The study area, located in the Yili River Valley (42°14′16′′–44°53′30′′N, 80°09′42″–84°56′50″E), is west of Tianshan Mountain in Xinjiang, China, which is the wettest area in Xinjiang. The average annual temperature in the area is 10.4°C and the total annual precipitation is 417.6 mm (Dong et al., 2020).

Three habitats commonly invaded by A. artemisiifolia, namely woodland (43°28′49″N, 83°20′16″E), roadside (43°27′48” N, 83°28′29″ E), and farmland edge (43°32′17″N, 83°15′37″E, hereafter referred to as farmland), were selected for the study (Figure 1), harboring 20, 22, and 18 plant species, respectively (see Supplementary Table 1 for details of these species). Perennial herbaceous plants comprised 55, 41 and 50% of all plant species in woodland, roadside and farmland habitats, respectively, with the remainder consisting of annuals and biennials. There were differences in species richness and community coverage across the experimental plots, where the number of A. artemisiifolia plants at different growth stages decreased abruptly from the seedling to the vegetative growth stage (Table 1). The farmland habitat was irrigated multiple times during the growing season, and there was a crop rotation of wheat and corn throughout the year. The roadside habitat was defined as an approximately 6-m-wide space between the main road and an approximately 5-m-wide windbreak belt separating the roadside space from the adjacent farmlands. The woodland was along the Kunes River, where trees were planted approximately 30 years ago to form an embankment, after which, natural succession has occurred unchecked up until the time when this study was conducted.

Herbicide screening was conducted in woodland habitats invaded by A. artemisiifolia a long time ago, with few native plants remaining. Four randomized blocks were established at random, comprising 12 (5 × 6 m) plots each, for a total of 11 herbicide treatments and a control. Three fixed quadrats (1 × 1 m) were set up in each plot for index observation and data collection.

The 11 herbicides used in the experiment are listed in Table 2. Two non-selective herbicides, glyphosate and glufosinate, were included in the herbicide screening test to verify the control effectivity of non-selective herbicides on A. artemisiifolia, and whether A. artemisiifolia plants at the study site show resistance to glyphosate.

A herbicide screening test was conducted in late July 2018 and applied upon the appearance of male flower buds (BBCH 51, Bae et al., 2017). Eleven commercial herbicides were prepared according to manufacturer instructions to select the most effective herbicide. A hand-powered knapsack sprayer with a capacity of 16 l was used for herbicide application, with a nozzle size of 1.0 mm, a working pressure of approximately 0.35 MPa, and a flow rate of approximately 400–540 ml min⁻¹. Spraying was performed uniformly, with a spray volume of 450 l ha⁻¹, and a spray volume of approximately 1.35 l per plot. After 1 month, plant control rate was evaluated for each treatment group, and all A. artemisiifolia plants in the three fixed quadrats in each plot were clipped at the seed stage to calculate seed yield.

The effects of herbicide dose and time of application on the control of A. artemisiifolia were evaluated. Nine blocks were randomly set up in each of the three habitats (woodland, roadside, and farmland). Three blocks were selected at a time for herbicide spraying of A. artemisiifolia during each growth stage (seedling, vegetative growth, and reproductive growth). Each block comprised six (5 × 6 m) plots for a total of five dose treatments and a control treatment. Three fixed quadrats (1 × 1 m) were set up in each plot for index observation and data collection.

Following the herbicide screening test in 2018, aminopyralid was identified as the most effective herbicide (Table 3). To determine the best time and dose for application, aminopyralid tests were performed in 2019, for which, three blocks were randomly selected at the seedling (BBCH 14), vegetative growth (BBCH 32–35, in mid-June, when rapid growth of A. artemisiifolia was recorded, with an average plant height of approximately 20, 35, and 70 cm in woodland, roadside, and farmland, respectively), and the reproductive growth stages (BBCH 60–61, in late July). Aminopyralid dose treatments included X (110 g ai ha⁻¹), 0.66X (73 g ai ha⁻¹), 0.45X (49 g ai ha⁻¹), 0.30X (33 g ai ha⁻¹), and 0.20X (22 g ai ha⁻¹). The GR₅₀ (herbicide application rate required for 50% growth reduction) of aminopyralid for A. artemisiifolia was 33 g ai ha⁻¹, as per results from the previous experimental study and the current study (Supplementary Table 2). The time of each application was evaluated using three blocks in each habitat. Except for measurement of aboveground biomass, all other data, including plant density, mortality rate, seedling regeneration, and plant height of A. artemisiifolia were collected from three fixed quadrats in each plot. A. artemisiifolia plant density was measured before each application. At 30 d after each treatment, the mortality rate, seedling regeneration, plant height, and aboveground biomass of A. artemisiifolia were recorded, along with the aboveground biomass of the native species. After herbicide treatment during the vegetative growth stage, the number of A. artemisiifolia plants that did not bloom was counted (n = 30). All the A. artemisiifolia plants in the three fixed quadrats in each plot were harvested during the seed stage to calculate seed yield.

