Rotation, Mulch and Zero Tillage Reduce Weeds in a Long-Term Conservation Agriculture Trial

Weed management is one of the main challenges of conservation agriculture. Although all three components of conservation agriculture (minimal tillage, permanent soil cover and crop diversification) can reduce weed populations, these effects may only become apparent in the medium to long term. This study evaluated weed biomass, density and diversity with and without herbicide control in a long-term trial initiated in 1991 in the Mexican Highlands to evaluate all three components of conservation agriculture. Data were collected in 2004, 2005, 2013, 2014 and 2015. Weed density and biomass were generally lower in conservation agriculture than with conventional tillage. The three components of conservation agriculture significantly reduced weed biomass, which was lower when all three components were applied together. When herbicides were applied, weed biomass in conservation agriculture was 91% lower in maize and 81% lower in wheat than in conventional tillage. Different treatments favored different weed species, but no trend toward increased perennial weeds was observed in conservation agriculture. These data supported claims stating that if adequate weed control is achieved in the initial years, weed populations in conservation agriculture systems are lower than in conventional tillage systems.


Introduction
Conservation agriculture (CA), a production system based on minimum tillage, permanent soil cover and crop diversification, increases yield and yield stability of maize and wheat under rainfed conditions [1,2]. Despite the potential benefits, uptake by smallholder farmers in the tropics and subtropics is limited. Competing interests using residue for fodder make it challenging for farmers to maintain adequate soil cover, and small farmers' lack of market access restricts the availability of profitable alternative crops for crop diversification [3,4]. However, when zero tillage (ZT) is applied without adequate soil cover and crop rotation, this can result in soil degradation, problems with weeds, pests and diseases and, consequently, lower yields [1,5,6]. Furthermore, farmers often lack access to knowledge regarding the management of CA-based production systems, which strongly differ from conventional practices [3]. Weeds are another major constraint for the widespread adoption of CA [7][8][9]. The main weed control methods in smallholder agriculture include tillage, hand weeding, mechanical control and cheap herbicides. In CA, tillage is not generally seen as an option for weed control. Instead, herbicides, mulching and cultural practices or integrated weed management are the primary means of weed management [10]. Weeds were reported to be problematic particularly in the initial years after changing from conventional agriculture to CA [10]. Therefore, the success of CA depends on optimum crop establishment and effective weed management. Improvement Center (CIMMYT) at El Batan, Texcoco, in the state of Mexico in Mexico. This experiment station was located in the semi-arid, subtropical highlands of Central Mexico (2240 masl; 19.318 N, 98.508 W), with a mean annual temperature of 14 • C and 616 mm average annual rainfall between 1995 and 2015, of which 567 mm fell on average during the growing season (May-October). The soil of the experimental plot is classified as a Haplic Phaeozem (clayic) in the World Reference Base system. The soil has 1.97% organic matter, 54.5 kg ha −1 phosphorus (P 2 O 5 ) and 210 kg ha −1 potassium (K 2 O).
The experimental design was a randomized complete block design with two replications. Individual plots measured 7.5 m by 22 m (165 m 2 ). The 32 treatments combined different wheat-maize rotations, tillage/planting methods and residue management practices. This study only included the first 16 treatments of the experiment, which consisted of the full factorial combination between the factors of 1) crop rotation, 2) tillage and 3) soil cover (Table 1). Crop rotation included continuous wheat, continuous maize (Mon) and wheat and maize rotations (Rot), with both phases of the double crop rotation included each year. Tillage was evaluated as CT or ZT with direct seeding. Conventional tillage consisted of one pass with a chisel plow to a depth of 30 cm depth, followed by two passes with a disk harrow to a depth of 20 cm and two passes with a spring tooth harrow to 10 cm. To prepare the seed bed before sowing, the tillage operations in December were repeated but including only one pass with the spring teeth harrow. No tillage operations were performed in ZT. The crop was planted by a custom-built, multicrop, multiuse, zero-till sowing machine. Soil cover was evaluated by keeping all residue on the field (K) or removing all residue for fodder (R) after grain harvest. Retained residues were incorporated in CT or left on the surface in ZT. The experimental design contained only two replicated plots, as the experiment was designed in 1991 to study the effect of CA practices on crop yields to provide agronomic recommendations for the Mexican highlands. Field experiments in agricultural research often use designs with two replicates to reduce costs while still allowing for testing of significant differences between treatments [26,27]; although the variation is larger when two replicates are used in a statistical analysis and "renders any test less powerful" [28], thus making it less likely to detect significant differences, it is still possible to determine the effect of an agricultural treatment in a two-replicate design (e.g., [29][30][31]).
