 6. Under the deionized water treatment (Figure 6a), sand-mixing had little effect on NO3--N content within the 0–40 cm layer. From 40 to 50 cm, NO3--N content increased with depth, and this increase was more pronounced with higher sand-mixing ratios. Under the F2 treatment (Figure 6b), NO3--N content decreased with depth within the 0–20 cm layer, with the highest levels found in the S0 treatment. NO3--N content decreased with increasing sand-mixing ratio. Between 20 and 40 cm, the S3 treatment exhibited relatively high NO3--N levels. Overall, NO3--N content showed limited variation from 0 to 40 cm, but at 50 cm depth, the S4 treatment had the highest NO3--N content. Under the F3 treatment (Figure 6c), the influence of sand-mixing ratio on NO3--N was pronounced. Within the 0–35 cm depth, higher sand-mixing ratios resulted in lower NO3--N content. Between 40 and 50 cm, the S3 treatment showed relatively high NO3--N levels. These results indicate that NO3--N was leached to deeper soil layers with the water flow. Furthermore, higher fertilizer concentrations resulted in significantly greater NO3--N accumulation at each corresponding depth.

Variation curves of NO<sub>3</sub><sup>-</sup>-N with soil depth under different treatments. **(a–c)** represent infiltration-water treatments F1, F2, and F3, respectively.
Figure 6. Variation curves of NO3--N with soil depth under different treatments. (a–c) represent infiltration-water treatments F1, F2, and F3, respectively.

3.4.2 Available phosphorus

The vertical distribution of available phosphorus (AP) content under different treatments is shown in Figure 7. The AP content exhibited distinct vertical distribution characteristics across soil depths. Under the F1 treatment (Figure 7a), the AP content in the 0–20 cm layer was generally lower than the initial soil AP level, except for the S0 treatment at 15 cm depth. Below 20 cm, the AP content increased with depth, indicating AP accumulation in the deeper soil layers. Under the treatments shown in Figures 7b, c, the trends were generally consistent. Within the 0–15 cm depth, AP content initially increased and then decreased with increasing depth. From 15 to 30 cm, AP content decreased with depth. Below 30 cm, the AP content approached the initial soil AP level. In both Figure 7b (0–20 cm) and 7c (0–25 cm), AP content was inversely related to the sand-mixing ratio but positively correlated with the fertilizer solution concentration at a given depth. A comparison between Figures 7b, c reveals that increasing the fertilizer concentration significantly enhanced the vertical migration of phosphorus within the soil profile.

Variation curves of AP with soil depth under different treatments. **(a–c)** represent infiltration-water treatments F1, F2, and F3, respectively.
Figure 7. Variation curves of AP with soil depth under different treatments. (a–c) represent infiltration-water treatments F1, F2, and F3, respectively.

3.4.3 Available potassium

The vertical distribution of available potassium (AK) content under different treatments is shown in Figure 8. A distinct vertical distribution pattern was evident. Under the F1 treatment (Figure 8a), the 0–10 cm soil layer exhibited a progressive increase in AK content with depth. Below 30 cm, sand-mixing had minimal influence on AK levels across the profile. The trends in Figures 8b, c were consistent. Within the 0–15 cm layer, AK content declined gradually with depth. At 20 cm depth, AK content was markedly lower (113–284 mg kg-¹) under the F2 treatment compared to the F3 treatment (777–958 mg kg-¹). At 25 cm depth, AK under the F2 treatment slightly exceeded the initial soil level, while under the F3 treatment it ranged from 138 to 239 mg kg-¹. Below 30 cm, AK content for all treatments returned to levels near the initial soil value at corresponding depths. In the top 20 cm, AK content was reduced by sand-mixing, showing a clear inverse relationship with the sand-mixing ratio. In contrast, below 20 cm, fertilizer concentration became the dominant factor influencing AK dynamics, overriding the effect of sand-mixing.

Variation curves of AK with soil depth under different treatments. **(a–c)** represent infiltration-water treatments F1, F2, and F3, respectively.
Figure 8. Variation curves of AK with soil depth under different treatments. (a–c) represent infiltration-water treatments F1, F2, and F3, respectively.

3.5 Evaluation of machine-learning algorithms for predicting soil physicochemical properties

This study assessed the performance of four machine-learning models for predicting key soil properties (water content, salt content, NH4+-N, NO3--N, AP, and AK) under various treatments. The evaluation metrics are summarized in Table 4.

Table 4. Comparison of predictive performance of machine learning models.

TargetModelWater content(%)Salt content(g·kg-1)NH4+-N(mg·kg-1)N03---N(mg·kg-1)AP(mg·kg-1)AK(mg·kg-1)
R2Linear Regression0.600.760.660.950.640.64
Random Forest0.910.931.000.990.981.00
XGBoost0.940.941.000.990.991.00
LightGBM0.770.880.950.950.960.96
RMSELinear Regression1.591.24180.4832.0298.11229.45
Random Forest0.750.6821.5511.9822.4520.43
XGBoost0.640.6419.6713.4413.2626.47
LightGBM1.220.8765.9934.5732.0677.69
MAELinear Regression1.361.04162.2025.7888.39207.75
Random Forest0.630.4212.348.0210.4712.81
XGBoost0.530.4110.359.067.6214.38
LightGBM0.920.7347.5625.6421.5857.68

The Linear Regression model exhibited the poorest prediction performance, with most R2 values ranging from 0.60 to 0.76. It also yielded markedly higher RMSE and MAE than the other models, suggesting that the linear model cannot capture the complex nonlinear relationships between influencing factors and soil properties. Both Random Forest and XGBoost achieved satisfactory performance across all targets, with R2 values above 0.91 and relatively low RMSE and MAE, which demonstrates the prominent advantages of tree-based models in handling high-dimensional interactions and nonlinear features. Among them, XGBoost slightly outperformed Random Forest, with R2 values higher than 0.93 for soil water content, salt content, and all nutrient indicators. The performance of LightGBM fell between the linear model and the above two tree-based models, with R2 ranging from 0.76 to 0.96. Although it performed better than Linear Regression, there was still a noticeable gap compared with Random Forest and XGBoost. By comprehensively considering R2, RMSE and MAE, XGBoost delivered the optimal or near-optimal prediction accuracy for all six prediction targets. Accordingly, XGBoost was selected as the optimal model for subsequent quantitative analysis of factor effects in this study.

