This study was performed at the Loess Plateau Research Station of Lanzhou University, Gansu, China (35°40′ N, 107°51′ E, elevation 1,297 m), a typical rainfed agricultural area with a semiarid continental monsoon climate and more than 60% of the precipitation occurring from July to September. The mean annual temperature and precipitation (1970–2021) are 9°C and 561 mm, respectively (Li et al., 2022). The soil is classified as silty loam soil based on the FAO/UNESCO soil classification (Table 1). The soil pH is 8.4 (0–20 cm).

Lucerne grassland (Medicago sativa L. cv. Gannong No.3) was established in April 2016 at a seeding rate of 1.5 kg ha⁻¹, sowing depth of 3 cm, and row spacing of 30 cm (Pan et al., 2016). The experimental field received 140 kg ha⁻¹ N and 70 kg ha⁻¹ P, as urea and calcium superphosphate at sowing stage, and no irrigation. Lucerne was harvested twice a year.

We conducted a preliminary study based on previous research to determine the types and combinations of herbicide (Davies and Peoples, 2003; Clark et al., 2021), from which three herbicides were selected: 41% glyphosate isopropylamine salt solution (Shandong Shengpeng Technology Co., Ltd., China, referred to herein as glyphosate), 90% 2,4-D isooctyl emulsifiable concentrate (Tianjin Boke Baisheng Technology Co., Ltd., China, 2,4-D), and 48% dicamba water agent (Zhejiang Baike Biotechnology Co., Ltd., China, dicamba).

The factorial experiment comprised two factors: (1) three herbicide type combinations [glyphosate and 2,4-D (G + 2), glyphosate and dicamba (G + D), 2,4-D and dicamba (2 + D)], with water (W) and conventional tillage (CT) used as controls; (2) three herbicide concentrations at 100, 200, and 300% of the recommended dose. The recommended doses for glyphosate, 2,4-D, and dicamba are 0.55 mL m⁻², 0.07 mL m⁻², and 0.09 mL m⁻², respectively. The experimental plots were arranged randomly with four replications of each treatment and 1 m walkways between each plot.

The treatments were applied on September 16, 2021, when the average height of lucerne was about 15 cm. The CT treatment was spaded to a depth of 40 cm to remove lucerne roots. Herbicide treatments involved spraying a tank mixture of the herbicide combination on the lucerne. Soil samples were collected in December 2021, January 2022, and March 2022 to assess dynamic changes in SOC and TN contents.

The normalized difference vegetation index (NDVI) was measured on September 20, September 30, and October 24, 2021, using GreenSeeker (RT100B, N-Tech, United States). The relative leaf chlorophyll content (characterized by SPAD) was measured with a portable SPAD chlorophyll meter (SPAD-502, Konica Minolta, Japan) on September 30, 2021. Five leaves in each plot were randomly selected to determine SPAD values, which were averaged.

The number of lucerne plants in a 1 m² quadrat was counted before the treatments were applied (September 2021) and after the regreening stage (April 2022) to calculate the regreening rate (Li, 2020):

Plant height was measured in May 2022 on four lucerne stems randomly selected in each plot, with the natural extension height from the root to the leaf tip measured with a ruler and the average value taken as plant height.

Soil samples at 0–10, 10–20, 20–30, 30–60, and 60–100 cm were collected with a soil drill (5 cm diameter). Part of the sample was air-dried and ground to pass through a 0.2 mm sieve to determine SOC and TN contents. Soil for ammonium-N (NH₄⁺-N) and nitrate-N (NO₃⁻-N) analysis was stored at 4°C.

SOC content was determined using the potassium dichromate external heating method (K₂Cr₂O₇-H₂SO₄ oxidation method) (Chan et al., 2001) and subsequently titrated with o-phenanthroline indicator and 0.2 mol L⁻¹ ferrous sulfate (FeSO₄) solution (Mccarty et al., 2010). TN content was determined using the Kjeldahl method (Kjeldahl, 1883). NO₃⁻-N and NH₄⁺-N contents were determined with an automatic high flux intermittent chemical analyzer (smart 450 auto analyst, AMS, Italy) with 50 mL of 2 mol L⁻¹ KCl extraction (Li et al., 2012). Soil pH was measured with an intelligent portable redox potentiometer (LD-QX6530, Laide-en, China).

NDVI, SPAD, and plant height at the regreening stage, regreening rate, and soil pH were analyzed by two-way ANOVA for herbicide type and concentration. SOC, TN, and NO₃⁻-N and NH₄⁺-N contents were analyzed by three-way ANOVA for herbicide type, herbicide concentration, and sampling date. The data were tested for normality and homogeneity before performing the ANOVA. Significant results were compared using Tukey’s HSD method with a significance level of 0.05. All data analyses were undertaken in the Agricola package (De Mendiburu, 2014) and plotted with the ‘ggplot2’ package (Wickham, 2016) in R (v4.1.3).