Maize and wheat yields and yield stability of the trial were reported in [5,31,32]. Apart from the treatment factors, crop management for each crop was the same in all treatments. Maize was planted at 75,000 plants ha −1 in rows 75 cm apart, and wheat in at 110 kg seed ha −1 in rows 20 cm apart. The experimental crops were planted around the beginning of the rainy season, usually in the last week of May or the first week of June. In 2004 and 2005, 120 kg N ha −1 applied as urea was used as fertilization at the 4-5 leaf stage in maize and at the first node stage in wheat. In 2013, fertilization consisted of 150 kg N ha −1 as urea and was incorporated during preplanting in wheat and sowing in maize. In 2014 and 2015, fertilization was the same as in 2013, with the addition of 40 kg ha −1 P 2 O 5 as triple superphosphate, which was applied before sowing. The experiment was rainfed, although in some years when precipitation was too low to allow germination in all treatments, one sprinkler irrigation of 10-20 mm was applied to save the experiment. The crops were generally harvested during late September for wheat and late November for maize; wheat was harvested with a combine harvester and maize was harvested manually.

Weed Management in the Long-Term Experiment
Herbicides were used for weed control during the growing season; during the winter fallow season, the spring tooth harrow was used when needed for weed control (typically twice) in CT, while weeds were controlled with glyphosate in ZT when necessary. Herbicide application was similar on all plots with the same crop and tillage for a given year, regardless of the level of weed infestation. In zero-tillage treatments, glyphosate was applied before sowing during the fallow season, depending on the presence of weeds following rainfall events. Glyphosate was also applied after harvesting when weeds and volunteers proliferated, which occurred when the end of the growing season was wet. Once the experimental crop was established, weeds were controlled mainly by applying atrazine or fluroxypyr in maize and bromoxynil and halosulfuron or bromoxynil, fluroxypyr and clodinafop in wheat in all treatments ( Table 2). The amounts of active ingredients applied for each herbicide were as follows: glyphosate: 907.5 g ha −1 ; atrazine: 1850 g ha −1 ; bromoxynil: 460.5 g ha −1 ; halosulfuron: 75 g ha −1 ; clodinafop: 62.25 g ha −1 ; fluroxypyr: 249.75 g ha −1 or 182 g ha −1 , depending on the product used; topramezone: 33.6 g ha −1 ; fenoxaprop: 115 g ha −1 ; and S-metolachlor: 1450 g ha −1 . One sowing machine width, which was 1.6 m in wheat (8 rows at 20 cm) and 1.5 m in maize (2 rows at 75 cm) on the border of the plot, was left without weed control in 2005, 2013, 2014 and 2015 after the presowing glyphosate application. In these untreated strips, biomass samples and weed counts without herbicide treatment were taken. There were no long-term plots without herbicide application. The herbicides used in the trial were the herbicides locally available at that time.

Data Collection
Weed density and dry biomass content were recorded for each of the experimental units in the study. Weed data were collected after treatments were in place for more than ten years to allow the study of long-term effects and when resources or resident students were available for data collection. The type of data collected differed between years because the data were collected by several resident students as part of their research projects. Weed observations were recorded from anthesis to milk stage in wheat and from V10 to tasseling in maize (Table 3). In maize, weeds were counted and collected between 2 rows with a width of 0.75 m (0.75 m × 1 m = 0.75 m 2 ), and in wheat, weeds were counted and collected between 5 rows with a width of 0.20 m between 2 rows (4 × 0.20 m × 1 m = 0.80 m 2 ) using a quadrate. Aboveground weed biomass was harvested in the three sampling areas per plot, then pooled and sundried. Samples were dried in a hot air oven at 70 • C for at least 24 h, after which the dry biomass was determined. When density and biomass were determined in the same year, weeds were counted in the collection area while being harvested for biomass. Weeds were identified using the keys provided by National Commission for knowledge and use of Biodiversity, Mexico [33].

