The study was conducted at the Huining Experimental Station (35°40′N, 105°06′E; 1800 m altitude) of Gansu Academy of Agricultural Sciences in Gaoling Village, Zhongchuan Town, Huining County, Gansu Province (Figure 1). The region is characterized by a temperate semi-arid continental monsoon climate. The site has a mean annual temperature of 6.8 °C and a frost-free period of approximately 130 days, with mean annual precipitation of 360 mm and evaporation of 1800 mm. The experimental soil was loessal soil with: 12.2 g kg⁻¹ organic matter, 0.83 g kg⁻¹ total N, 44.6 mg kg⁻¹ available N, 0.83 g kg⁻¹ total P, 3.79 mg kg⁻¹ available P, 20.6 g kg⁻¹ total K, and 178 mg kg⁻¹ available K. Average temperatures and precipitation for 2023 and 2024 are shown in Figure 2.

A split-plot field experiment was conducted under the complete plastic film mulching with double ridges and furrow sowing system. The main plots were assigned to three fertilization regimes: N1 (organic fertilizer only), N2 (inorganic fertilizer only), and N0 (no fertilizer). Subplots were allocated to three planting densities: D1 (200,000 plants hm⁻²), D2 (400,000 plants hm⁻²), and D3 (600,000 plants hm⁻²) (Table 1). The experiment comprised a total of nine treatments, each replicated three times in a randomized complete block design. The planting was performed by hand dibbling in holes spaced 20 cm apart within rows, with row spacing of 40 cm and a ridge width of 60 cm. Each subplot measured 5 m in length and 4 m in width, with a total area of 20 m². Sowing was performed in mid-May each year. For the fertilization treatments, nitrogen was applied basally before planting at a rate of 120 kg N hm⁻²: in N1, this was supplied by a commercial organic fertilizer containing 5.0% total nitrogen, 2.5% total phosphorus (P₂O₅), and 2.0% total potassium (K₂O); in N2, nitrogen was supplied by urea (46% N). The organic fertilizer was evenly broadcast and incorporated into the top 0–20 cm soil layer during field preparation, while urea was band-applied next to the sowing holes at a depth of 5 cm. No additional topdressing was applied during the growing season. Two widely cultivated foxtail millet varieties, ‘Nianbocaign’ and ‘Hongniangu’, were tested concurrently under all treatment combinations.

Data were analyzed using SPSS 17.0 (SPSS Inc., Chicago, IL, USA). Treatment differences were assessed by Duncan’s multiple range test at P<0.05. Data analysis revealed no significant inter-annual interaction effect on the experimental results. Structural equation modeling (SEM) was performed using AMOS 22.0 (IBM Corporation, Armonk, NY, USA) to examine causal relationships among variables.

Fertilization (N), density (D), and their interaction (N×D) significantly affected foxtail millet yield and water use efficiency (Figure 3). The water use efficiency (WUE) was calculated using Equation 6. For variety B, under different fertilization levels, the yield of N2 treatment increased by 3.05%−3.92% and 25.45%−26.43% compared to N1 and N0, respectively. Under different density levels, the yield of D2 treatment increased by 2.48%−3.09% and 3.87%−4.67% compared to D1 and D3, respectively. Under the interaction of fertilization and density (N×D), the yield of N2D2 treatment increased by 0.75%−2.47%, 2.29%−4.33%, 2.30%−4.60%, and 26.57%−38.62% compared to N2D1, N2D3, N1D2, and N0D2, respectively. Variety H also exhibited a similar pattern, with the yield of N2D2 treatment increasing by 1.04%−1.06%, 3.63%−3.66%, 3.61%−3.71%, and 24.20%−24.39% compared to N2D1, N2D3, N1D2, and N0D2, respectively. The N×D effect also significantly influenced millet WUE (Figure 3). For variety B, under N×D, the WUE of N2D2 treatment increased by 3.41%−5.21%, 7.09%−9.38%, and 34.69%−35.64% compared to N2D1, N1D2, and N0D2, respectively. For variety H, under N×D, the WUE of N2D2 treatment increased by 3.62%−3.81%, 6.77%−8.56%, and 32.15%−32.37% compared to N2D1, N1D2, and N0D2, respectively. In summary, the N2D2 treatment was beneficial for improving millet yield and WUE.