After the optimal application period was determined, in-depth analysis of the data was undertaken to determine the relationship between native species richness and A. artemisiifolia fitness, and whether effective control might be achieved with lower doses. During the vegetative growth stage of A. artemisiifolia, when other plants in the community were growing and could be identified, the species richness of each habitat (number of species in habitat), the species richness (number of species in each plot) and the number of individuals in each plot were calculated.

A linear mixed model analysis was used to examine whether under the same aminopyralid dose, there was an increase in the species richness of native plants, and an improved control effect had been achieved.

After the optimal application time and dose had been determined, the experiment continued in 2020 using the appropriate application time and dose. Aminopyralid treatments were applied in 2019 and 2020. Changes in the relative cover of A. artemisiifolia, the native plants and the bare patch were examined in each group of blocks annually before chemicals were applied. Data collection was also undertaken in 2021 to assess the effect of the herbicide on the control of A. artemisiifolia and on the restoration of native plants.

The plant community was divided into two parts: A. artemisiifolia and native plants. Because of the random distribution of native plant species, it is difficult to quantitatively observe the response of each plant species before and after herbicide treatment. The aboveground biomass was, therefore, used to represent the impact of herbicides on native plants because some plants were adversely affected but not killed by the herbicide treatment. The interannual variability of native plants under herbicide influence was expressed as relative coverage at the community level (represented by blocks in the paper). To identify the most effective herbicide and the most suitable control period, one-way ANOVA and Fisher’s least significant difference (LSD) multiple comparisons of means were used to analyze the effects of the 11 herbicides on A. artemisiifolia plants control and seed yield (Equation 1), compared to the unsprayed control. Similarly, the effects of herbicide treatment and the application period on A. artemisiifolia growth, mortality rate (plant height, aboveground biomass), reproduction (number of flowering plants and seed yield), and native plant species growth (aboveground biomass), were calculated.

where E stands control rate of A. artemisiifolia plants or seed yield reduction rate.

CK1 stands the number of A. artemisiifolia plants (or seeds) in the control group without herbicide application. Pt1 stands for the number of A. artemisiifolia plants (or seeds) in the herbicide application treatment.

To determine the effects of native species richness and number of individuals on A. artemisiifolia fitness, linear mixed model analysis was conducted with the lmerTest package in R 3.6.3 using data on optimal application time. A. artemisiifolia plant control and seed yield were used as dependent variables, while habitat and herbicide dose were the fixed effects, and species richness and the number of individuals in each plot and block were defined as random effects. To examine the impact of species richness on A. artemisiifolia seed yield, correlation analyses of “seed yield” and “species richness” data were conducted for each of the three habitats. The reasons are as follows: (1) Under the same chemical dose treatment, seed yield was found to be different with the assumption of a negative correlation. (2) Seed yield of A. artemisiifolia varied significantly among habitats; thus, the same plant species richness may correspond with different seed numbers. Therefore, some factors showed a significant correlation, while others showed no significant correlation (Table 4).

ANOVA followed by the LSD test were used to compare changes in the relative coverage of A. artemisiifolia, native species, and bare ground, and to evaluate community recovery after 2 years of weed control.

All data followed a normal distribution except for the relative cover data of A. artemisiifolia after control. For data that did not follow a normal distribution, the Kruskal-Wallis test was used.

Aminopyralid was found to be the most effective of the 11 herbicides evaluated (Table 3). At the recommended dose (110 g ai ha⁻¹), it resulted in over 90% control of A. artemisiifolia plants when applied upon the appearance of male flower buds (BBCH 51); furthermore, it caused a 98% reduction in A. artemisiifolia seed yield compared to the control. Glyphosate and glufosinate also reduced seed yield by more than 94%, implying that, A. artemisiifolia has not developed glyphosate resistance at this site. Nonetheless, as non-selective herbicides, they were deemed unsuitable for use in habitats with high levels of plant species richness.