Herbicide type, herbicide concentration, and their interaction significantly affected the NDVI of lucerne grassland (Figure 1), which decreased over time (p < 0.05). The G + 2 treatment had a lower NDVI than the 2 + D treatment (p < 0.05). Regardless of herbicide type and concentration, the herbicide treatments had lower NDVI values than the W control (NDVI = 0.918 ± 0.010, p < 0.05).

Herbicide type, herbicide concentration, and their interaction significantly affected the SPAD value of lucerne (Figure 2). Except for the G + 2 (100%) treatment, the herbicide treatments had significantly lower SPAD values than the W control (49.91 ± 8.68). The G + 2 (200%) and G + 2 (300%) treatments had lower SPAD values than the G + D (200%) treatments (p < 0.05).

All herbicide types and concentrations significantly decreased the regreening rate of lucerne compared to the W control (regreening rate > 90%) (Table 2). There was no significant difference between herbicide types and concentration levels, and regreening rates under CT and herbicide treatments irrespective of types and concentrations were lower than 8%.

All herbicide types and concentrations significantly decreased lucerne height at the regreening stage compared to the W control (45.25 ± 1.13 cm) (Table 3). There was no significant difference between herbicide types and concentrations, and lucerne height under CT and herbicide treatments irrespective of types and concentrations were in the range of 10–25 cm, which was significantly lower than the W control.

Herbicide type, herbicide concentration, and their interaction significantly affected soil pH from 0–100 cm (Figure 3). The CT and W controls had significantly higher soil pH values than the herbicide treatments. The G + D treatment had the lowest soil pH (p < 0.05). The herbicide treatments at 100% concentration had significantly higher soil pH values than the 200 and 300% concentrations, with no significant difference between the 200 and 300% concentrations. The 2 + D treatment (100, 200 and 300%) had higher soil pH values than the G + D treatment (100 and 300%, p < 0.05).

Herbicide type, sampling date, and their interaction significantly affected SOC content (Figure 4). The 2 + D treatment had higher SOC content than the G + 2 treatment and W control (p < 0.05). The 2 + D treatment had the highest SOC content from 30–100 cm (p < 0.05). December had higher SOC contents from 0–10 cm than January and March, while March had higher SOC contents from 20–100 cm than January and December (p < 0.05). The 100% herbicide concentration had the highest SOC contents from 0–10 cm (10.65 ± 1.50 g kg⁻¹), even higher than the CT and W controls (p < 0.05).

Herbicide type, sampling date, and their interaction significantly affected soil TN content (Figure 5). The CT control and 2 + D and G + D treatments had higher TN contents than the W control (p < 0.05). Soil TN contents did not significantly differ in December and January but were higher than in March (p < 0.05). The herbicide types had similar soil TN contents in December, higher than the W control (p < 0.05). In January, the CT control had significantly higher TN content than the G + 2 treatment and W control, while the W control had significantly lower TN content than the G + D and 2 + D treatments. In March, no significant differences in TN content occurred between herbicide types.

Herbicide type, herbicide concentration, sampling date, and their interaction significantly affected soil NO₃⁻-N content (Figure 6). All herbicide treatments significantly increased soil NO₃⁻-N content compared to the W control but did not significantly differ from the CT control (p > 0.05). The 100% herbicide concentration increased soil NO₃⁻-N content more than the 200 and 300% concentrations (p < 0.05). January had higher soil NO₃⁻-N contents than December (p < 0.05). Regardless of herbicide type, herbicide application produced higher soil NO₃⁻-N contents than the W control in January and March (p < 0.05). The G + D (100%) and G + 2 (300%) treatments had higher soil NO₃⁻-N contents than the W control (p < 0.05).

Herbicide type, herbicide concentration, sampling date, and their interactions significantly affected soil NH₄⁺-N content (Figure 7). The G + 2 treatment had the highest soil NH₄⁺-N contents (7.94 ± 1.45 mg kg⁻¹) (Figure 7, p < 0.05), while CT had the lowest (6.46 ± 1.54 mg kg⁻¹). Regardless of herbicide concentration, the herbicide treatments had higher soil NH₄⁺-N contents than the CT and W controls (p < 0.05). December and March had lower soil NH₄⁺-N contents than January (p < 0.05).

The NDVI and SPAD values indicated that the three herbicide combinations had a higher efficiency of terminating and inhibiting lucerne growth in a short time. While it is difficult to successfully terminate lucerne growth with a single herbicide, selecting an appropriate herbicide combination can inactivate lucerne roots (Buhler and Mercurio, 1988). Studies have reported a synergistic relationship between glyphosate and auxin herbicides, which may induce lucerne withering (Wehtje and Walker, 1997; Flint and Barrett, 1998).