There were no significant differences in SWS between the 0–40 cm and 40–100 cm soil layers under different treatments at the sowing stage. The soil water storage (SWS, mm) was calculated using Equation 1. However, at the maturity stage, N, D, and N×D significantly affected SWS in both the 0–40 cm and 40–100 cm soil layers (Figure 4). For variety B, under different fertilization levels in the 0−40cm soil layer, the SWS of N2 treatment increased by 4.64%−4.83% and 7.77%−8.25% respectively compared to N1 and N0. At different density levels, the SWS of D2 treatment increased by 2.21%−2.98% compared to D1 and decreased by 0.71%−1.05% compared to D3. At N×D, the SWS treated with N2D2 increased by 10.32%−10.69% and 12.71%−13.57% respectively compared to N1D2 and N0D2. In the 40−100cm soil layer, for different fertilization levels, the SWS of N2 treatment increased by 4.92%−5.31% and 8.35%−8.74% respectively compared to N1 and N0. At different density levels, the SWS of D2 treatment increased by 2.52%−3.37% compared to D1 and decreased by 0.41%−0.73% compared to D3. At N×D, the SWS treated with N2D2 increased by 3.25%−4.08%, 6.92%−7.86%, and 9.73%−10.15% respectively compared to N2D1, N1D2, and N0D2. Variety H also maintains a similar pattern of change. Under the N×D soil layer of 0−40cm, the SWS treated with N2D2 increased by 7.86%−9.98% and 13.28%−15.10% respectively compared to N1D2 and N0D2. In the 40−100cm soil layer under N×D, the SWS of N2D2 treatment increased by 3.33%−3.79%, 4.55%−6.97%, and 10.27%−11.63% compared to N2D1, N1D2, and N0D2, respectively. Overall, N2D2 treatment can maintain a high soil water storage capacity.

N, D, and N×D significantly affected E and ET (Table 3). The total evapotranspiration (ET, mm) was calculated using the water balance equation (Equation 2). For variety B, at the fertilization level, the E and ET of N2 treatment decreased by 8.33%−8.54% and 2.83%−2.86%, 13.68%−13.71% and 4.05%−4.35%, respectively, compared to N1 and N0. At different density levels, the E and ET of D2 treatment decreased by 1.44%−2.51% and 1.41%−1.69%, 0.82%−0.95% and 0.16%−0.42%, respectively, compared to D1 and D3. At N×D, E and ET treated with N2D2 decreased by 11.46%−11.57% and 4.37%−4.48%, 16.28%−16.47% and 5.18%−6.02%, respectively, compared to N1D2 and N0D2. For variety H, under N×D, E and ET treated with N2D2 decreased by 11.40%−11.48% and 2.96%−4.47%, 16.45%−16.51% and 5.97%−6.03%, respectively, compared to N1D2 and N0D2. The varieties H, N, D, and N×D have a significant impact on T. The transpiration (T, mm) was derived as Equation 3. At the fertilization level, the T treated with N2 increased by 1.15%−1.85% and 2.75%−2.79% respectively compared to N1 and N0. At different density levels, the T of D2 treatment increased by 0.59%−1.63% and 0.33%−1.54% respectively compared to D1 and D3. At N×D, the T treated with N2D2 increased by 0.52%−3.15% and 1.88%−0.95% respectively compared to N1D2 and N0D2. For varieties B, N, and D, they have a significant impact on T, but N×D has no significant impact on T. In summary, N2D2 can reduce E and increase T to maintain a lower ET.