3.6 Effects of sand-mixing ratios and fertilization treatments on foxtail millet growth

The results presented in Table 5 indicate that both sand-mixing ratio and fertilization rate had significant effects on plant height and stem diameter of foxtail millet at maturity (p< 0.05), and highly significant effects on panicle weight (p< 0.01). No significant interaction effects were observed between sand-mixing ratio and fertilization rate on plant height, stem diameter, or panicle weight. The effects of the two factors on these indicators were independent and additive rather than synergistic.

Table 5. Variance analysis of foxtail millet growth indicators at the maturity stage.

Source of variationFreedomPlant heightStem diameterPanicle weight
Mean squarep-valueMean squarep-valueMean squarep-value
Sand468.9140.0010.4820.0165.279<0.001
Fertilization283.0060.0037.8600.00216.462<0.001
Sand × Fertilization810.0280.5440.4880.8551.0250.439
Error3011.3900.1340.698

Plant height, stem diameter, and panicle weight of foxtail millet at maturity under different sand-mixing ratios and fertilizer application rates are presented in Table 6. Plant height is an important indicator reflecting the growth status of foxtail millet plants, and its variation is jointly regulated by sand-mixing ratio and fertilization rate. Under the same fertilization level, plant height generally increased first and then decreased with increasing sand-mixing ratio. Plant height decreased notably at S4, while S1 and S2 maintained relatively high plant height values, indicating that moderate sand-mixing (S1, S2) favored plant height, while excessive sand-mixing (S4) led to reduced plant height. This inhibitory effect persisted even under high fertilization (F3) conditions. For most sand-mixing treatments, plant height tended to increase with increasing fertilization level, generally following the order F3 > F2 > F1, except for the S4 treatment. This indicates that fertilization could promote the vertical growth of foxtail millet. However, differences between most treatments did not reach statistical significance, indicating that the effect of sand-mixing ratio on plant height gradually diminished with increasing fertilization level. High fertilization conditions effectively alleviated growth differences caused by variations in soil texture, resulting in more stable plant height performance.

Table 6. Growth indicators of foxtail millet at the maturity stage.

TreatmentPlant height(cm)Stem diameter(mm)Panicle weight(g)
S0F1133.76 ± 2.15aAB7.07 ± 0.33aA20.97 ± 0.20cB
S1F1136.69 ± 2.54aA7.56 ± 0.17aA21.83 ± 0.02bB
S2F1136.46 ± 1.10bA7.56 ± 0.35bA23.59 ± 0.86aA
S3F1133.62 ± 5.60aAB7.22 ± 0.50aA22.12 ± 1.02aB
S4F1128.61 ± 3.34aB7.09 ± 0.53aA21.87 ± 0.98aB
S0F2134.35 ± 2.91aB7.49 ± 0.07aA22.11 ± 0.60bB
S1F2137.14 ± 3.60aAB7.65 ± 0.22aA22.00 ± 1.08bB
S2F2140.07 ± 2.81aba7.83 ± 0.52abA24.33 ± 1.22aA
S3F2138.69 ± 4.72aAB7.65 ± 0.02aA22.55 ± 0.56aB
S4F2134.41 ± 4.50aAB7.27 ± 0.31aA22.01 ± 0.86aB
S0F3139.14 ± 5.50aA7.56 ± 0.49aB23.68 ± 0.37aA
S1F3139.42 ± 0.88aA7.72 ± 0.33aAB24.25 ± 0.64aA
S2F3140.85 ± 1.37aA8.38 ± 0.18aA24.48 ± 0.65aA
S3F3139.95 ± 1.55aA7.77 ± 0.61aAB23.52 ± 1.58aA
S4F3133.23 ± 4.11aB7.72 ± 0.20aAB22.99 ± 0.39aA

Within the same sand-mixing ratio, different lowercase letters denote significant differences (p< 0.05) among fertilization levels. Within the same fertilization level, different uppercase letters denote significant differences (p< 0.05) among sand-mixing ratios.

Stem diameter is an important indicator reflecting the robustness and lodging resistance of foxtail millet plants. Under low fertilization (F1) conditions, there were no significant differences in stem diameter among different sand-mixing treatments. Under medium fertilization (F2) conditions, stem diameter slightly decreased at S4, but no significant differences were detected across sand-mixing treatments. Under high fertilization (F3) conditions, the S2 treatment obtained the maximum stem diameter and was significantly higher than S0, whereas stem diameter of S1, S3 and S4 showed no statistical difference compared with S0. The regulatory effect of sand-mixing on stem diameter was only prominent for the S2 treatment under high-fertilization conditions. Under the same sand-mixing treatment, fertilization rate only had a significant regulatory effect on stem diameter in the S2 treatment, with no significant effects on the other sand-mixing treatments. This indicates that the regulatory effect of fertilization rate on foxtail millet stem diameter is dependent on sand-mixing ratio, and the overall regulatory effect is relatively weak. Stem diameter growth may be influenced by a combination of factors, including its own growth rhythm and other soil environmental factors. The treatment with the highest stem diameter was S2F3 (8.38 ± 0.18 mm), which was significantly higher than that of S0F3, demonstrating that moderate sand-mixing and high fertilization together contributed to superior stem diameter.