The best time for chemical control of A. artemisiifolia was during the vegetative growth stage (BBCH 32–35) in mid-June. Although low-dose (0.2 and 0.3X) applications killed A. artemisiifolia plants at the seedling stage (BBCH 14, Figure 2), application during this period also caused the greatest harm to native species, reducing the aboveground biomass of native plants by more than 60% (Supplementary Table 2), while regeneration of A. artemisiifolia seedlings was also high (> 800 seedlings/m² on average). Herbicide application during the reproductive growth (BBCH 60–61) period was found to be ineffective at preventing seed production, even at the highest dose tested (Figure 3). Therefore, the optimum growth stage for herbicide application was the vegetative growth stage (BBCH 32–35), when herbicide application was found to inhibit vegetative growth and to reduce the number of plants entering the flowering stage (Figure 4). Thus, at low doses (0.2X and 0.3X), aminopyralid inhibited A. artemisiifolia seed production while having minimal effects on native plants species.

With the same dose of aminopyralid, the higher the species richness, the greater the reduction in A. artemisiifolia fitness (Table 4). Model M1 indicated that the same dose of aminopyralid did not significantly affect A. artemisiifolia control rate, regardless of habitat, whereas species richness of native plants showed a significant effect. The greater the native species richness, the higher the control rate on A. artemisiifolia. Results from Model M2 indicated that the seed yield of A. artemisiifolia was significantly affected by habitat type and herbicide dose. Further, correlation analysis indicated that with the same dose of aminopyralid, A. artemisiifolia seed yield was negatively correlated with species richness (Figure 5).

After 2 years of control, the relative coverage of A. artemisiifolia significantly decreased and was almost eradicated from the roadside habitat, whereas there was a significant increase in the relative coverage of native species (Figures 1, 6).

Invasive annual plants constitute a substantial proportion of the herbaceous invasive plant species found worldwide. In China, annual and biennial herbs together account for 46.1% of invasive plants. More than 60% of invasive plants such as A. artemisiifolia can cause severe negative effects on ecosystems (Zhou et al., 2020). The number of seeds and the germination characteristics play key roles in invasion by A. artemisiifolia. This species can produce hundreds to tens of thousands of seeds per plant, depending on the habitat and plant size. Seeds also have different dormancy characteristics (Bazzaz, 1974; Baskin and Baskin, 1980), with some seeds of invasive species germinating even after remaining dormant for up to 39 years (Baskin and Baskin, 1980). Propagule pressure has a substantial influence on the success of species invasion (Simberloff, 2009) and is a strong predictor of invasiveness (Colautti et al., 2006). Even in invaded habitats, propagule pressure can sustain invasive populations, and A. artemisiifolia can invade a wide variety of habitats after disturbance, thereby accumulating large amounts of biomass.

Our findings suggest that A. artemisiifolia seeds can germinate at different times (Figure 2), with germination having a clustering effect, meaning that the propagule pressure of A. artemisiifolia may be reflected in the number of seeds that can germinate at the same time. Tens of thousands of A. artemisiifolia seedlings can emerge simultaneously per square meter, occupying bare ground, before the numbers rapidly decrease to approximately 1,500–3,000 seedlings/m². Following the death of large numbers of A. artemisiifolia plants, additional seeds underground can then germinate rapidly to cover the area. After A. artemisiifolia seedlings have been killed by herbicide application, additional seeds in the seed bank germinated quickly, with the new seedlings filling the available space. However, the ability of the seed bank to take advantage of these openings decreased with time. After entering the vegetative growth stage, the number of viable seeds significantly decreased, and after entering the reproductive growth stage, almost no new seedlings appeared. Finally, after 2 years of herbicide application, the relative coverage of A. artemisiifolia significantly decreased in the study area (Figure 6).

These results indicate that abundant seed production and high seed germinability constitute a mechanism through which A. artemisiifolia can rapidly take advantage of ecological opportunities. Therefore, effective control can be achieved by understanding this germination strategy and depleting the seed bank by inhibiting seed setting through herbicide application at the optimal time.

The success of different weed control methods largely depends on the growth stage of the plant at the time of application. Some studies have compared the effectiveness of chemical control at different growth stages of A. artemisiifolia, but not much consideration was given to reducing the herbicide dose (Gauvrit and Chauvel, 2010; Bae et al., 2017). For example, Gauvrit and Chauvel (2010) tested herbicides in four stages of the A. artemisiifolia life cycle: early vegetative (BBCH 14), bud appearance (BBCH 53–55), pollen production onset (BBCH 61), and mid flowering of female flowers (BBCH 73–77). Using two agents, glufosinate and glyphosate, their study showed that the best time to apply herbicides was at BBCH 53–55. Bae et al. (2017) conducted experiments at stages BBCH 51 and BBCH 61–63 with a total of five agents. Their results showed that the application of tank mixtures containing 2,4-D or dicamba had the potential to limit seed production of glyphosate-resistant A. artemisiifolia when applied on or before stage BBCH 51.