Figure 5 showed the water consumption characteristics of foxtail millet under different treatments. The results indicated that for the two varieties B and H, the E/ET ratio was lowest under the N2D2 treatment, reaching 37.79%–38.56%. Meanwhile, the T/ET ratio was highest under the N2D2 treatment, reaching 61.44%–62.03%. It can be seen that the N2D2 treatment can optimize the water consumption characteristics of foxtail millet.

Under different treatments, the LAI of foxtail millet showed a trend of first increasing and then decreasing as the growth stage progressed (Figure 6). The leaf area index (LAl) was calculated using Equations 4, 5 and N×D had a significant effect on the LAI. During the flowering stage, LAI reached its highest value. For variety B, the N2D2 treatment increased by 6.86%−6.99% and 13.87%−14.25% respectively compared to the N1D2 and N0D2 treatments. For variety H, N2D2 treatment increased by 5.61%−7.24% and 11.84%−13.92% respectively compared to N1D2 and N0D2 treatments. Subsequently, the LAI of each treatment decreased until maturity. For variety B, the N2D2 treatment increased by 6.70%−6.73% and 12.27%−12.30% respectively compared to the N1D2 and N0D2 treatments. For variety H, N2D2 treatment increased by 6.49%−6.53% and 12.28%−16.26% respectively compared to N1D2 and N0D2 treatments.

N, D, and N×D significantly affected the aboveground biomass and root biomass of foxtail millet (Figure 7). For variety B, under different fertilization levels, the aboveground biomass of foxtail millet treated with N2 increased by 2.18%−3.26% and 25.31%−26.19% respectively compared to N1 and N0. At different density levels, the aboveground biomass of foxtail millet treated with D2 increased by 0.09%−0.10% and 0.24%−0.32% respectively compared to D1 and D3. At N×D, the aboveground biomass of foxtail millet treated with N2D2 increased by 2.32%−4.43%, 2.24%−4.60%, and 26.62%−29.62%, respectively, compared to N2D3, N1D2, and N0D2. Variety H also maintained a similar pattern of change. Under N×D, the aboveground biomass of foxtail millet treated with N2D2 increased by 3.66%−4.65%, 3.71%−4.62%, and 24.39%−26.91%, respectively, compared to N2D3, N1D2, and N0D2. Similarly, for variety B, under N×D, the root biomass of foxtail millet treated with N2D2 increased by 1.91%–7.89% and 25.17%–33.28% compared to N1D2 and N0D2, respectively. For variety H, under N×D, the root biomass of foxtail millet treated with N2D2 increased by 3.63%–4.09% and 24.21%–28.97% compared to N1D2 and N0D2, respectively.

Linear functions were used to fit the relationships between each factor and WUE (Figure 8). The results indicated that for the two varieties B and H, WUE was significantly and negatively correlated with E under all treatments, with the regression explaining 82.1%–84.1% of the variation in WUE. Meanwhile, T, LAI and aboveground biomass were significantly and positively correlated with WUE, with the regressions explaining 38.2%–56.5%, 82.9%–84.2% and 96.8%–97.8% of the variation in WUE, respectively. It is evident that WUE is influenced by multiple factors.

SEM was used to analyze the comprehensive response of each factor to WUE (Figure 9). The results indicate that the interaction effect between fertilization and density mainly affects the RB (SPC = 0.83**) of foxtail millet by optimizing the SWS of the 0−40cm soil layer, which in turn affects the AB (SPC = 0.84**) and LAI (SPC = 0.81**) of foxtail millet. However, the improvement of WUE is mainly influenced by the positive effect of AB (SPC = 0.79**) and the negative effect of regulating T through LAI (SPC= -0.57*).