Panicle weight is a core indicator for assessing the yield potential of foxtail millet and is the most sensitive to sand-mixing ratio and fertilization rate. Under the same fertilization level, the effect of sand-mixing ratio on panicle weight varied. Under low fertilization (F1) and medium fertilization (F2) treatments, panicle weight initially increased and then decreased with increasing sand-mixing ratio, with the moderate sand-mixing (S2) treatment achieving the highest panicle weight, which was significantly higher than that of other sand-mixing treatments. Both excessive sand-mixing (S3, S4) and no sand-mixing (S0) significantly reduced panicle weight. Under high fertilization (F3) treatment, there were no significant differences in panicle weight among different sand-mixing treatments, indicating that high fertilization attenuated the regulatory effect of sand-mixing ratio on panicle weight. Under the same sand-mixing ratio, the effect of fertilization level on panicle weight exhibited a gradient pattern. Under no sand-mixing (S0), panicle weight increased significantly with rising fertilization level. For low sand-mixing (S1), panicle weight under F3 was significantly higher than those under F1 and F2, while no obvious difference was found between F1 and F2. Under moderate sand-mixing (S2) conditions, panicle weight was relatively high, and there were no significant differences among fertilization levels, indicating limited potential for further increase in panicle weight through fertilization. Under high sand-mixing (S3, S4) conditions, panicle weight tended to increase with increasing fertilization rate, but there were no significant differences between medium fertilization (F2) and high fertilization (F3), suggesting that excessive fertilization did not result in further significant gains in panicle weight.

In summary, fertilization exerted an overall positive effect on plant height of foxtail millet, while its effect on stem diameter was mainly observed under the S2 sand-mixing condition. Both indicators generally increased with the rise of fertilization rate, and the differences among sand-mixing treatments gradually diminished as fertilization rate increased. The moderate sand-mixing (S2) treatment resulted in the best performance for both indicators. For panicle weight, high fertilization (F3) attenuated the effect of sand-mixing ratio, resulting in no significant differences across sand-mixing treatments. Panicle weight performed best under the coupling of S2 with F2 and F3. Under low fertilization conditions, panicle weight was sensitive to sand-mixing; moderate sand-mixing significantly increased panicle weight, whereas high sand-mixing limited panicle weight potential. Increasing fertilization alone could not fully offset this negative effect. The comprehensive performance of foxtail millet at the maturity stage indicates that the combination of moderate sand-mixing (S2) with medium to high fertilization levels (F2, F3) was the optimal treatment for promoting foxtail millet growth and panicle matter accumulation. The underlying reasons are that moderate sand-mixing improves soil particle size distribution, reduces soil bulk density, and enhances soil aeration, while appropriate fertilization further improves nutrient use efficiency.

4 Discussion

4.1 Soil water and salt transport under sand-mixing and infiltration-water treatments

This study demonstrates that sand-mixing in the top 20 cm of moderately saline-alkali soil enhances wetting front migration, with a higher sand-mixing ratio leading to a shorter time to reach the same depth. This effect is attributed to alterations in soil particle composition and its consequent impact on infiltration capacity (Sun et al., 2022; Zhou et al., 2020; Zhang et al., 2019). Specifically, as noted by Bergström and Shirmohammadi (1999), sand-mixing increases soil macroporosity, facilitating rapid water movement through larger pore channels. While Xia and Tian (2020) reported that wetting front velocity decreases with increasing clay content within a range of 4.5% to 12.0%, our findings reveal a different pattern. In this present experiment, clay contents in the sand-mixed treatments were calculated as 11.5%–14.1% based on the measured particle size distributions of the original soil and river sand, using a weighted average method. Within this range, the correlation between wetting front velocity and clay content weakened significantly. This indicates that the influence of clay content may be subject to a threshold; beyond this point, factors such as soil structure and sand content become the dominant factors governing water movement. Qin et al. (2019) reported that urea solutions (300–900 mg·L-1) reduced the infiltration rate in saline-alkali soils, whereas this study observed that higher fertilizer concentrations promoted infiltration in moderately saline-alkali soil. This discrepancy can be explained by the properties of the medium- and high-concentration solutions (F2, F3). As strong electrolytes, their high ionic strength compresses the electrical double layer (Hillel, 2008; Brady and Weil, 1960), thereby diminishing electrostatic repulsion and promoting flocculation and aggregation of fine particles. This process improves soil porosity, thereby stabilizing soil structure and enhancing water infiltration.

The composition of soil particles determines porosity, which governs soil water-holding capacity (Su et al., 2013; Williams et al., 1983). As reported by Li et al. (2009), water-holding capacity decreases with increasing sand content. This occurs because sand particles possess a low specific surface area and form point-to-point contacts, creating a pore network dominated by macropores. Consequently, capillary and adsorptive forces for water retention are diminished. Under F1 conditions, surface sand-mixing reduced the moisture content within the sand-mixed layer while increasing it in the underlying soil. This is because the underlying soil layer has a relatively higher clay content and larger specific surface area, giving it stronger water adsorption and retention capacity. After water rapidly infiltrates downward, it is not easily transported further into deeper layers and therefore accumulates in this layer, resulting in an increase in its water content, which is consistent with earlier reports (Sun et al., 2022; Zhang et al., 2019). A sand-mixing ratio of 10–15% proved particularly effective in enhancing overall soil water retention. Under fertilizer solution infiltration conditions, although the test soil contains a certain amount of sodium ions, NH4+ and K+ in the fertilizer solution have smaller hydrated radii and stronger abilities to compress the double layer and promote clay flocculation than Na+. Under the high concentration conditions of this experiment, the flocculation effect induced by NH4+ and K+ dominates, and its intensity exceeds the potential dispersive effect of sodium ions. This promotes the formation of soil aggregates, improves soil pore structure, accelerates water infiltration, and consequently increases the water content across all soil layers.