While focusing on the chemical control of A. artemisiifolia, the weed cannot be considered as a separate element, as it is also under constant pressure by its surrounding environment. To adapt to complex natural environments, plants have evolved sophisticated mechanisms to balance growth and defense responses. The activation of defense mechanisms at the cost of suspending growth is known as the growth–defense tradeoff phenomenon (Coley et al., 1985; Figueroa-Macías et al., 2021). In the absence of pathogens or other environmental stimuli, young tissues must suppress immune or adaptation responses to maximize growth, whereas mature organs can allocate more resources to defense (Wang and Wang, 2014). In the present study, clear growth–defense tradeoffs were observed at different developmental stages in A. artemisiifolia. During the vegetative growth stage (BBCH 30–39), A. artemisiifolia exhibited rapid growth (Supplementary Table 2). According to this growth–defense tradeoff theory, A. artemisiifolia reduces its investment in defense and maximizes growth, resulting in a high level of sensitivity to chemical agents and greater susceptibility to low-dose herbicide treatment. Our findings showed that following treatment, the vertical growth of A. artemisiifolia substantially decreased, likely due to the triggering of defense mechanisms. In response to low-dose herbicide treatment, most A. artemisiifolia plants survived but to do so, growth and subsequent reproduction were sacrificed. Conversely, herbicide application during the reproductive growth stage did not have this effect, suggesting that the more mature organs had increased defensive capability.

Long-term effective management of A. artemisiifolia must rely on considerably reducing seed production to deplete the seed bank. Using this growth–defense tradeoff mechanism for A. artemisiifolia, effective control can be rapidly achieved, thus reducing the possible negative effects of the long-term use of low-dose chemicals that might entail the risk of herbicide resistance or plant hormesis (Busi et al., 2013; Jalal et al., 2021). The invasiveness of A. artemisiifolia can be reduced primarily by preventing seed production using low herbicide doses and relying on the resilience of native plants. After two years of herbicide treatment, A. artemisiifolia was almost entirely eradicated at three sites in the present study (Figure 6). Furthermore, this approach reduces the likelihood of the development of resistance in A. artemisiifolia and native plants due to the short-term use of low-dose herbicides. Thus, in the third year (2021), when no herbicide was applied, native plants effectively occupied the habitat (Figures 1, 6), demonstrating that this method is highly efficient and eco-friendly.

Reduction of A. artemisiifolia seed production, or even completely inhibiting it, means that the number of offspring that can develop resistance is reduced, and when combined with government increasing phytosanitary policies, the risk of spreading resistant offspring is greatly reduced. In the worst-case scenario, even if A. artemisiifolia becomes resistant to the herbicide, we can still increase the dose of the herbicide, as we would be using a low dose. This means that the duration of an effective agent can be extended, thus allowing the necessary time for the discovery or development of new agents.

In addition to directly reducing the fitness of A. artemisiifolia, herbicide application also mediated interspecific competition with native plants, which indirectly reduced the fitness of A. artemisiifolia (Table 4; Figure 5). According to the biotic resistance hypothesis (Elton, 1958; Levine and D'Antonio, 1999; Jeschke et al., 2012; Henriksson et al., 2016), plant communities with high species diversity are more resistant to invasion by invasive species than communities with low species diversity, with species diversity being negatively correlated with invasion. This hypothesis has been supported experimentally and theoretically by some studies (Dukes, 2002; Kennedy et al., 2002; Beaury et al., 2020). Mwangi et al. (2007) tested the hypothesis by transplanting species into experimental grassland communities, and showed that invasion resistance is related to the degree of niche overlap between resident species and invaders. This niche overlap can be high due to generally low amounts of empty niche space in species-rich resident communities. However, there are cases that do not support this hypothesis (Levine et al., 2004; Akin-Fajiye et al., 2021). Collins et al. (2007) found that species diversity does not appear to be an important factor for Imperata cylindrica invasion in the southern United States, and reasons for why no relationship was observed may be simply due to the tremendous competitive ability of I. cylindrica. At present, this is controversial.

Our study better fit here. On a small scale, high species richness reduced A. artemisiifolia fitness. The richness of native species was negatively correlated with A. artemisiifolia seed yield, regardless of herbicide treatment. It is likely that plant species in more diverse communities occupy a wider range of ecological niches and are more apt to compete with A. artemisiifolia (Byun et al., 2013, 2020). Diversified interspecific competition largely reduces the fitness of A. artemisiifolia, thereby providing a solid basis for effective weed control using herbicides at low doses prior to A. artemisiifolia outbreaks.