Crop yield is primarily determined by dry matter accumulation, with optimized planting density serving as a key strategy to enhance population-level biomass production (Zhang et al., 2020; Liu et al., 2023b). Our results demonstrate that the combination of inorganic fertilizer (N2) and moderate planting density (D2, 400,000 plants hm⁻²) achieved optimal growth and yield performance (Figure 3). This treatment significantly increased both crop biomass and leaf area index (LAI), consistent with findings by Wu et al. (2024a) in China’s Loess Plateau region. The mechanisms involve: (1) Inorganic nitrogen’s rapid availability promoting leaf expansion and photosynthetic efficiency (Wang et al., 2022); and (2) Moderate density balancing resource utilization (avoiding D1’s underuse and D3’s excessive competition) (Xu et al., 2024). More importantly, their interaction creates a synergistic effect: chemical nitrogen ensures sufficient nutrient supply for individual plant growth, while moderate density optimizes the population structure, allowing plants to more efficiently convert nutrients and moisture into biomass under limited resource conditions. The yield improvement stems from chemical nitrogen meeting crop nutrient demands while optimal density enhances photosynthetic performance. However, some long-term studies suggest organic fertilizers better improve soil structure and nutrient availability−a divergence potentially attributable to our shorter study duration highlighting inorganic nitrogen’s immediate effects, particularly critical in arid/semi-arid regions with limited precipitation windows (Manna et al., 2007).

Notably, the fully mulched double-ridge and furrow sowing system may have amplified these effects by improving soil moisture retention. Practical applications require consideration of local climate, soil texture, and agronomic practices.

Water is a critical factor influencing crop growth and yield formation. Both increased planting density and optimized nitrogen management must carefully consider water supply and utilization patterns, especially in dryland cropping systems where efficient water use is fundamental for achieving high and stable yields while improving nutrient utilization efficiency (Wang et al., 2019; Shen et al., 2024; Liu et al., 2025). Our results demonstrate that chemical nitrogen fertilization combined with moderate planting density effectively optimized water consumption patterns in foxtail millet, reducing total seasonal water use while significantly improving water use efficiency (Figure 3 and Table 3). These findings align with Tian et al. (2024), who reported that appropriate chemical nitrogen application promotes soil aggregate formation. The water-holding pores within these aggregates not only retain soil moisture but also preserve water-soluble nutrients, thereby achieving simultaneous improvements in both water and nutrient retention (Bai et al., 2021; Guo et al., 2022). Systematic water monitoring in our study revealed that the chemical nitrogen with moderate density treatment achieved comprehensive improvements in water regulation, manifested through three key increases and one decrease: enhanced soil water storage capacity, increased productive plant transpiration, improved water use efficiency, coupled with reduced unproductive soil evaporation losses. These observations are consistent with Tong et al. (2025), who found that optimized nitrogen application reduces non-productive soil water evaporation while improving crop water productivity.

The underlying mechanisms involve several aspects. First, moderate planting density creates an optimal canopy architecture that improves the utilization efficiency of light and thermal resources. Second, appropriately intensified plant competition under nitrogen-sufficient conditions enhances the crop’s capacity for water and nitrogen acquisition, thereby improving water use efficiency (Fang et al., 2021). Furthermore, chemical nitrogen application facilitates the retention of infiltrated water in upper soil layers, increasing the proportion of water allocated to productive transpiration (Xu et al., 2021). Notably, our study found that the combination of high planting density and organic fertilizer application disrupted the balance between individual plant performance and population productivity, ultimately reducing water use efficiency. This can be attributed to several interconnected factors. Under high-density conditions, the nutrient release from organic fertilizer is too slow to meet the peak nitrogen demand of the dense crop stand, leading to constrained dry matter accumulation during both vegetative and reproductive growth stages (Wu et al., 2024b; Wang et al., 2025). Simultaneously, high planting density intensifies competition for limited soil water, accelerating moisture depletion in the root zone. This water stress, coupled with nutrient limitation, reduces the photoprotective capacity of leaves, accelerates leaf senescence, and ultimately impairs grain filling (Guo et al., 2021). Moreover, the intensified competition under high density increases total seasonal water consumption without a corresponding yield benefit, resulting in a lower proportion of water being used for productive transpiration (Wei et al., 2019; Li et al., 2020).