4.2 Soil water–salt–nutrient relationships and model predictions

Existing studies have demonstrated that sand-mixing reduces soil salinity (Chen et al., 2022; Qi et al., 2021; Chi and Wang, 2009). Column experiments confirmed that soil salt content decreased with increasing sand-mixing ratio. Notably, the difference in salinity between 15% and 20% sand-mixing was marginal. Therefore, considering cost-effectiveness and soil moisture retention, a 15% sand-mixing ratio is proposed as the optimal practice for remediating moderately saline-alkali soils in the Datong Basin. However, in this experiment, salts migrated downward with percolating water and accumulated in the 40–50 cm soil layer, indicating the formation of a subsurface salt reservoir. Although no upward salt movement was observed during the experiment, such subsoil salt storage represents a potential long-term risk under field conditions. In this soil-column system, the sand-mixed upper layers had higher infiltration capacity, while the underlying non-sand-mixed soil retained its original pore characteristics. This inter-layer difference in hydraulic properties may retard further salt leaching and trigger salt accumulation at this depth. Under water deficit or tillage disturbance, the stored salts could be remobilized and may degrade root-zone soil quality in subsequent growing seasons. This risk is particularly prominent in regions with shallow, highly mineralized groundwater tables, such as the Datong Basin, where the subsurface salt reservoir can be continuously supplemented via groundwater capillary rise. This pattern of “surface leaching and subsurface accumulation” benefits root-zone desalination in the short term. However, in the long term, if subsequent irrigation or rainfall is insufficient, the accumulated salt may rise back to the root zone via capillary rise, causing secondary salinization (Lv et al., 2017; Rengasamy, 2006). Due to experimental limitations, the transport and redistribution of salt at deeper depths (>50 cm) could not be observed. Consequently, in practical field applications, long-term monitoring of salt dynamics in the 40–60 cm soil layer should be strengthened, combined with management measures such as drip irrigation and plastic mulching, to prevent subsoil salt from moving upward and damaging the crop root system.

The correlations between soil moisture content and salt content, NO3--N, NH4+-N, AP, and AK under various treatments were analyzed. As presented in Table 7, the correlation between soil moisture and salt content was consistently positive across all sand-mixing ratios, displaying a trend that initially strengthened and then weakened. With the exception of a negative correlation between moisture content and NO3--N under the S4 treatment, moisture content was positively correlated with all other measured indicators. Soil salt content increased with rising fertilizer solution concentration, a finding consistent with previous studies (Alomar and Jena, 2024; Yan and Shi, 2014; Jiang et al., 2019). Under deionized water irrigation, moisture content and salt content were significantly negatively correlated. However, this relationship shifted to a significantly positive correlation as the fertilizer solution concentration increased. For different fertilizer solution treatments (Figure 9), the correlation coefficient between NO3--N and NH4+-N transitioned from negative to positive and back to negative with increasing fertilizer concentration. The underlying mechanisms are interpreted as follows: During deionized water infiltration, NH4+ undergoes exchange adsorption with other cations on soil colloids, which repels NO3-, resulting in a non-significant negative correlation under low-concentration nitrogen solutions. Under the F2 treatment, as the soil’s adsorption capacity for NH4+ gradually approached saturation, the exogenously applied NH4+ and NO3- ions were co-transported by water flow. Consequently, the positions of their transport fronts in the soil profile tended to converge, and their concentration variations exhibited a positive correlation. At the highest concentration (F3), soil hydrodynamic processes become the dominant factor governing nitrogen transport differentiation. As a non-adsorbable anion, the NO3- transport front advances in close synchrony with the wetting front (Zhou and Selim, 2001). In contrast, the high concentration of NH4+ displaces native adsorbed cations via cation exchange, generating a “desorption pulse peak” that lags behind the water flow (Van Genuchten and Wierenga, 1976). This differential transport behavior creates a strong negative correlation between their concentration profiles. Upon dissociation, K+ from KH2PO4 is predominantly adsorbed by soil colloids, whereas PO43- ions are susceptible to chemical fixation (e.g., precipitation or specific adsorption) under alkaline conditions. In short-duration, high-intensity ponding infiltration experiments, a competition arises between the hydrodynamic transport of these ions and adsorption. Increasing the fertilizer solution concentration sharply elevates the influx of externally applied P and K. This saturates the instantaneous fixation capacity at the soil solid-liquid interface, resulting in a fraction of the ions persisting in the soil solution as free or weakly bound species that are co-transported with the infiltrating water flow. Thus, as the concentration of the fertilizer solution increased, AP and AK exhibited a significantly positive correlation. The correlations of NO3- with both AP and AK reversed as the fertilizer concentration increased. The critical fertilizer concentration at which this reversal occurs requires further investigation.

Table 7. Correlations between soil moisture and solutes under different treatments.

TreatmentMetrics (%)Salt content(g·kg-1)NO3--N(mg·kg-1)NH4+-N(mg·kg-1)AP(mg·kg-1)AK(mg·kg-1)
S0Water content0.2960.1080.705**0.692**0.710**
S10.3030.1630.623**0.600**0.623**
S20.366**0.1690.614**0.604**0.605**
S30.1840.1170.3280.3020.326
S40.04-0.0590.3010.2960.324
F1Water content-0.581**-0.711**0.376**-0.337*-0.223
F20.081-0.361*0.591**0.597**0.630**
F30.380**-0.1980.565**0.575**0.587**

**Indicates significant correlation at the 0.01 level (two-tailed). *Indicates significant correlation at the 0.05 level (two-tailed).

Correlations between soil nutrients under different fertilizer solution concentrations. **(a–c)** represent infiltration-water treatments F1, F2, and F3, respectively.
Figure 9. Correlations between soil nutrients under different fertilizer solution concentrations. (a–c) represent infiltration-water treatments F1, F2, and F3, respectively.

Overall, the results indicate that there may be an appropriate range of sand-mixing ratios. Excessive sand mixing, such as the 20% treatment used in this study, may result in the coupled effects of enhanced nutrient transport to deeper soil layers and reduced soil water-holding capacity. Regarding nutrient transport, a high sand-mixing ratio may increase the proportion and connectivity of large pores, thereby facilitating the downward transport of water and solutes and reducing the contact time between the soil solution and colloidal surfaces. Consequently, weakly adsorbed and highly mobile ions, such as NO3-, may be more readily transported to deeper soil layers with infiltrating water, increasing the risk of their accumulation in the subsoil or subsequent leaching (Zeng et al., 2021; Glæsner et al., 2018). Excessive sand mixing may further reduce soil water-holding capacity, resulting in rapid downward water movement and insufficient water storage in the surface soil.

