The study area was located at 850 m a.s.l. Fifteen kilometers (15 km) to the west of Salamanca city, Spain (its coordinates: 40˚54'00''N, 5˚45'30''W), where a 1 ha experi- mental plot was chosen at the Muñovela experimental farm (C.S.I.C). The plot is eda- phically homogenous, with a dehesa-like woodland. Previously, it was heavily grazed although it is now fenced off to prevent the access of domestic animals.

The climate of the zone features rainy winters and hot summers and may be classified as semi-arid Mediterranean (C₁ B’₁ S₂ b’₄). Long-term mean rainfall and temperature have values of 500 mm and 12.3˚C, with November being the rainiest month (99 mm) and July the driest (17 mm). January is normally the coldest month (0.8˚C).

The tree covering comprises Quercus rotundifolia Lam, with a density of 98 trees ha⁻¹, a mean height of 5.9 m and mean diameter of 29.1 cm. Chorologically, the area lies in the Mediterranean Region, Carpetano-Ibérico-Leonesa province, Salmantino sector, Genisto hystricis-Querceto-rotundifoliae sigmetum series. The estimated mean age of the trees was 150 years. Other characteristic species defining the series are: Doryncium pentaphyllum, Thymus zygis, T. mastichina and Crataegus monogyna.

The soil is a chromic Luvisol [41] , developed over red clays and Miocene conglomerates. Soil texture A/B is loam/clay, the slope of the plot is 2%. pH H₂O: 6.3, % CO₃Ca: 0.07, % Organic matter: 2.14, P total (mg per 100 g): 298, N total (mg per 100 g): 731, Na (mEq per 100 g soil): 39.6, K (mEq per 100 g soil): 56.3, Mg (mEq per 100 g soil): 260, CaCl₂ (mEq per 100 g soil): 16.

A field experiment was carried out on abandoned arable land with sown low and high diversity treatments and natural colonisation following typical farming practice for the site. In April-May 1996, experimental plots were installed on former agricultural land that had been cropped with (a rotation of) monocultures until the end of 1995. The experiment was organised according to a block design with five replicate blocks. Within each block, four plots measuring 10 × 10 m were marked out and each of the three treatments was randomly assigned to one plot. All plots were separated by 2 m walkways. The three treatments (LD-low diversity seed mixture, HD-high diversity seed mixture and NC-natural colonization) were randomly allocated to the plots in each block.

As the initial vegetation development at abandoned land is usually highly unstable and unpredictable, late-successional types of functional groups of plants were experimentally sown in both low and high diversity mixture. Based on the specific characteristics of plants, the functional groups most widely recognised in tempered grassland communities and used in this study are (1) grasses, (2) legumes and (3) other forbs. The low and high diversity mixtures consisted of the same amounts of seed (grasses: 2500 seeds・m⁻², legumes: 500 seeds・m⁻², and other forbs also 500 seeds・m⁻²). Fifteen species (five per functional group) were sown as the high diversity sown treatment. For the low diversity sown treatment, low diversity seed mixtures (two grasses, one legume and one other forb species) were used as random choices from the total set of plants available for each replicate in order to take account of the sampling effects. The plant mixtures used consisted of species typical of later successional stages expected (Table 1).

The data were analysed by repeated measures ANOVA (P-values), for number of sown species, number of unsown, i.e. colonizing species, total number of plant species and proportion of the sown species calculated on the basis of cover. In the model, treatment (NC, LD, HD) was the fixed factor, block nested was the random factor while year was the repeated measures factor. Number of species applies for the pooled sample of 12 quadrants, 1 m² each, in the whole-plots and in subplots (averaged over the 12 plots). The ecologically most interesting tests are for the effect of treatment, and the interaction between treatment and year. Significant effects of treatment and treatment by year interactions; would show that there is a consistent effect of treatment, either constant over years (treatment), or variable over the years (treatment × year interaction). To test

the effect of sowing different numbers of species (same number of seeds of the three functional groups, but differing in the diversity within the functional groups), an ANOVA comparing LD and HD plots was also carried out. Since in LD plots, sown species composition differed among the replicates, the performance of the “best” LD plot (i.e. the plot showing maximum proportion of sown species, maximum suppression of natural colonizers, etc) with the HD treatment was also compared.

