Effects of several establishment modes of Miscanthus × giganteus and Miscanthus sinensis on yields and yield trends

Miscanthus is a C4 perennial grass originating from East Asia, the yields of which progressively increase in the first years of growth. Several species for bioenergy have been studied since the mid‐1980s in Europe, in particular (Miscanthus × giganteus [M. × giganteus]), due to its high yields. M. × giganteus is mainly cultivated in France and established from rhizomes. Our study aimed to assess, in field conditions, alternative establishment methods combined with an alternative species, Miscanthus sinensis (M. sinensis). We set up a multi‐environment experimental network. On each trial, we tested two treatments with M. × giganteus, established from rhizomes (G_r‐sd) and from plantlets obtained from rhizomes (G_p‐sd), and two treatments with M. sinensis seedlings transplanted in single (S_p‐sd) and double density (S_p‐dd). ANOVA was performed to compare establishment and regrowth rates across treatments, as well as yields across treatments and site‐years. A logistic model was used to describe yield trends and to compare the maximum yield reached and the rate of yield increase of both species. Results showed that miscanthus establishment from plantlets resulted in higher establishment (between 87% and 92%) and regrowth (between 91% and 94%) rates compared to establishment from rhizomes. Treatments with M. × giganteus obtained higher average yields across site‐years than those with M. sinensis, but more variable yields across site‐years. We showed a strong species effect on yields, yield components (shoot weight, shoot density and shoot number per plant) and light interception (through leaf area index). Lastly, to use M. sinensis established from transplanted plantlets as an alternative to M. × giganteus, research would be required on the breeding of M. sinensis sterile seeds to avoid risks of invasiveness.

M. × giganteus encounters difficulties, however, which impede its expansion in farmers' fields. First of all, the high cost of establishment (1,904-3,376 €/ha) due to rhizome costs (Xue, Kalinina, & Lewandowski, 2015) is a major constraint for farmers during the first year of production. In addition, the use of rhizomes often results in a low establishment rate because the quality of the rhizomes is not easy to check: rhizome quality depends on the rhizome harvesting period, harvesting technology, storage time between harvesting and planting, storage conditions and rhizome size (Xue et al., 2015). This low establishment rate leads to additional costs for the extra supply of rhizomes, or else to lower yields. Lastly, the genetic diversity of M. × giganteus is very limited because it is a sterile clone derived from M. sinensis and M. sacchariflorus, and only vegetative propagation is possible (Lewandowski, 1998;O'Loughlin, McDonnell, & Finnan, 2017).
M. sinensis has similar characteristics as M. × giganteus, including low nitrogen requirements and a high lignocellulose content. However, M. sinensis is an earlier species than M. × giganteus due to its lower base temperature (Zub & Brancourt-Hulmel, 2010). It is also more tolerant to water stress  and would therefore be more suited than M. × giganteus to soils with a low soil water content. Finally, M. sinensis is more tolerant to frost than is M. × giganteus Farrell, Clifton-Brown, Lewandowski, & Jones, 2006). However, yields of M. sinensis are lower than those of M. × giganteus in temperate climate (Zub & Brancourt-Hulmel, 2010).
Several establishment methods of M. × giganteus and M. sinensis were studied to find alternatives to the establishment from rhizomes (Atkinson, 2009;Xue et al., 2015;Xue, Lewandowski, & Kalinina, 2017). However, the success of these methods across different types of soils has not yet been documented. Moreover, the evolution of the biomass produced from alternative establishment methods over several growth years is not addressed in the literature (O'Loughlin et al., 2017). Only yields over the short term (first 2 years) for new establishment methods have been discussed for M. × giganteus (Boersma & Heaton, 2014a;Lewandowski, 1998).
It is therefore necessary to find alternatives to M. × giganteus, to facilitate the establishment phase and to broaden genetic diversity while maintaining satisfactory yields (Clifton-Brown et al., 2001). M. sinensis, despite having lower yields than M. × giganteus, could achieve more stable yields under various soil and climate conditions. It could, moreover, be interesting to explore new establishment methods to overcome the problems raised by the establishment from rhizomes. The use of transplanted plants could for instance allow for a better establishment of miscanthus and therefore less investment for re-establishment in case of non-emergence.
Our study aims to assess the effect of several modes of establishment of M. × giganteus and M. sinensis on yield and yield trends under field conditions and various environments in the north and centre of France.