Data Analysis
Indices were calculated as follows: The Shannon-Weaver diversity index (H ) [34] was calculated using the following equation: where p i is the relative density of the i-th species. Species evenness was calculated using the following equation: Weed control efficiency was calculated as: Weed control efficiency= Weed biomass without herbicides − Weed biomass with herbicides Weed biomass without herbicides The complete dataset is available from Dataverse [35]. Statistical analysis of the effect of the experimental factors on weed biomass and weed density was performed using R 3.1.1 [36]. Generalized linear models and analysis of variance (ANOVA) were used to assess the effects of the components of CA on weed biomass, weed density, H' and J', using the "glm" and "aov" functions from the "stats" package. Treatment effects were analyzed per crop and with and without weed control using the following model: where R is the factor "Residue" (keeping all residues in the field or removing all residues), C is the factor "Crop rotation" (monocropping or wheat-maize rotation) and T is the factor "Tillage" (conventional tillage or zero tillage). The year of sampling and repetition were considered to be random effects. In the case of significant interactions between the factors, the datasets were split according to the respective factors and analyzed per factor level. The p level of significance for all effects mentioned in the text was 0.05. Post-hoc analysis was performed using the "HSD" (Honestly Significant Differences) function from the "agricolae" package, which makes multiple comparisons of treatments by means of Tukey.

Results
Over the years, 40 different weed species were observed in the trial (Table S1). Most species were annual dicots or broadleafs. Of all the identified species, 31 were annuals and 9 were perennials, with 30 dicots and 10 monocots. Only 10 weed species (Portulaca oleracea, Amaranthus hybridus, Sonchus oleraceus, Verbena bipinnatifida, Euphorbia stictospora, Galinsoga parviflora, Malva parviflora, Oxalis corniculata, Eragrostis mexicana and Bromus carinatus) were observed every year; of these, O. corniculata and E. mexicana were the most dominant. O. corniculata was mainly associated with maize, while E. mexicana was mainly associated with wheat.
Herbicides were generally effective in weed suppression. The subplots with herbicides showed significantly lower weed biomass (31 ± 5 g m −2 ) than the subplots without herbicide application (188 ± 16 g m −2 ). Lower weed biomass was recorded in maize treated with herbicides than in wheat treated with herbicides (17 ± 5 g m −2 and 34 ± 8 g m −2 , respectively, excluding observations from 2014, when data were only collected from wheat). There was a significant relationship between herbicide application and the tillage system on weed biomass, so data with and without herbicide were analyzed separately. Weed control efficiency was 93% in maize on average, but lower in wheat at 68%, mainly due to presence of E. mexicana; the minimum occurred in 2014, when it was only 33%. Tillage, residues and rotation did not influence weed control efficiency of herbicide applications. The treatment with the lowest weed control efficiency was W-Mon-ZT-K with an average of 52%, which demonstrated significantly lower weed control efficiency than all maize rotation treatments. Weed density was significantly higher without herbicide application than with herbicide application, resulting in 208 ± 29 plants and 134 ± 23 plants m −2 , respectively.