If there is an important species identity effect, then the performance of particular LD mixtures can be predicted from the species performances in the HD plots. The following procedure was used: for each of the sown species, the average “importance” (average value of cover abundance) over all the HD plots was calculated. For each LD mixture, the sum of importance of its constituent species was then calculated, and this value was used as a predictor of the performance of LD mixture. The species importance in HD plots should be proportional to its competitive ability, and consequently, the total for the particular LD species combination should be a good predictor of its ability to suppress the other species (note that it is important that this total is completely independent of the performance of species in LD plots).

The first comparison was by varying the number of treatments. One analysis was performed with natural colonization (NC), low diversity (LD) and high diversity (HD) to compare the effect of sowing with spontaneous development of vegetation. A second analysis concerning LD and HD only, to examine the effect of diversity was also performed. All treatment means were compared by Tukey’s HSD.

Various diversity indexes (Shannon [42] , Camargo [43] and Q statistics [44] and (Shannon and Berger-Parker [45] based on the proportional abundance of species were established (Table 2).

The heterogeneity of the vegetation might be an important factor for increasing the richness of vegetation at the scale of the whole-plot. Heterogeneity was characterized by taking the average of standardized Euclidean distance (chord distance) between all pairs of subplots within a whole-plot. Chord distance is classical ED calculated after standardization by sample norm, i.e. after each value is divided by the norm of the vector characterizing the community where x_i is representation of i-th species in the sample [46] .

Between both sowing treatments, no significant differences in the number of colonizing species were detected (F = 0.81, 1 df, P = 0.42). During all experimental process, a similar behavior was observed (Figure 1), such that no significant differences were found between the treatment × year interaction (F = 6.46, 6 df, P = 0.87). The number of colonizing species in the treatments of sowing LD-HD species mixtures for each of the blocks (B) sampled for seven years was recorded (Figure 2). To eliminate the blocks effect, an analysis (PCA) with the composition of colonizing species was established (Figure 3), which highlights the heterogeneity shown by the communities, such that

the correlation between species and treatments was very high (Axis 1 = 0.97, Axis 2 = 0.85).

The relationship between suppression of colonizing species, expressed as the ratio of the average number per year of colonizing species in HD of about NC (HD/NC) species, and the dominance index D1 found for the NC, based on the dominance of the most abundant species in each block is shown in Figure 4.

The annual results expressed as block showed that blocks of the same year presented a totally different behaviour even in the NC treatment (Figure 5). These results were

corroborated relating the index D1, based on the behavior of the most abundant species, with the number of dominant species by NC block calculated according to Camargo (1992) dominance index (Figure 6).

A significant and negative relationship (r = −0.77, P< 0.001) between the number of colonizing species in LD and performance of sowing species in HD was obtained (Figure 7), so the competitive ability of the sown species in different mixtures in LD, valued on the basis of their performance in HD, is a good predictor of their ability to suppress the weed crops species. This was also corroborated by the high relationship between the

efficiency of the sown species in both treatments of species mixture (R² = 0.51, F (1, 33) = 34.56, P < 0.001).

PCA explains 64% of the variability of sown species cover (eigen values: Axis 1 = 0.36 and Axis 2 = 0.28), in which a complete separation between the blocks of both treatments was not observed. In all experimental time, only the confluence of blocks 2 and 3 was presented. Instead, the blocks 1, 4 and 5 are located on opposite sides of the main axis 1, because the cover of certain species was more important in HD and little or none in LD (Figure 8).

Similarly, there were no significant differences between LD and HD treatments with respect to mean and minimum biomass (F = 2.38; 1 d.f ; P = 0.20), but occasionally, there were significant differences for same years (F = 28.14; 1 d.f ; P < 0.000) and also in the treatment× year interaction (F = 3.55; 1 d.f; P = 0.01). In 1998, the biomass in-

creased significantly in the two sown treatments and remained constant in the others (Figure 9). The grasses are the group that greatly contributes to the biomass increase in 1998; they also presented consistent differences in the treatment × year interaction (F = 1.81; 1 d.f.; P = 0.04), whereas it was nearly significant for the other forbs group (F = 2.39; 1 d.f.; P = 0.06) and null for the legumes group (F = 0.23; 1 d.f.; P = 0.96).

The blocks presented behaviours that are more homogeneous among themselves in HD sown species, whereas there were consistent differences between blocks in LD treatments, such that only the blocks 2 and 3 were similar in eliminating colonizer plant species in HD. However, 1998 was the only year where the large total biomass attained in both treatments was correlated with the high suppression of colonizer plant species by the combined effect of sown species and dominant species in natural conditions (Figure 10).