| A multi-environment experimental network
A multi-environment experimental network was set up on seven sites in France from 2013 to 2019: six sites in the Ile de France region and one site in the Centre region.
We explored a diversity of soil types, with various textures, depths and stoniness (Table 1) to compare the yield variability of the treatments across sites. In addition to these varying soil characteristics, some trials were located in agricultural lands, that is, lands that were cultivated with annual crops before carrying out the experiment (Bioferme, La Bondue), but also in marginal lands such as Chanteloup, with a polluted soil due to the presence of Metallic Trace Elements, Polycyclic Aromatic Hydrocarbons and PolychloroBiphenyl, and Marne et Gondoire, which was located between several roads/highways and was therefore difficult to access with agricultural equipment.
Before setting up the trials, Subdray was managed as a set-aside land. La Bondue and Episy had been managed as set-aside lands since 1992. Marne et Gondoire and Evry had also been maintained as set-aside lands since 2003 and 2005 respectively. The preceding crop was winter barley (Hordeum vulgare) for Bioferme and Chanteloup (Table 1) Three sub-plots were positioned on each strip after the establishment phase. For single density (G_r-sd, S_p-sd, G_p-sd), five rows and five plants per row were selected for a sub-plot area of 16 m 2 (distance between plants and between rows: 0.8 m). For the double density (S_p-dd), the sub-plot area was 15 m 2 with five rows and 10 plants per row (distance between plants: 0.5 m; distance between rows: 0.6 m).

| Crop management over the whole trial
The establishment of miscanthus was carried out manually in March 2013 for all sites. During the first year of growth, all treatments were protected against weeds. Weed management was chemical and/or manual, depending on the amount of weeds observed in the trial. Given the low N requirements of miscanthus, no nitrogen fertilizer was applied throughout the experiment. 30 and 300 kg/ha of P 2 O 5 and K 2 O, respectively, were applied after harvest time in the third year of growth, except for the Bioferme, Episy and La Bondue sites, which were fertilized in the fourth growth year.

| Measurements
The aerial biomass of the whole trial was shredded the first year (end of winter 2014) and left on the soil as a mulch. To assess this initial aerial biomass (AB 1 ) without modifying the effect of the mulch inside the sub-plots, 10 plants inside each strip (treatment) but outside the sub-plots were selected, cut at 15 cm from the ground and weighed (SW s ). We counted the shoot number (SN s ) for these 10 plants. A 2-3 kg sample was put in the oven at 80°C for 48 hr to determine the DM. The number of plants on each sub-plot was counted (SN p ) and the yields measured in kg/m 2 . The sub-plot area (A p ) was 15 or 16 m 2 . We multiplied by 10 the formula described below to express the aerial biomass in t/ha. The aerial biomass produced during the first year was obtained as follows: where AB 1 is the aerial biomass the first year of production (t/ha of DM), SN s is the shoot number per m 2 of the sample, SW s is the shoot weight of the sample (kg), SN p is the shoot number of the sub-plot (kg), A p is the area of the sub-plot (m 2 ) and DM is the dry matter (%).
The establishment rate was measured in July 2013 (i.e. the first growth year) by determining the average ratio (over the three sub-plots of a given treatment) between the number of plants that have emerged or survived and the number of transplanted plants (for S_p-sd, S_p-dd and G_p-sd) or planted rhizomes (G_r-sd). For each sub-plot, 25 plants at single density and 50 plants for double density were considered for this measurement.
The regrowth rate was measured at the beginning of the second growing year at leaf development stage 1-12 according to the BBCH scale (Tejera & Heaton, 2017) to assess the quality of the regrowth of miscanthus. This corresponds for each treatment to the average ratio between the number of plants at leaf development stage in the second year and the number of plants at leaf development stage in the first growth year (over three sub-plots of a given treatment).
The LAI was assessed each growth year (i.e. from 2013 to 2018) between mid-July and mid-August. We collected all the green leaves of 10 shoots from each plot. The length and width of each leaf were measured and the leaf area index was calculated as follows: where LAI is the leaf area index, L is the length (m), W is the width (m), SN is the shoot number per m 2 , k is the correction coefficient for plant leaves varying across species. k amounts to 0.68 (Clifton-Brown, Neilson, Lewandowski, & Jones, 2000) and 0.61 (Laurent, pers. comm.), respectively, for M. × giganteus and M. sinensis.
Every year, from the second year of growth (i.e. 2015) until 2019, yield and yield components (shoot weight, shoot number per plant, shoot height, plant density) were estimated from the three sub-plots of each treatment before the harvest of the entire trial (i.e. in February or March). To assess yields, on each sub-plot we cut all shoots higher than 1 m (15 cm from the ground). The fresh matter of the biomass (FM) was calculated in the field using a scale. A sample of 1.5-2 kg of fresh matter (FW) was taken. This sample was immediately weighed in the laboratory to avoid water loss. The sample was then put in the oven for 48 hr at 80°C and weighed to get the dry weight of the sample. Sub-plot yields were in kg/m 2 , and the plot area (A) was 15 or 16 m 2 . We multiplied the formula described below by 10 to express yields in t/ha. The following equation was used to calculate the yield in t of DM per ha: where Y is the yield (t/ha of DM), FM is the fresh matter of the sub-plot (kg), FW is the fresh weight of the sample (kg), DW is the dry weight of the sample (kg) and A is the sub-plot area (m 2 ). Shoot number counting consisted in enumerating the shoots over 1 m high on each sub-plot. We then counted the number of plants on each sub-plot to determine the crop density (number of plants per m 2 ). Shoot height was measured using a tape measure. Lastly, shoot weight (kg) was assessed by dividing yield (expressed in kg/m 2 ) by the shoot number per m 2 .