In the correlation analysis, soil water content was negatively correlated with NO3-–N under S4 (20% sand mixing). Moreover, the 15% and 20% sand-mixing treatments produced similar surface-soil desalination effects but differed in their water-retention performance. These results support the above mechanistic interpretation to some extent. Under the experimental conditions, the 15% sand-mixing treatment achieved a relatively favorable balance among promoting salt removal from the surface soil, maintaining soil water-holding capacity, and reducing the risk of nutrient transport to deeper soil layers. In contrast, the 20% sand-mixing treatment exhibited some evidence of reduced water-holding capacity and an increased risk of downward nutrient transport. Therefore, from the perspective of sustainable soil amelioration, it is not advisable to seek greater surface-soil desalination solely by further increasing the sand-mixing ratio.

To evaluate the accuracy of the XGBoost model in predicting soil physicochemical indicators under different treatments, scatter plots with fitted lines were generated to compare the predicted values against the measured values, as shown in Figure 10 (using the F2 concentration treatment as an example, with similar results observed for other treatments). In the plots, points represent the measured values, while lines indicate the predicted values. The results demonstrate that the predicted values under various treatments show good agreement with the measured values, indicating that the model can be effectively used to predict changes in soil water, salt, and nutrient content at different depths across treatments.

Prediction of soil water content, salt content, and nitrogen, phosphorus, and potassium indicators under F2 treatment using the XGBoost model.
Figure 10. Prediction of soil water content, salt content, and nitrogen, phosphorus, and potassium indicators under F2 treatment using the XGBoost model.

4.3 Foxtail millet growth responses to sand mixing and fertilization

The results of this experiment indicate that both sand-mixing ratio and fertilization rate had significant effects (p< 0.05) on plant height, stem diameter, and panicle weight of foxtail millet at the maturity stage, with a highly significant effect (p< 0.001) on panicle weight. The significant promoting effect of fertilization on plant height and panicle weight is consistent with findings from previous studies (Gao et al., 2021; Pavithran et al., 2022; Han et al., 2020). Appropriate fertilizer application supplies key nutrients such as nitrogen, phosphorus, and potassium required for foxtail millet growth and development, promotes the synthesis and transport of photosynthetic products, and thereby enhances vertical growth and panicle development. As fertilization rate increased, the differences in growth indicators among sand-mixing treatments gradually diminished. The underlying mechanism is that under high fertilization conditions, sufficient soil nutrient supply effectively compensates for variations in soil texture caused by different sand-mixing ratios, thereby alleviating the constraints of soil physical properties on foxtail millet growth. This finding is consistent with the conclusions of Du et al. (2022).In this experiment, the regulatory effect of fertilization on stem diameter was relatively modest, showing no significant promoting or inhibiting trend. This may be because stem diameter growth in foxtail millet is more dependent on its own growth rhythm and is more significantly influenced by soil physical properties such as aeration and bulk density (Lipiec et al., 1996). Under the 20% sand-mixing ratio, panicle weight in the F2 and F3 treatments was lower than in all other sand-mixing ratios at the same fertilization levels, indicating that a 20% sand-mixing ratio is excessive under medium-to-high fertilizer input conditions. Beyond this threshold, the positive effect of sand-mixing on improving soil permeability is offset by its negative effects, ultimately leading to a decline in crop growth indicators. In contrast, the 10% sand-mixing ratio exhibited the best overall performance across different fertilization levels, with particularly prominent advantages in plant height and panicle weight. Moderate sand-mixing combined with appropriate fertilizer application effectively optimizes the agronomic traits of foxtail millet, promotes robust plant growth, and significantly increases panicle weight, providing technical support for high-yield and high-quality production of foxtail millet.

5 Conclusions

This study used soil-column infiltration and field experiments to investigate the effects of sand mixing and infiltration-water treatments on water, salt, and nutrient transport in moderately saline–alkali soil in northern Shanxi, as well as the effects of sand mixing and fertilization on foxtail millet growth. The main conclusions are as follows:

  • Increasing the sand-mixing ratio in the surface soil and the nutrient concentration in the infiltration water accelerated wetting-front movement. A higher nutrient concentration in the infiltration water increased soil water content in the soil profile, while sand mixing promoted the downward movement of water and salt and altered the vertical distribution of soil nutrients. An appropriate sand-mixing ratio facilitated surface-soil desalination and water storage in deeper layers, whereas excessive sand mixing may reduce soil water-holding capacity and increase the risk of salt and nutrient transport to deeper soil layers.
  • Among the four models, XGBoost exhibited the best overall predictive performance for soil water, salt, and nutrient contents under different treatments and at different soil depths.
  • Fertilization generally promoted foxtail millet plant height and panicle weight, while sand mixing promoted crop growth at appropriate ratios but inhibited it at excessive ratios. The combination of 10% sand mixing and a fertilizer application rate of 1500 kg·hm-2 showed the best overall performance. Considering both soil improvement and crop growth, a sand-mixing ratio of 10%–15%, combined with appropriate water and fertilizer management, is recommended for moderately saline–alkali soil in northern Shanxi under the experimental conditions.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Haichao Li, Swedish University of Agricultural Sciences, Sweden

Reviewed by: Jiefei Mao, Chinese Academy of Sciences (CAS), China

Yongmei Xu, Xinjiang Academy of Agricultural Sciences, China

Footnotes

AP, Available phosphorus; AK, Available potassium.

Statements

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Funding

The author(s) declared that financial support was received for this work and/or its publication. This work was supported by Basic Research Program of Shanxi Province (NO.202403021222125) and the earmarked fund the Modern Agricultural Industry Technology System Project of Shanxi Province (Grant No. 2026CYJSTX04-21).

Data availability

The original contributions presented in the study are included in the article/supplementary material. Further inquiries can be directed to the corresponding author.

Author contributions

WW: Conceptualization, Funding acquisition, Writing – review & editing, Data curation, Software, Writing – original draft, Methodology. SQ: Funding acquisition, Resources, Writing – review & editing. SY: Writing – review & editing, Visualization, Software, Data curation. XD: Validation, Conceptualization, Writing – review & editing. ZL: Writing – review & editing, Validation, Methodology. QC: Project administration, Resources, Writing – review & editing, Supervision.