Diversity profiles of the sown species in LD and HD treatments during the experimental period, for each block and all experimental sampling period average cover degree is plotted against its rank within the sown species (Figure 11).

Relationship between the cover of sown species in LD and HD treatments for the experimental sampling period is shown in Figure 12. B. inermis was perhaps, the only species which attained high values of cover in both treatments, and therefore displayed competitively other sown species, but only those of the same functional group.

The individual behaviour of sown species was estimated on the basis of their performance in HD (Table 3). At the end of the experimental period, the mixed sown species in LD which had the best performance coincided with the best performance of the same species in HD. The more successful species during all experimental period was B. inermes, followed by Plantago lanceolada.

The individual performance comparison of sown species according to their functional group differs considerably, for example, many grasses disappear quickly without becoming dominant (Figure 13).

Mostly, in all the years of the study, there was a repetitive model in response to treatments of total biomass (Figure 14) and the number of colonizer species (Figure 15), because the treatment × year interaction for these two variables was not significant.

Comparing this interaction and taking into account only the sowing of later-succes- sional plant species in LD and high-diversity HD treatments, the number of colonizing species follows without any significant interaction, just as the cover ratio of sown species. Lastly, the significance of total biomass analyzing both treatments separately has to be emphasized. Among them, the annual behaviour was similar, but in 1998 there was a great increase of biomass in LD sowing mixtures, which allows the significant interaction (Figure 14).

Results of repeated measurement ANOVA (P-values) are shown in Table 4. The

ANOVA model included three main plot factors (treatment, year and treatment-year interaction). The increase of colonizing species cover was significantly lower in LD, which again supports the increase of biomass in 1998, being also lower in HD. Therefore, this increase in the community is more prominent in LD treatments. The effect of sown species was not so important in this increase.

Suppression of early colonizing plant species was presumed to be one of the primary mechanisms driving secondary succession in old-field plant communities [8] [27] . The

development of plant communities is often affected by the presence of strong dominants [47] . At locations with invasion potential of the species from a nearby target community, one should be aware that sowing mixtures could suppress desired natural colonization. It seems that highly suppressive mixtures were either the high diversity ones, or the ones composed of strong competitors. However, the presence of a strong competitor in a mixture not only suppresses the colonizers, but also leads to decreased survival of other sown species.

The sown species failed to establish successfully and were unable to eliminate the weeds for competition. Until 1998, the sown species were successful in suppressing the colonizer species, but their competitive effect decreased noteworthy the following year, when the number of colonizer species increased significantly (F = 30.30; 6 d.f.; P < 0.000). From this year until the end of the experiment in 2004, the number of colonizing species diversity in both plots stabilized because they did not differ significantly from the peak reached in 1999First, the suppressing effect of the high diversity treatment showed more homogeneity among replicates than that of the low diversity mixtures. This demonstrates that the different low diversity mixtures had variable effects on weed suppression. Starting from the second season, the number of colonizers began to differentiate according to the treatments, being on the lowest average in the HD and highest in the NC plots. The development of the community is often affected by the presence of strong dominants, and it depends on the local conditions and on the presence of species that are able to attain high dominance under given condition [23] .

When there was a significant increase in the number of colonizer species in high diversity seed mixtures (1999 and 2004), it is largely due to the divergence in the number of colonizer species among blocks, but most of all the blocks 1, 3 and 5 contributed mainly to this differentiation (Figure 2). For the other years when the number remained constant in all blocks, there was a similarity in the behavior of all blocks. On the other hand, in low diversity seed mixtures, the number of colonizer species was different for each of the blocks, for each of the years, due to the effect of different sown species mixtures. Nevertheless, the common behavior of treatment was due to combined effect of four blocks. The block 2 where there was mixed B. inermis and Sanguisorba minor, was only taking a low number of colonizer species during all experimental process, even in 1999 and 2004 when there was a significant increase in the number of those colonizer species.

For every year, except 1996, the low values of index D₁ (<0.4) found showed that in all communities not subjected to sowing effect (NC), there was no dominant species that affects the growth of the other species by competitive exclusion. Similarly, the high level of variability of HD/NC ratio among years; demonstrates that in some of them, the sown species were more effective in suppressing the numbers of colonising species (low values), especially in 1997 and 1998. The general found model shows that in most of years, there was no dominant specie in NC, with the exception of the first year of experimentation (1996), where only the perennial colonizer grass Poa sp. species showed a high dominance. In this same year, the annual other-forb, Polygonum aviculare, was also dominant, but only in this year for both sowing of low and high diversity seed mixtures treatments.