| Statistical analysis
The analysis consisted of two steps: (a) we analysed the effect of treatments on establishment rate, regrowth rate, yield and yield components across site-years; and (b) we modelled the temporal evolution of yields to compare yield trends between treatments. Data analysis was implemented using R software 3.5.2 (R Core Team, 2018).

| Analysing the effects of treatments on establishment rate, regrowth rate, yield and yield components
Multi-factor ANOVAs were performed to compare the effect of the treatments on establishment rate, regrowth rate and yield.
where E ij is the establishment rate (%) in 2013 or the regrowth rate (%) in 2014 for the ith treatment of the jth site, T i is the ith treatment, S j is the jth site, µ E is the average establishment rate, e E,ij is the residual associated with E ij and α T and α S are the parameters associated with each factor. E ij is assumed to follow independent Gaussian distribution with mean zero and constant variance.
where Y ij is the yield or cumulative yield (t/ha of DM) for the ith treatment of the jth site-year, T i is the ith treatment, SA j is the jth site-year, µ y is the average yield, e y,ij is the residual associated with Y ij are the parameters associated with each factor. Y ij is assumed to follow independent Gaussian distribution with mean zero and constant variance. e y,ij ~ N(0, σ 2 y ).
The significance of the difference between treatments was evaluated using a Tukey test with a 0.05 confidence level. The average comparisons between treatments were performed with the multicomp package of R (release 1.4-10).
We performed single linear regressions between: (a) yield and shoot number; and (b) yield and shoot weight, to compare yield build-up between treatments and identify which yield component explains better yield variability.
These regressions were performed according to the following formula: where Y i is the explained variable (yield, LAI) for the ith site, X i is the explanatory variable (shoot number, shoot weight, LAI, shoot height) for the ith site, µ is the global average of the explained variable and α is the parameter associated with the explanatory variable. Y i is assumed to follow independent Gaussian distribution with mean and constant variance. e i ~ N (0, σ 2 ).
Clifton-Brown et al. (2000) assumed the yield to be a logistic function of the LAI. We used the SSlogis function of stats package of R (released 3.5.2) to parameterize this relationship. Logistic growth function g describing miscanthus yield trends over LAI. g is defined as follows: where LAI i is the leaf area index for the ith site. P 1,i , P 2,i and P 3,i are function parameters for the ith site. As we will not use the parameters of the function g, we will not detail them.
Based on the data design (repeated yield measurements over time for each site), the model was defined as a mixed effects model, that is, with one (or more) fixed and random parameters. In this model, three random parameters have been defined for the logistic growth: where T ij is the jth year of growth for the ith site. φ i is the function parameter vector (y max,i , φ 1,i , φ 2,i ) for the ith site. The first parameter (y max,i ) is the deviation asymptote from the curve towards the plateau, that is, the maximum yield for a mature miscanthus crop. The second parameter (φ 1,i ) represents the inflection point, that is, the year in when the crop has reached half of its maximum yield. The third parameter (φ 2,i ) gives an indication of the spread of the function, but it can also be interpreted as the time elapsed between reaching the second and the third quarter of the maximum yield. Equation (2) describes the distribution of yields within a given group (i.e. within a site-year). In this formulation, the sites were nested within the years to calculate the average values of the associated parameters. Y ij is the jth yield data collected in the ith site-year. Y ij is linked to the time of year since establishment T ij as follows: where Y ij is the yield of the miscanthus (t/ha of DM), f is the function linking Y ij to T ij and to a set of specific parameters φ i and e ij is the residual normally distributed term. All residuals are assumed to be independent, their distribution following a variance σ 2 .
Equation (3) describes φ k (i.e. y max , φ 1 and φ 2 parameters) by distinguishing fixed and random effects: The fixed effect β k represents the average values of the parameter φ k on all sites and b k represents the local φ k estimated deviation. It was assumed that the random effect b of the different parameters in φ i was related by a multivariate normal distribution of estimated variance-covariance matrix ψ.
This model was previously adapted to M. × giganteus by Miguez et al. (2008) and Laurent et al. (2015). We validated its use for M. × giganteus and its extension to M. sinensis by verifying that residuals did not have a particular structure and that their distribution was normal (Shapiro-Wilk test). We also compared the prediction error of the model by using the root mean square error (RMSE) obtained for the two species.
Finally, we performed an ANOVA on the estimated model parameters (y max , φ 1 , φ 2 ) to assess the differences in the multi-year trends across treatments.