Effects of Conservation Agriculture and Its Components on Weed Biomass and Density
In maize with herbicide application, a significant effect of crop rotation was shown only on weed biomass, while the effects of tillage practice, residue management and interaction effects were not significant. Crop rotation in maize significantly reduced weed biomass by 87%, resulting in 31 ± 10 g m −2 in monoculture and 4 ± 1 g m −2 in rotation on average ( Figure 1, Table 4). ZT and CT in maize resulted in similar weed biomasses on average (18 ± 6 g m −2 in CT and 18 ± 9 g m −2 in ZT). Treatment effects were generally consistent over the measured seasons, although weed pressure magnitude differed widely throughout the years ( Figure S1). In herbicide-applied wheat, a significant effect of rotation was observed, with a weed biomass of 58 ± 13 g m −2 in the monoculture treatments and 24 ± 5 g m −2 in the treatments with rotation. Weed biomass was also significantly higher in wheat with CT than with ZT. There was no overall significant effect of residue in wheat, although weeds were significantly reduced in W-Mon-ZT-K compared with W-Mon-ZT-R (treatment abbreviations in Table 1). Weed biomass was higher in the treatments with wheat monocropping, except in W-Mon-ZT-K, mainly because of an infestation with E. mexicana which was not controlled by the clodinafop or fenoxaprop herbicides used.
In maize subplots without herbicide application ( Figure 1), a significant interaction was observed among all factors (tillage, rotation and residue), making interpretation of the overall effects complicated. Treatments with CT resulted in higher weed biomass than ZT treatments (220 ± 27 compared to 178 ± 43 g m −2 ), with treatments with rotation having lower weed biomass on average than monocultures (176 ± 29 compared to 222 ± 41 g m −2 ), but no apparent difference was observed due to residue retention (200 ± 35 and 198 ± 37 g m −2 with K and R, respectively). Effects of the treatments varied over the years; in 2005, the effects of all factors were significant, but this was not the case for 2013 and 2015 ( Figure S1).
In wheat without herbicide application, weed biomass was significantly higher in the CT treatments compared to the ZT treatments throughout all four years, but there were no significant effects due to rotation or tillage. Treatment effects varied over the years; in 2005 and 2014, there was only an effect due to crop rotation, but 2005 demonstrated higher weed dry biomass with crop rotation, probably due to a combination of initial drought and higher soil fertility with crop rotation, whereas lower weed biomass with crop rotation was observed in 2014. In 2013, all factors produced significant effects on weed biomass in wheat, and in 2015, rotation and ZT resulted in significantly reduced weed biomass ( Figure S1).  In herbicide-applied wheat, a significant effect of rotation was observed, with a weed biomass of 58 ± 13 g m −2 in the monoculture treatments and 24 ± 5 g m −2 in the treatments with rotation. Weed biomass was also significantly higher in wheat with CT than with ZT. There was no overall significant effect of residue in wheat, although weeds were significantly reduced in W-Mon-ZT-K compared with W-Mon-ZT-R (treatment abbreviations in Table 1). Weed biomass was higher in the treatments with wheat monocropping, except in W-Mon-ZT-K, mainly because of an infestation with E. mexicana which was not controlled by the clodinafop or fenoxaprop herbicides used.
In maize subplots without herbicide application (Figure 1), a significant interaction was observed among all factors (tillage, rotation and residue), making interpretation of the overall effects complicated. Treatments with CT resulted in higher weed biomass than ZT treatments (220 ± 27 compared to 178 ± 43 g m −2 ), with treatments with rotation having lower weed biomass on average than monocultures (176 ± 29 compared to 222 ± 41 g m −2 ), but no apparent difference was observed due to residue retention (200 ± 35 and 198 ± 37 g m −2 with K and R, respectively). Effects of the treatments varied over the years; in 2005, the effects of all factors were significant, but this was not the case for 2013 and 2015 ( Figure S1).
In wheat without herbicide application, weed biomass was significantly higher in the CT treatments compared to the ZT treatments throughout all four years, but there were no significant effects due to rotation or tillage. Treatment effects varied over the years; in 2005 and 2014, there was only an effect due to crop rotation, but 2005 demonstrated higher weed dry biomass with crop rotation, probably due to a combination of initial drought and higher soil fertility with crop rotation, whereas lower weed biomass with crop rotation was observed in 2014. In 2013, all factors produced significant effects on weed biomass in wheat, and in 2015, rotation and ZT resulted in significantly reduced weed biomass ( Figure S1).