Crop biomass and LAI are key indicators characterizing crop growth and photosynthetic capacity, both closely related to the formation of WUE (Bennetzen et al., 2012; Lin et al., 2024; Sharma et al., 2025; Wang et al., 2023). Results from correlation and SEM in this study confirmed that aboveground biomass (AB) is the core direct factor driving the improvement in WUE, while LAI indirectly influences WUE by regulating the water consumption structure of the population.

AB serves as the fundamental basis for enhancing WUE. In this study, a significant positive correlation was observed between WUE and AB, indicating that efficient biomass production is crucial for achieving high yield per unit of water consumed. The N2D2 treatment (chemical nitrogen fertilizer 120 kg hm⁻²+400, 000 plants hm⁻²) resulted in the highest aboveground biomass, which can be attributed primarily to the rapid nutrient supply from chemical nitrogen fertilizer, which sustained vigorous photosynthesis, and the optimal planting density, which improved population structure by avoiding light resource competition at high density and resource wastage at low density, thereby maximizing the accumulation of photosynthetic products (Guo et al., 2020). SEM analysis further clarified the direct positive path of aboveground biomass on WUE, indicating that under the optimized fertilizer-density regime, crops allocated limited water resources more effectively toward biomass production, establishing a virtuous cycle of “water promoting fertilizer efficiency and fertilizer regulating water use,” ultimately leading to a significant increase in WUE.

LAI indirectly enhances WUE by optimizing the water consumption components (T and ET). The findings of this study revealed a significant positive correlation between LAI and WUE; however, SEM indicated that LAI primarily influenced WUE through its effect on transpiration rather than through a direct path. This elucidates the physio-ecological mechanism by which LAI affects WUE: an appropriate LAI signifies a canopy structure with reasonable closure. Such a structure effectively covers the ground surface, significantly reduces ineffective soil evaporation, and promotes a shift in water consumption toward plant transpiration, thereby optimizing the water consumption structure and increasing the transpiration ratio (Zhang et al., 2016; Wang et al., 2024b). Consequently, the core role of a moderate increase in LAI lies in converting “ineffective” physical evaporation of soil moisture into “effective” physiological transpiration, creating favorable water use conditions for efficient dry matter accumulation.

From an applied perspective, the chemical nitrogen with moderate density regime (120 kg N hm⁻² + 400, 000 plants hm⁻²) developed in this study shows significant practical value for dryland foxtail millet production in the Loess Plateau region. However, special consideration should be given to adjustment of planting density to 350,000 plants hm⁻² during extremely dry years (growing season precipitation<250 mm) to maintain water balance. We acknowledge that the relatively short experimental duration (1–2 growing seasons) may limit comprehensive evaluation of fertilizer-density interaction stability across different climatic conditions, particularly regarding soil water dynamics and crop responses during extreme drought or wet years, which require further investigation. While this study demonstrates that the combination of chemical nitrogen fertilization (N2) and moderate planting density (D2) synergistically enhances yield and water use efficiency in foxtail millet under the fully mulched double-ridge system, it is important to acknowledge its limitations. The experimental design employed three discrete levels of nitrogen (N0, N1, N2) and three discrete density levels (D1, D2, D3). Although this approach successfully identified the optimal treatment combination among those tested, it does not allow for the precise quantification of continuous interaction effects between nitrogen application rate and planting density. Consequently, the exact agronomic optimum within a continuous gradient—such as the specific nitrogen rate that maximizes WUE at a given density, or the density that best balances yield and water use under a given nitrogen supply—remains to be determined. Future research should conduct multi-level gradient experiments, systematically changing nitrogen input and planting density simultaneously on a wider range, and combining the “water−fertilizer−density” synergistic regulation quantitative model for simulation and optimization. Such experiments would not only refine the functional relationships between these two key management factors but also help to identify context-specific recommendations under different soil moisture regimes and climatic conditions, particularly in the face of increasing climate variability in arid and semi-arid regions.