| Establishment and regrowth rates
Establishment rates of all sites were higher than 75%. However, the establishment rate of M. × giganteus from rhizomes (77%) was significantly lower (p = .008) than the establishment rate of M. × giganteus from transplanted plants (87%; Figure 1a). The establishment rates of both M. sinensis treatments (S_p-sd and S_p-dd) were not significantly different and were similar to the establishment rate of M. × giganteus from transplanted plants (Figure 1a).
One year after the crop establishment, the regrowth rate of M. × giganteus established from rhizomes (86%) was significantly lower (p = .001) compared to the regrowth rate of M. × giganteus established from plants (94%; Figure 1b). Regrowth rates of both M. sinensis treatments (92% and 91%, respectively, for single and double density) were not significantly different.
Lastly, establishment and regrowth rates of M. × giganteus from rhizomes were more variable across sites than the treatments established from transplanted plants. The coefficients of variation of G_r-sd were 23% and 11%, respectively, for establishment and regrowth rates. These coefficients of variation were higher than those of G_p-sd (14% and 8% for establishment and regrowth rates, respectively) and much higher than those of S_p-sd (4% and 5% for establishment and regrowth rates, respectively) and S_p-dd (7% and 6% for establishment and regrowth rates, respectively). (2)

| Effects of treatments, sites and years on yields
Yields obtained for each treatment varied across sites and years ( Figure 2). Note that the 'yield' in 2014 corresponds to the quantity of shredded biomass for the first growth year. Yield trends of G_p-sd treatment in Chanteloup site are missing because the establishment of this treatment did not succeed (Figure 2b).
Years, sites and treatments had a significant effect on yields when taken alone and considering their interaction in pairs (Table 2). Yields were indeed affected by treatment × sites (p = 3.92 e-15 ), treatment × year (p = 1.39 e-7 ) and year × sites (<2.2 e-16 ; Table 2). We did not do a third level of interaction (treatment × sites × year) because we did not have enough degrees of freedom to do so.
G_r-sd obtained both the highest yield (20.5 t/ha of DM) and the lowest yield (2 t/ha of DM) of the trial network. G_p-sd yields ranged from 2.1 to 19.5 t/ha of DM across siteyears. S_p-sd and S_p-dd obtained yields ranging from 1.8 to 11 t/ha of DM and from 1.4 to 11.9 t/ha of DM, respectively, across site-years ( Figure 2).
Over the site-years, G_r-sd and G_p-sd (both in M. × giganteus) achieved similar yields (Figure 3a). We conclude that M. × giganteus can be established using transplanted plants from rhizomes without impacting yields compared to conventional establishment (from rhizomes).
There were, moreover, no significant differences in yields between S_p-sd and S_p-dd treatments (Figure 3). As a result, transplanting M. sinensis plants at double density (S_p-dd) did not produce higher yields than transplanting it at single density (S_p-sd).
Average yields of treatments with M. × giganteus (G_r-sd and G_p-sd) were significantly higher than those with M. sinensis (S_p-sd and S_p-dd): they amounted to 8.1 and 8.6 t/ha of DM for G_r-sd and G_p-sd, respectively, compared to 5.6 and 5.8 t/ha of DM for S_p-sd and S_p-dd, respectively