The negative relationship (r = −0.87) found supports that, low values of index D_1, the number of dominant colonizer species by block increased for most of the years. This fact was very relevant, because the sowing effect may be explained not only by suppressing colonizer species, but also for its effect on the number of dominant species income, that can be considered as a positive effect on the increase of diversity by sowing.

At the beginning of the experiment, only Polygonum aviculare was dominant and was quickly replaced by the sown species, but mostly by B. inermis and Sanguisorba minor. Nevertheless, at the same time that some species disappeared, other species began to dominate in the established community in HD from 1999; the other annual forbs Barkhausia taraxacifolia and Lactuca serriola were the most important in this year. In 2002, the annual grass Vulpia bromoides was emphasized. In 2003, B. taraxacifolia y L. serriola were again dominant, together with a high number of colonizer species that remained until the following year (2004), but the other annual-forb species Geranium molle and the other perennial-forb species Carduus tenuiflorus were the most important among the colonizer dominant species.

Large differences were recorded among individual sown species. Some of them were abundant in sown plots, but rarely or never spread to other plots. This suggests that dispersal limitation is an important factor for determining the composition of plant community [48] [49] . In general, plant traits are important determinants of species success in restoration [50] , and the ability to spread is amongst the most important ones.

The main goal that was obtained through diversity manipulation was to identify the effect of biodiversity on ecosystem processes and its implications in the restoration of abandoned arable lands. In natural conditions, there are many limitations due to other factors correlated with diversity, species composition and other characteristics common to herbaceous plant species, and can minimize the importance of diversity on some ecosystem processes [51] [52] . Experimentally, this problem can be solved, but further effect remains; the species identity [33] , which in turn can also be solved if in the experimental design, it was considered that each LD treatment contains sown species mixtures that were a subset of the species sown mixtures in HD treatment, so that each one have the same importance in all examined levels.

The cover of sown species in LD and HD treatments for the experimental sampling period was compared with a PCA analysis, in order to estimate which of them showed a similar performance under conditions of different sowing, and to separate in this way, the diversity effects and the species identity on the removal of weeds. The model allows us to know if the cover of sown species in HD were independent of cover of the same species in the LD blocks.

The HD mixtures were more successful when comparing the success of sowing of HD and LD mixtures, measured either by proportion of sown species, by suppression of natural colonizers, or by the total biomass, on average. However, the most successful LD mixture was usually equally successful as the HD mixtures. This suggests that diversity was important primarily for its “insurance effect” [23] [53] . Whilst some of the LD mixtures failed to establish successfully, the HD mixtures were always able to compensate the failure of some species to establish. Often, the presence of the dominant grass was enough for the LD mixture qualify as “the best”, even if the other species in the combination exhibited low cover, or even fail to survive. However, weed suppression and later-succession species invasion depended on the performance of species introduced. The strongest effect on weed suppression may result from the growth characteristics of the species used.

Community assembly often leads to a single stable equilibrium, such that the conditions of the environment and interspecific interactions determine which species will exist there. In such cases, regions of local communities with similar environmental conditions should have similar community composition. The resulting community depends on the assembly history, even when all species have access to the community. Both regional and local factors should determine the patterns by which communities assemble, and the resultant degree of similarity or dissimilarity among localities with similar environments.

Within the group of grasses, which competes properly in natural conditions, its participation in the community matrix during the experimental period was especially noticeable. Festuca rubra, Phleum pratense and Poa pratensis were successfully established in the plots, but they disappeared quickly, without becoming dominants, whereas Poa trivialis remains with very low values for cover during all experimental process. Within the group of other forbs, S. minor was important during all experimental time, just as P. lanceolata, but without maintaining its character of dominant until the end of the experiment. Among the remaining three species, none was important, and even the seeds of Galium verum were unable to grow. Within the group of sown legumes, Lotus corniculatus was the only species well established, but the persistence varied considerably over time. Medicago lupulina and the three species of Trifolium disappeared quickly.

The treatment × year interaction was highly significant for most of the variables tested, indicating that the growth of vegetation for individual year have high specific properties (20). Besides, the cover of colonizing species varied with respect to treatments and years, but it does not vary with respect to the number of species which naturally colonize the plots, this means that there was no sown effect over species richness.