References

  1. Alomar R., Jena D. (2024). Potassium and brackish water irrigation influence electrical conductivity, potassium dynamic, and maize yield in coastal saline soil. Commun. Soil Sci. Plant Anal. 55, 723 – 734. doi: doi:10.1080/00103624.2023.2276270
  2. Ashraf M., Awan A. R., Anwar S., Khaliq B., Malik A., Ozturk M. (2021). “ Economic utilization of salt-affected wasteland for plant production,” in Handbook of Halophytes (Cham: Springer), 2149 – 2172. doi: doi:10.1007/978-3-030-57635-6_87
  3. Bergström L. F., Shirmohammadi A. (1999). Areal extent of preferential flow with profile depth in sand and clay monoliths. Soil Sediment. Contamination (formerly J. Soil Contamination) 8, 637 – 651. doi: doi:10.1080/10588339991339513
  4. Brady N. C., Weil R. R. (1960). The Nature and Properties of Soils (New York: Macmillan Company).
  5. Chai H. X., Guo X. L., Wu W., Zhang X., Gao P. H., Gao P. L. (2023). The effect of different amendments on saline soils and wheat growth indicators. J. Shandong Univ. Technol. (Natural Sci. Edition) 37, 39 – 45. doi: doi:10.13367/j.cnki.sdgc.2023.06.008
  6. Chen P., Sun J., Ma L., Chen Y., Xia J. (2022). Effects of shell sand content on soil physical properties and salt ions under simulated rainfall leaching. Geoderma 406, 115520. doi: doi:10.1016/j.geoderma.2021.115520
  7. Chen T. Q., Guestrin C. (2016). “ Xgboost: a scalable tree boosting system ”, in: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (San Francisco: Association for Computing Machinery), 785 – 794. doi: doi:10.1145/2939672.2939785
  8. Chi C. M., Wang Z. C. (2009). Effect of sand on saturated hydraulic conductivity and salts leaching of sodic soil. J. Soil Water Conserv. 23, 99 – 102. doi: doi:10.3321/j.issn:1009-2242.2009.01.021
  9. Du X. R., Bai Y. X., Wang G., Jiang Y. L., Tong W. J., Zhao Z. X.. (2022). Effects of sand mixing in tobacco fields on soil physicochemical properties and nutrient accumulation of flue-cured tobacco. Chin. Tobacco Sci. 43, 32 – 39. doi: doi:10.13496/j.issn.1007-5119.2022.04.005
  10. Eck D. V., Qin M., Hirmas D. R., Giménez D., Brunsell N. A. (2016). Relating quantitative soil structure metrics to saturated hydraulic conductivity. Vadose Zone J. 15, 1 – 11. doi: doi:10.2136/vzj2015.05.0083
  11. Eswar D., Karuppusamy R., Chellamuthu S. (2021). Drivers of soil salinity and their correlation with climate change. Curr. Opin. Environ. Sustainability 50, 310 – 318. doi: doi:10.1016/j.cosust.2020.10.015
  12. Fan Z. (2023). “ Planting the hope of harvest in saline-alkali land——Shanxi agricultural university's saline-alkali land improvement technology team boosts sustainable agricultural development,” in Shanxi Daily (Taiyuan, China), 011. doi: doi:10.28713/n.cnki.nshxr.2023.001891
  13. Feng J., Ren Y. M., Yang Z., Guo R. F., Zhang S., Ren G. B. (2020). Adaptability analysis of millet in northern Shanxi. Anhui Agric. Sci. Bull. 26, 59 – 60. doi: doi:10.16377/j.cnki.issn1007-7731.2020.05.019
  14. Gao S. L., Wang C. X., Yang X., Ji T. Y., Shang S. Q., Li S.. (2025). Adaptability of foxtail millet varieties based on photosynthetic performance and agronomic traits. Plants 14, 2502. doi: doi:10.3390/plants14162502
  15. Gao X. F., Yan B. S., Wu C. X., Wang G. L. (2021). Effects of long-term fertilization on soil quality and millet yield on slope farmland in loess hilly areas. Agric. Res. Arid Areas 39, 76 – 83. doi: doi:10.7606/j.issn.1000-7601.2021.05.10
  16. Glæsner N., Diamantopoulos E., Magid J., Kjaergaard C., Gerke H. H. (2018). Modeling solute mass exchange between pore regions in slurry-injected soil columns during intermittent irrigation. Vadose Zone J. 17, 1–10. doi: doi:10.2136/vzj2018.01.0006
  17. Guo W. C., Fan G. S. (2011). Investigation of salinity accumulating and its movement characters in original saline wasteland. Trans. Chin. Soc. Agric. Eng. (Transactions CSAE) 27, 84 – 88. doi: doi:10.3969/j.issn.1002-6819.2011.03.015
  18. Han F., He W., Zhang H., Zhuge Y. P., Lou Y. H., Sun M. J.. (2020). Effects of different fertilization patterns on soil reclamation and foxtail millet growth in saline-alkali soils. Chin. J. Soil Sci. 51, 860 – 865. doi: doi:10.19336/j.cnki.trtb.2020.04.14
  19. He J. Y., Tian J. C. (2015). Economic effects and coupling models of water and fertilizer of rice cultivated in aerobic soil. J. Drainage Irrigation Machinery Eng. 33, 716 – 723. doi: doi:10.3969/j.issn.1674-8530.15.0044
  20. Hillel D. (2008). Environmental Soil Physics (Pittsburgh: Academic Press).
  21. Huang L., Fan X. K. (2018). The effects of phosphate and potash fertilizers in different facilities on soil nutrient migration. J. Soil Water Conserv. 32, 184 – 190. doi: doi:10.13870/j.cnki.stbcxb.2018.02.027
  22. Jiang J., Zhai D., Zhang C. B. (2019). Effects of irrigation and fertilizer levels on the distribution of water and salt in saline field and maize yield. J. Appl. Ecol. 30, 1207 – 1217. doi: doi:10.13287/j.1001-9332.201904.039
  23. Li X. S., Dang H. K., Song N., Shen X. J., Gao Y., Sun J. S. (2020). Effects of biological organic fertilizer and Yellow River sediment mixture on water consumption and growth of winter wheat in saline-alkali land. Trans. Chin. Soc. For. Agric. Machinery 51, 272 – 284. doi: doi:10.6041/j.issn.1000-1298.2020.05.030
  24. Li Z., Feng H., Wu P. T., Zhao X. N., Guo Z. (2009). Simulated experiment on effects of soil clay particle content on soil water holding capacity. J. Soil Water Conserv. 23, 204 – 208. doi: doi:10.13870/j.cnki.stbcxb.2009.03.034
  25. Li L., Yang S., Hu X., Li Z., Chen H. (2024). The combined application of salt-alkali tolerant phosphate solubilizing microorganisms and phosphogypsum is an excellent measure for the future improvement of saline-alkali soils. Front. Microbiol. 15, 1364487. doi: doi:10.3389/fmicb.2024.1364487
  26. Lipiec J., Ishioka T., Szustak A., Pietrusiewicz J., Stepniewski W. (1996). Effects of soil compaction and transient oxygen deficiency on growth, water use and stomatal resistance of maize. Acta Agriculturae Scandinavica 46, 186 – 191. doi: doi:10.1080/09064719609413132
  27. Lu H. Q., Wang Z. H., Zhang J. H., Guo L., Yin F. H., Zhang J. Z. (2026). Process, experiences, and insights of saline-alkali land governance in Xinjiang Production and Construction Corps of China. Trans. Chin. Soc. Agric. Eng. (Transactions CSAE) 42, 152 – 163. doi: doi:10.11975/j.issn.1002-6819.202508084
  28. Lv Z. Z., Yang J. S., Liu G. M., Li J. B., Liu H. Q., Li B. (2017). Relationship between soil salinization and groundwater characteristics in the Yellow River delta. Acta Pedologica Sin. 54, 1377 – 1385. doi: doi:10.11766/trxb201701160401