F I G U R E 2 Yield trends for
G_r-sd (a), G_p-sd (b), S_p-sd (c) and S_p-sd (d)  ( Figure 3a). Lastly, yield variability was high for the G_r-sd (coefficient of variation = 73%) and G_p_sd (coefficient of variation = 64%) compared to S_p-sd (coefficient of variation = 55%) and S_p-dd (coefficient of variation = 54%). Both coefficients of variation of M. sinensis treatments showed that yields were more stable than M. × giganteus treatments. Cumulative yields of G_r-sd (49.9 t/ha) and G_p-sd (54.8 t/ha) treatments were not significantly different (Figure 3b). There were also no significant differences between the cumulative yield of both treatments of M. sinensis (S_p-sd = 34 t/ha and S_p-dd = 36.1 t/ha). The cumulative yield of G_p-sd was significantly higher than the cumulative yields of both treatments of M. sinensis (S_p-sd and S_p-dd), whereas the cumulative yield of G_r-sd was significantly higher than the cumulative yield of S_p-sd.

| Treatments effect on yield components
We observed a strong species effect on the relationships between yield and shoot weight (Figure 4a), between yield and shoot number per ha (Figure 4b), between shoot number per ha and plant density (Figure 5a), between yield and LAI ( Figure 6a) and between LAI and shoot height (Figure 6b). M. × giganteus G_r-sd and G_p-sd modalities did indeed have similar yield build-up, differing from M. sinensis S_p-sd and S_p-dd modalities, which also had similar yield build-up.
Yields of M. × giganteus were more correlated to shoot weight (R 2 = .79 and .76, respectively, for G_r-sd and G_p-sd) than to shoot number (R 2 = .63 and .44, respectively, for G_r-sd and G_p-sd). Yields of M. sinensis were more correlated to shoot number per ha (R 2 = .63 and .63, respectively, for S_p-sd and S_p-dd) than to shoot weight (R 2 = .26 and .51, respectively, for S_p-sd and S_p-dd). The slopes of the linear regressions described in Figure 4a (0.50 for G_r-sd and 0.43 for G_p-sd) showed that yields of M. × giganteus increased more slowly with shoot weight than yields of M. sinensis (0.86 for S_p-sd and 1.31 for S_p-dd). The slopes of the linear relationship between yield and the shoot number showed that yields of M. × giganteus increased more rapidly with shoot weight (4.04 e-5 and 3.88 e-5 for G_r-sd and G_p-sd) than those of M. sinensis (6.62 e-6 and 7.32 e-6 for S_p-sd and S_p-dd; Figure 4b).

F I G U R E 3 Effects of treatments on miscanthus average yields (a) and cumulative yields (b) from seven sites across 5 years (2015-2019).
Boxplots with the same letter are not significantly different at the 5% threshold.  (Figure 5a). At the same plant density, M. sinensis (in S_p-sd) produced more shoots per ha than M. × giganteus (in G_r-sd and G_p-sd; Figure 5a). Lastly, despite an increased plant density, the number of shoots per ha of M. sinensis at double density (S_p_dd) was similar as M. sinensis at single density (S_p-sd; Figure 5a).
The number of shoots per plant (between 5 and 46 shoots per plant) of M. × giganteus was lower compared to that of M. sinensis (between 25 and 110 shoots per plant; Figure 5b). G_r-sd and G_p-sd treatments obtained similar numbers of shoots per plant. M. sinensis at single density produced more shoots per plant than M. sinensis at double density (Figure 5b).
M. × giganteus observed yields were better correlated with simulated yields following a logistic function of the LAI (R 2 = .90) than were M. sinensis observed yields (R 2 = .67; Figure 6a). At similar LAI values, M. × giganteus (G_r-sd and G_p-sd) obtained higher biomass than did M. sinensis (S_p-dd and S_p-sd; Figure 6). The highest yields were obtained for LAI values greater than 5 for both species.
Shoot heights of M. × giganteus were more positively correlated to LAI (R 2 = .76) than those of M. sinensis (R 2 = .60). The slope of the regression line of M. × giganteus (2.28) showed in Figure 5b was lower than that of M. sinensis (5.04). The highest LAIs were obtained for the largest shoot heights. This relationship could be explained by leaf production, which increased as the plant grew (Figure 6b). and G_p-sd). M. sinensis had an RMSE of 1.1 and 1.2 t/ha of DM, respectively, for S_p-sd and S_p-dd. The prediction error ratio (RRMSE between 37% and 41%) is close for both species, suggesting that the model is just as efficient for M. sinensis yields as it is for M. × giganteus yields, or even better.