All herbicide treatments significantly inhibited lucerne’s regreening rate and plant height at the regreening stage, with the lowest NDVI and SPAD values in the G + 2 treatment (Figure 2), particularly at higher concentrations (200, 300%). The selected herbicides in this experiment are relatively safe. For instance, glyphosate has a half-life in soil and water that is relatively short (averaging about 30 d in temperate climates), and it does not bioaccumulate in organisms due to its high water solubility (Duke, 2020). Commercial 2,4-D is easily soluble in water and other solvents result in rapid penetration through the leaves/roots, and runoff of 2,4-D can be proper irrigation management (Islam et al., 2018). Dicamba is a chlorobenzene acid herbicide, and the half-life of dicamba in soil would be <12 weeks under typical conditions (Caux et al., 1993; Ochsner et al., 2006). Nonetheless, the recommended dosage, 100%, already significantly decreased regreening rate (Table 2) and height at the regreening stage (Table 3). Thus, in light of environmental and economic costs, applying the recommended dosage of glyphosate and 2,4-D in autumn can mostly terminate lucerne growth on the Loess Plateau. However, the general outcome of this study may vary depending on the climate and the specific aspect being studied. Further research is warranted to validate the relevant findings across multiple sites and years. It is worth noting that one-year trials have limitations in terms of their ability to fully capture the long-term effects of treatments, and the results should be interpreted with caution when making policy decisions.

Soil C and N contents in the herbicide treatments significantly differed from the W and CT controls (Figures 4–7). For instance, the 2 + D treatment had significantly higher SOC contents from 30–100 cm than the CT and W controls (Figure 4), possibly because the tillage destroyed soil aggregate structure, releasing and mineralizing SOC. A study on lucerne showed that the herbicide treatment had higher SOC content than the tillage treatment (Malhi et al., 2010). In our study, the herbicide treatments had lower soil pH values than the CT and W controls, possibly due to the degradation of lucerne roots (killed by the herbicides) by microorganisms, subsequently releasing carbon dioxide from organic acids and reducing soil pH (Hinsinger et al., 2003). This suggested that lucerne roots are gradually apoptotic, releasing C and N substances in the herbicide treatments. However, soil C and N contents did not significantly differ between the different herbicide treatments (Figure 8). This may be because soil organic and labile compounds can persist not for weeks but years, with the initial stages of litter decomposition unable to explain the persistence of soil organic compounds in the long term (Schmidt et al., 2011). Therefore, treatment effects were not evident for organic C inputs or outputs during the seven-month experiment, possibly because significant changes in SOC take time. Additionally, this study did not monitor gaseous soil C and N losses that significantly contribute to soil C and N dynamics. Thus, further study is warranted to assess gaseous and labile soil C and N dynamics in response to lucerne termination with herbicides.

The herbicide treatments and CT control increased soil TN and NO₃⁻-N contents in lucerne grassland relative to the W control, with significantly higher soil NO₃⁻-N content in the CT control than the herbicide treatments. This may be because CT adversely affects soil structure, disturbing aggregate formation (Six et al., 2002), increasing soil aeration, and releasing physically protected organic matter, thus promoting organic residue decomposition and mineralization (Jia, 2013). Nonetheless, lucerne termination with CT generally causes soil erosion, with large amounts of NO₃⁻-N runoff due to leaching and denitrification (Malhi and Nyborg, 1986; Nyborg et al., 1997). Herbicide-based lucerne termination can improve soil water conservation capacity, reduce water and soil erosion, and reduce carbon dioxide and nitrous oxide emissions relative to CT (Toderi et al., 2022). Although the termination effect of CT treatment was acceptable in this experiment, spading at 40 cm is a high energy consumption tillage operation, that is difficult to achieve in high clay soils.

March had significantly higher SOC contents from 20–100 cm under lucerne grassland than January and December. From December to March, the soils on the Loess Plateau experience severe freezing–thawing cycles (Li and Fan, 2014), which can increase SOC content by destroying soil aggregate structure and releasing physically protected SOC (Wang et al., 2012; Ma et al., 2021). Soil freezing also facilitates root death (Henry, 2007; Kreyling et al., 2012), increasing SOC content. The effect of soil frost on root mortality or function can also affect C dynamics because root respiration accounts for 40–60% of soil C losses (Hogberg et al., 2001).

Similarly, January had higher soil NO₃⁻-N and NH₄⁺-N contents than December, with soil NH₄⁺-N content decreasing in March relative to January, possibly due to the increasing temperature from January to March increasing nitrification (Hoyle et al., 2006; Wang et al., 2006), thus decreasing NH₄⁺-N content (Nicolardot et al., 2001; Wang et al., 2006). Besides, the decrease in soil pH could convert NH₄⁺ to NH₃, reducing soil N content. Changes in soil TN content depend on N accumulation and consumption. In this study, the change in soil TN content lagged behind that of mineral N, reflecting a relatively stable soil TN content, but the change in soil N supply capacity is difficult to predict over a short period (Kaye et al., 2002).