In maize and wheat, weed biomass was markedly lower when crops were planted in CA (a combination of ZT with retained residue and crop rotation) compared with treatments representing the conventional production system of the Mexican highlands, i.e., CT with residue removal and monoculture of maize or wheat. When herbicide was applied, the weed biomass was 34 ± 18 g m −2 in M-Mon-CT-R, with M-Rot-ZT-K showing a biomass of 3 ± 1 g m −2 , 91% lower; similarly in wheat, weed biomass was 77 ± 33 g m −2 in W-Mon-CT-R, while W-Rot-ZT-K showed a result 81% lower at 15 ± 13 g m −2 . Differences were similar without herbicide application; weed biomass was 299 ± 79 g m −2 in M-Mon-CT-R, while in M-Rot-ZT-K this was 66% lower at 102 ± 29 g m −2 . Similarly, in wheat, weed biomass was 288 ± 37 g m −2 in W-Mon-CT-R, while this was 59% lower in W-Rot-ZT-K at 118 ± 75 g m −2 .
Regardless of herbicide application, no significant differences were observed in weed density between factors or between treatments throughout the whole time period (Table 5) due to the large variation in weed density between the years (Figure 2, Figure S2). Without herbicide application, significantly fewer weeds were produced in ZT compared to CT in 2013, but no significant effects of the factors were observed in the other years. With herbicide application, significantly fewer weeds were observed in maize than in wheat in 2015, with 63 ± 21 and 136 ± 49 plants m −2 , respectively. In maize, treatments with rotation showed similar weed densities, while maize monoculture produced a significantly lower weed density in CT and with residue retention. In wheat, there were significantly more weeds after monoculture treatments than with rotation. Treatments with monoculture resulted in significantly more weeds in ZT than in CT, while rotation treatments with tillage did not affect weed density. In 2013, significantly fewer weeds were observed in maize with crop rotation than with monoculture, 14 ± 4 and 122 ± 30 plants m −2 , respectively, while rotation did not affect weed density in wheat. In contrast, in 2004 and 2014, no significant effects on weed density were observed as a result of any of the factors. Agronomy 2020, 10, x FOR PEER REVIEW 10 of 16 variation in weed density between the years (Figure 2, Figure S2). Without herbicide application, significantly fewer weeds were produced in ZT compared to CT in 2013, but no significant effects of the factors were observed in the other years. With herbicide application, significantly fewer weeds were observed in maize than in wheat in 2015, with 63 ± 21 and 136 ± 49 plants m −2 , respectively. In maize, treatments with rotation showed similar weed densities, while maize monoculture produced a significantly lower weed density in CT and with residue retention. In wheat, there were significantly more weeds after monoculture treatments than with rotation. Treatments with monoculture resulted in significantly more weeds in ZT than in CT, while rotation treatments with tillage did not affect weed density. In 2013, significantly fewer weeds were observed in maize with crop rotation than with monoculture, 14 ± 4 and 122 ± 30 plants m −2 , respectively, while rotation did not affect weed density in wheat. In contrast, in 2004 and 2014, no significant effects on weed density were observed as a result of any of the factors.

Effects of Conservation Agriculture and Its Components on Weed Diversity
Overall, in ZT, perennial weeds were of minor importance in the weed community counts. Of the perennial weed species, Oxalis spp. was the most common, followed by Alternanthera caracasana and B. carinatus. With herbicide application, only the number of perennial weeds was significantly influenced by crop rotation and were shown to be significantly more abundant in treatments with maize monoculture compared to the other treatments. This was mainly due to the high incidence of O. corniculata in the maize monoculture treatments. Without herbicide application, there was a significant effect from the rotation × tillage interaction on the density of perennial weeds, with perennial weeds more abundant in maize monoculture with CT, while in the maize-wheat rotation there were no differences in perennial weed density.
In maize, an average of 45% of counted weeds were Oxalis spp., with 13% E. mexicana. In wheat, 30% of the weeds were E. mexicana and 24% were Oxalis spp. Other common weeds included Euphorbia stictospora, with 5% of observations in maize and 7% in wheat. In maize, Portulaca oleracea was another common weed, making up 5% of the observations, and in wheat Acalypha mexicana was common, with 7% of the observations. None of the other weeds observed formed more than 5% of the counted weed population in any of the years of observation. Because of the dominance of Oxalis spp. in maize and E. Mexicana in wheat, monocots were more dominant in the weed community in wheat, while dicots were more dominant in maize. Without herbicide application, significantly more monocots were observed in ZT, while there were no differences in monocot density due to tillage when herbicides were applied. Sedges were not commonly observed in the trial.