| Multi-year yield trends
Neither of the two parameters (y max ; representing the maximum yields achieved) were significantly different (a) between M. × giganteus treatments or (b) between both treatments of M. sinensis. Additionally, y max values of G_r-sd (12.2) and G_p-sd (13.7) were significantly higher than those of S_p-sd (7) and S_p-dd (7.2; Table 4). Parameters φ 1 and φ 2 , indicating the crop biomass evolution rate, were significantly different across treatments. M. × giganteus-G_r-sd and G_p-sd took longer than M. sinensis (S_p-sd and S_p-dd) to reach its maximum yield (Table 4).
M. × giganteus yields were beginning to reach the plateau in the sixth growth year. M. sinensis achieved a plateau between the second and third year of growing and then remained stable.
M. sinensis established from plantlets at double density tended to reach its maximum yield slightly faster than M. sinensis established from plantlets at single density ( Figure 7). However, the maximum yield was not significantly different for both treatments.

| Benefits of establishing miscanthus from plantlets
M. × giganteus established from rhizomes resulted in higher losses (23%) compared to the treatments established from plantlets (13%). Boersma and Heaton (2014a) found a similar result (equal to 23%) for establishment from rhizomes. To overcome such losses, Christian, Yates, and Riche (2009) recommend using 14% more rhizomes. Losses of plants during the establishment from rhizomes have been explained by intrinsic properties of the rhizomes, lack of active buds, small rhizome size, loss of germination capacity due to bacterial or fungal attacks during storage (Covarelli, Beccari, & Tosi, 2012;Xue et al., 2015) and soil and climate conditions (Boersma & Heaton, 2014b). Establishment from plantlets overcomes these difficulties. Transplanted plantlets also eliminate the need for a re-establishment worksite and make it possible to avoid a significant delay in the stage between the plants resulting from the establishment and the re-establishment.
The regrowth rate of miscanthus established from transplanted plantlets is also higher than establishment from rhizomes. The order of magnitude of regrowth rate of miscanthus established from rhizomes observed in our study (88%) is consistent with the one estimated by Pyter, Dohleman, and Voigt (2010), who found a regrowth rate of 88%. In contrast, Boersma and Heaton (2014a) found a regrowth rate higher than those we encountered (98.8%). This discrepancy with our results could be explained by the management method chosen by these authors, who replaced the dead plants before the winter so as to have 100% survival at the end of the first year. Plant losses can be due to drought or frost. The main source of plant losses during regrowth is due to winter frost. Devastating winter losses have been widely discussed in the literature (Clifton-Brown, 1997;Clifton-Brown, Breuer, & Jones, 2007;Kucharik, VanLoocke, Lenters, & Motew, 2013;Purdy et al., 2013). Losses related to the establishment from rhizomes could lead to gross margin reductions of up to 50% over the whole miscanthus production period (Zimmermann, Styles, Hastings, Dauber, & Jones, 2014). The better establishment from transplanted plants could reduce these economic losses.