  29. Ma D. X., Jiang Y. L., Wang X. P., Wang Z. P., Li Y. Y., Yao R. J.. (2025). Construction of an interpretable machine learning model for soil salinity inversion in the Hetao irrigation district of Inner Mongolia. J. Agric. Sci. 46, 20 – 29. doi: doi:10.13907/j.cnki.nykxyj.2025.04.003
  30. Pavithran A., Krishnan R., Somasundaram E., Chandrasekhar C. (2022). Evaluation of foliar nutrition for yield maximization in foxtail millet (Setaria italica). Int. J. Plant Soil Sci. 34 (21), 571–576. doi: doi:10.9734/ijpss/2022/v34i2131302
  31. Qadir M., Oster J. D. (2004). Crop and irrigation management strategies for saline-sodic soils and waters aimed at environmentally sustainable agriculture. Sci. Total Environ. 323, 1 – 19. doi: doi:10.1016/j.scitotenv.2003.10.012
  32. Qi B., Li J. J., Gao S., Meng S. H., Yang J. H., Lei T. W. (2021). Influence mechanism of sand mixing on soil clod disintegration and soil salt leaching of saline soil. J. Soil Water Conserv. 35, 369 – 375. doi: doi:10.13870/j.cnki.stbcxb.2021.06.049
  33. Qin X. Y., Wang C. X., He X. L. (2019). Effect of nitrogen application on infiltration and water and salt transport in saline alkali soil. J. Soil Water Conserv. 33, 248 – 252. doi: doi:10.13870/j.cnki.stbcxb.2019.01.039
  34. Rengasamy P. (2006). World salinization with emphasis on Australia. J. Exp. Bot. 57, 1017 – 1023. doi: doi:10.1093/jxb/erj108
  35. Song R. Q., Chu G. X., Zhang R. X., Bai L., Yang J. S. (2012). Effects of sand mulching on soil infiltration, evaporation and salt distribution. Acta Pedologica Sin. 49, 282 – 288. doi: doi:10.11766/trxb201010190436
  36. Su Y., Zhu J., Wang P., Wang D. B., Lin Y. (2013). Research progress on soil water holding capacity. Chin. Agric. Sci. Bull. 29, 140 – 145. doi: doi:10.11924/j.issn.1000-6850.2012-4022
  37. Sun C. T., Deng Y. P., Sun Y. X., Mao W. B., Mi Z. R., Zhang Q.. (2022). Effects of sand mixing on soil water-salt distribution and soil evaporation characteristics for salinized soil. J. Drainage Irrigation Machinery Eng. 40, 287 – 293. doi: doi:10.3969/j.issn.1674-8530.20.0109
  38. Tan J. L., Wang X. N., Tian J. C., Su X. L. (2018). Effect of gravel-sand mulching on movements of soil water and salts under different amounts of brackish water. Trans. Chin. Soc. Agric. Eng. 34, 100 – 108. doi: doi:10.11975/j.issn.1002-6819.2018.17.014
  39. Tian R. R., Wang S. J., Liu J., Xu L. Z., Sun Z. T., Li Y.. (2024). Applying biochar and flue gas desulfurization gypsum in the root zone to improve saline-alkali soil quality and sunflower yield. Trans. Chin. Soc. Agric. Eng. (Transactions CSAE) 40, 148 – 157. doi: doi:10.11975/j.issn.1002-6819.202310167
  40. Trinco F. D., Zeraatpisheh M., Turner H. C., El Mujtar V., Tittonell P. A., Galford G. L. (2025). High-resolution soil organic carbon mapping for enhancing predictive accuracy of environmental drivers in heterogeneous and mountainous landscapes in Patagonia. CATENA 259, 109353. doi: doi:10.1016/j.catena.2025.109353
  41. Van Genuchten M. T., Wierenga P. J. (1976). Mass transfer studies in sorbing porous media I. Analytical solutions. Soil Sci. Soc Am. J. 40, 473 – 480. doi: doi:10.2136/sssaj1976.03615995004000040011x
  42. Wang Q. J., Wang W. Y., Lv D., Wang Z. R., Zhang J. F. (2000). Water and salt transport features for salt effected soil through drip irrigation under film. Trans. Chin. Soc. Agric. Eng. 16, 54 – 57. doi: doi:10.3321/j.issn:1002-6819.2000.04.014
  43. Wang L. C., Zhu Z., Kong Q. Q., Peng Y. Q., Lu Y. X., Zhu L. Y. (2025). Characteristics of water infiltration and evaporation under different modes of mixing between soil and sand based on HYDRUS-1D. J. Agric. Sci. Technol. 27, 183 – 193. doi: doi:10.13304/j.nykjdb.2023.0496
  44. Williams J., Prebble R. E., Williams W. T., Hignett C. T., Williams J., Prebble R. E.. (1983). The influence of texture, structure and clay mineralogy on the soil moisture characteristic. Soil Res. 21, 15 – 19. doi: doi:10.1071/SR9830015
  45. Xia T., Tian J. C. (2020). Using clay content to predict water infiltration in sandy loam soils. J. Irrigation Drainage 39, 90 – 96. doi: doi:10.13522/j.cnki.ggps.2019340
  46. Xu G. W., Wang H. Z., Zhai Z. H., Sun M., Li Y. J. (2015). Effect of water and nitrogen coupling on root morphology and physiology, yield and nutrition utilization for rice. Trans. Chin. Soc. Agric. Eng. 31, 132 – 141. doi: doi:10.11975/j.issn.1002-6819.2015.10.018
  47. Xu C., Zhao H. X., Yan W. P., Li Q., Liu X. L., Wang J. P.. (2020). Study on yield formation and irrigation yield potential of maize under irrigation quota condition. Water Saving Irrigation 45, 22 – 26. doi: doi:10.3969/j.issn.1007-4929.2020.05.005
  48. Yan J. W., Shi H. B. (2014). Effect of nitrogen fertilizer on yield of wheat and soil moisture and salinity of different salinity soils. J. Irrigation Drainage 33, 50 – 53. doi: doi:10.13522/j.cnki.ggps.2014.04/05.011
  49. Yang Y., Zhang M. J., Wang S. J., Qu D. Y., Zhang Y., Wang J. X. (2023). Soil moisture variability affected by sand mulch: An isotope‐based assessment of irrigated farmland in Northwest China. Ecohydrology 16, e2477. doi: doi:10.1002/eco.2477
  50. Yue W. J., Zhang F. C., Li Z. J., Zou H. Y., Gao Y. (2015). Effects of water and nitrogen coupling on nitrogen uptake of muskmelon and nitrate accumulation in soil. Trans. Chin. Soc. For. Agric. Machinery 46, 88 – 96. doi: doi:10.6041/j.issn.1000-1298.2015.02.014
  51. Zeng H., Wen N., Zhang J. F., Zhang J., Hu K. L., Liu G. (2021). Effect of macropore preferential flow on nitrogen leaching in a North China Plain farmland. Chin. J. Eco-Agriculture 29, 66 – 75. doi: doi:10.13930/j.cnki.cjea.200508
  52. Zhang Y. H., Gao P. L., Zhang Q. W., Zhao Y. D., Yang D. M. (2019). Effect of mixing sand with surface soil on moderately salinized soil water-salt transport and summer maize growth. Soil Fertilizer Sci. China 56, 83 – 90. doi: doi:10.11838/sfsc.1673-6257.18189
  53. Zhang Z. Q., Yang L. L., He X. L., Li X. L. (2024). Effects of the regulation of water-salt-nitrogen on water-salt nutrient transport and yield in drip-irrigated cotton fields. Arid Zone Res. 41, 876 – 893. doi: doi:10.13866/j.azr.2024.05.15
  54. Zheng M. N., Liang X. Z., Han Z. S., Kang J. H., Chen Y. N. (2021). Effects of different improvement measures on the mitigating soil salinity and N uptake of silage corn in salt-alkali soil. Acta Agrestia Sin. 29, 2871 – 2877. doi: doi:10.11733/j.issn.1007-0435.2021.12.029
  55. Zhou L. Y., Li R. P., Miao Q. F., Dou X., Tian F., Yu D. D.. (2020). Effects of different sand ratios on infiltration and water-salt movement of heavy saline-alkali soil in Hetao irrigation area. Trans. Chin. Soc. Agric. Eng. 36, 116 – 123. doi: doi:10.11975/j.issn.1002-6819.2020.10.014
  56. Zhou L. Z., Selim H. M. (2001). Solute transport in layered soils. Soil Sci. Soc Am. J. 65, 1056 – 1064. doi: doi:10.2136/sssaj2001.6541056x