The weed community diversity per treatment was rather low in the trial, with the number of species observed per plot ranging between 0 and 12 in any given year. When herbicides were applied, the number of species was significantly impacted by management factors (Table 6). In wheat, the number of species was significantly higher in treatments with rotation compared to monoculture and with CT compared to ZT. In maize, the number of species was higher in ZT than in CT. There was also a significant rotation × residue interaction observed, with the number of species higher with residues than without in maize monoculture treatments, while rotation treatments demonstrated no effect of residue management. Without herbicides, species diversity and evenness only differed significantly with rotation, with the effect of residue management on species diversity appearing marginally significant (p = 0.06). There was higher species diversity and evenness in treatments with rotation compared to monoculture, and higher species diversity and evenness in treatments with residue retention compared to residue removal.

Discussion
This study demonstrated that, at least in the long term, weeds are not necessarily problematic in CA, and CA can lead to lower weed pressure compared to CT if adequate weed management is performed, as proposed in literature reviews regarding this subject [8,10]; this is true both with herbicide control and without. All three CA components showed potential to decrease weed populations. Even when the herbicides could not control weed populations adequately (e.g., the treatments with wheat monocropping showed that herbicides did not control all weed species), the combination of ZT and residue retention resulted in a significantly lower weed biomass, thereby demonstrating the usefulness of the individual CA components in weed suppression. The treatments with all three CA components generally resulted in the lowest weed biomass over the years, which may have been a result of the weed-reducing effect of all three components over the 25-year course of the experiment, leading to reduced weed seedbanks and lower weed populations in the treatments with adequate weed control through the CA components, while weed problems accumulated in treatments without adequate control, particularly in wheat.
The effect of residue retention on weeds was dependent on tillage; in CT, crop residues were incorporated by tillage, therefore, they did not form a mulch layer that suppressed weeds. Treatments with ZT resulted in low weed biomass, regardless of residue retention or removal. Weed germination, growth and development were lower in ZT, which may have been dependent on the residue management. In treatments with residue retention and ZT, soil fertility and soil water content were considerably higher compared to treatments with residue removal [5], leading to better crop growth and an improved ability of the crop to compete with weeds. The mulch cover, especially after maize, prevents weeds from germinating and may facilitate higher seed predation due to favorable conditions for soil fauna, such as ants and beetles [8,37]. In treatments with ZT but without residue retention, the soil was degraded and showed poor fertility and lower infiltration [5], which created an unfavorable environment for weeds to germinate and grow. Wheat was shown to leave more stubble when harvested than maize, and previous studies showed that ZT-treated soil with wheat monoculture or maize-wheat rotation without residue was less degraded than treatments with maize monoculture [38,39]. Therefore, in ZT treatments with residue removal, weed biomass was the lowest in M-Mon-ZT-R treatments, higher in M-Rot-ZT-R and W-Rot-ZT-R treatments, and highest in the W-Mon-ZT-R treatment, proving that weed management must be considered when designing CA systems, since the crops available for rotation and the residue that they produce affect weed control and growth.