Citation

Wang, W., et al. (2026). Impact of sand-mixing and nutrient treatments on the growth of foxtail millet in moderately saline-alkali soil. Frontiers in Agronomy, Volume 8. https://doi.org/
Received
Accepted
Published
Volume
8 - 2026

Copyright

Copyright © 2026 Wang, Qiu, Yang, Duan, Liu and Cui.

License: https://creativecommons.org/licenses/by/4.0/

*Correspondence: Qingliang Cui

Article metrics

Views
0
Downloads
0
Citations
0
Shares
0

Related articles

Original ResearchIntegrating fungal virulence, formulation, and volatile-mediated attraction for an effective attract-and-kill strategy against Agriotes spp.Ismahen Lalaymia, Fanny Ruhland, Nolan Regnier, Elise Delahaut, Antoine Boullis, Clément Martin, François J. Verheggen, Stéphane DeclerckOriginal ResearchImproving energy efficiency and carbon balance through conservation and precision agriculture with respect to conventional agriculture: a methodological approachFrancisco Márquez-García, Miguel Á. Repullo-Ruibérriz de Torres, Gottlieb Basch, Emilio J. González-SánchezCorrectionCorrection: Identification of pathogens causing mulberry leaf spot and resistance evaluation of mulberry germplasms in Hechi, ChinaFrontiers Production Office