The dominant weed species were different in maize and wheat, and their presence was related to management factors. In wheat, E. mexicana was the dominant weed, with mulching with residues helping to suppress this problem. In the W-Mon-ZT-R treatment, weed biomass was high due to the infestation with E. mexicana, while in W-Mon-ZT-K, it was significantly lower due to the presence of residues. This may have been due to the residues impeding germination, the small seeds having insufficient reserves to push through the residue cover, higher seed predation with residues present, or because of the residues' allelopathic effects. In wheat, the same herbicides (e.g., fenoxaprop or clodinafop) were used since the beginning of the experiment for weed control, but they did not control E. mexicana, leading to an increase in the presence of this weed. Alternative active ingredients available on the local market were tested in separate experiments, but none were able to control E. mexicana. In maize, E. mexicana was not a problem due to S-metolachlor effectively controlling the weed, combined with the higher competitiveness of maize with E. mexicana. Similar results were obtained in an experiment with winter wheat in Alberta, where populations of Bromus tectorum increased over the course of six years in wheat monoculture treatments, but not in a wheat-canola rotation, where it could be controlled with herbicides in canola [40]. In maize, Oxalis spp. weeds were the most common. As Oxalis spp. do not compete strongly with maize, especially later in the growing season, applying additional herbicides to control Oxalis spp. in maize was not considered to be necessary, leading to an increase in Oxalis spp. propagules in maize monocultures. In maize, treatments with monoculture had higher weed density on average due to presence of Oxalis spp., especially in CT. Crop rotation helped suppress the presence of this weed in maize; Oxalis spp. was not common in wheat.
Treatments with rotation resulted in larger weed diversity, which may have helped the crops avoid weed problems. The more diverse the weed community in a field, the less likely one dominant weed is to cause serious problems [41]. The diversity of weeds in a field depends on the local ecology and on station conditions. Changes appeared to occur in the weed species present in the trial since its initiation in 1992. While P. oleracea, A. hybridus, S. oleraceus, V. bipinnatifida, E. stictospora, G. parviflora, M. parviflora, O. corniculata, E. mexicana and B. carinatus were the most common weeds observed in this study, Elusine, Eragrostis, Setaria, Cyperus, Oxalis, Alcalypha, Galinsoga, Taraxacum, Portulaca and Sonchus were reported to be the most common weed genera in the 1992-1995 period [42]. Taraxacum was only observed in 8 out of 256 plots sampled over the years, while Setaria was not observed during our study. These shifts in populations were similar to those reported by [43,44], who reported rapid changes in dominant weed species during a five-year trial with tillage and herbicide treatments.
No indication of a higher incidence of perennial weeds in CA was found, which was similar to the findings of other studies; for example, a study on tillage and rotation in Argentina did not find a consistent relationship between tillage reduction and the presence of perennial weeds [20]. This is, of course, not a general conclusion, as it is highly dependent on the weed species present, the climate and the field management. The most common perennial weed species at the site, O. corniculata, was more abundant in CT, especially in maize. These observations were considered to be long-term effects of CA, as the treatment was in place for more than 25 years and occurred in a research station in which weeds were controlled for over 50 years. In Mexico, agroecological conditions vary widely across the country. Therefore, to gain more conclusive results, it is necessary to study the effects of CA in on-farm trials, different agroecologies and on different time scales.

Conclusions
Weed density and biomass were lower in long-term conservation agriculture than in conventional tillage, confirming the hypothesis described previously. The three components of conservation agriculture (zero tillage, residue retention and maize-wheat rotation) reduced weed biomass, which was lower when all three components were applied together. The exception was the treatment with maize monoculture, zero tillage and residue removal, where low weed densities were observed due to severe soil degradation. Different treatments favored different weed species, but no trend toward increased perennial weeds was observed in conservation agriculture. The data supported claims stating that if adequate weed control is achieved in the initial years, weed populations in conservation agriculture systems would be lower than in conventional tillage systems. Given the weed-controlling effects of conservation agriculture, this practice will likely lead to lower herbicide use in the long run.
Supplementary Materials: The following are available online at http://www.mdpi.com/2073-4395/10/7/962/s1. Table S1: Weed species observed and identified in the trial (some minor weed species could not be identified due to lack of flowers). Figure  Funding: This work was implemented by CIMMYT as part of the project "MasAgro Productor", made possible by the generous support of the Mexican Government through SADER. The project was part of the CGIAR Research Program on Maize (MAIZE), which was generously supported by W1 and W2 donors, including the Governments of Australia, Belgium, Canada, China, France, India, Japan, Korea, Mexico, Netherlands, New Zealand, Norway, Sweden, Switzerland, U.K., U.S., and the World Bank. Any opinions, findings, conclusions, or recommendations expressed in this publication are those of the authors and do not necessarily reflect the view of the donors mentioned previously.