| A strong genotype effect on yield build-up
M. × giganteus obtained higher average yields than M. sinensis. Our results are in line with those of Clifton-Brown et al. (2001) and Lewandowski et al. (2003) obtained in European conditions. In Asian conditions, M. sinensis obtained higher yields than M. × giganteus: Nakajima et al. (2018) and Yu, Ding, Huai, and Zhao (2013) observed higher average yield for M. sinensis (~30 t/ha of DM) than for M. × giganteus (~20 t/ha of DM). Asian conditions are characterized by much higher annual rainfall (exceeding 1,000 mm) than in Northern Europe (where average annual rainfall amounted to 600 mm), which could explain the opposite yield differences between M. sinensis and M. × giganteus. Higher yields of M. × giganteus could also be explained by its lateness, as late genotypes have been shown to achieve higher yields (Gauder, Graeff-Hönninger, Lewandowski, & Claupein, 2012;Jensen et al., 2013;Lewandowski et al., 2000).
Yield differences between the treatments were linked to a strong genotype effect, which in particular determined yield potential. M. sinensis showed variability between plants within the same strip. In particular, shoot diameter, shoot height and shoot number per plant differed across plants within a given treatment-site-year. In contrast, M. × giganteus showed an architectural uniformity of plants within a plot. Plant height, leaf and shoot proportion are key factors explaining the variability in biomass production | 535 OUATTARA eT Al.
between species (Zub & Brancourt-Hulmel, 2010). Lim et al. (2014) found similar results from the analysis of 66 miscanthus genotypes: they showed that plant height and shoot weight were positively and significantly correlated with yield.
Our results showed that yields of M. × giganteus were linked to shoot weight while those of M. sinensis were essentially determined by the number of shoots per ha. A canopy architecture specific to each species is therefore established, determined by plant height (3.5 m for M. × giganteus compared to 2.5 m for M. sinensis) and by the number of shoots per plant, which was higher for M. sinensis (between 25 and 110 shoots per plant) than for M. × giganteus (between 5 and 46 shoots per plant). Similar results were found by Zapater et al. (2017)  The LAI values that we observed reinforce this species effect. At similar LAI value, M. × giganteus produced higher yields than did M. sinensis. Our results showed that maximum yields were reached when LAI is higher than 5 for both species, and with a more efficient use of the solar radiation intercepted by M. × giganteus compared to M. sinensis. Beyond that value, we observed a yield plateau. Our results are consistent with those of Defra (2007) that showed that the LAI value for which miscanthus yields are maximum varied between 5 and 6. Our results are also consistent with those of Davey et al. (2017) who found maximum LAI values of 6 for M. × giganteus and 4.3 for M. sinensis with more efficient use of the incident Photosynthetic Active Radiation (PAR) of M. × giganteus compared to M. sinensis. Davey et al. (2017) noted that from an LAI of 3.5, M. × giganteus could intercept 90% of the incident PAR. The yield plateau we then described for higher LAI values could be due to self-shading of the leaves which limits photosynthesis and no longer allows an increase in yields (Davey et al., 2017;Defra, 2002).

| Yield variability across site-years
Beyond the effect of treatments, we observed a variability related to sites and years, with a higher effect of years than sites. Our results also showed a greater inter-annual and inter-site variability of M. × giganteus than M. sinensis. High average yields of M. × giganteus were associated with high yield variability across sites, as shown by Lesur et al. (2013) and Miguez et al. (2008). Similar results to those of our study were obtained by Laurent et al. (2015) and Clifton-Brown et al. (2007), which showed that yields of M. × giganteus varied across sites and years. Kalinina et al. (2017) also found interannual and inter-site variability for several M. sinensis genotypes.
The inter-site variability of yields could be related to the edaphic variability of our trial network, characterized by contrasting soil types that can generate a range of crop sensitivity to stress, particularly water stress. As the annual yield variability could be linked to crop age and to climatic variability, it is not straightforward to explain them. Further studies on our trial network could be carried out to identify the main factors responsible for the variability in miscanthus yields across siteyears, in order to draw conclusions for a better management of miscanthus and an anticipation of its location.

| Which perspectives for the management of miscanthus?
Describing yield trend curves could be useful for the management of M. × giganteus and M. sinensis established from plantlets.
Despite the losses at establishment and regrowth periods, M. × giganteus established from plantlets showed a yield increase rate close to that of M. × giganteus established from rhizomes. According to the model, the maximum yield of M. × giganteus is reached after 6 years of cultivation or more. Our results are consistent with those of Lesur et al. (2013) and Christian, Riche, and Yates (2008) who found that the maximum yields of M. × giganteus were reached after 8 years of growth. This proximity of yields trends could be explained by a similarity of the physiology of the two establishment modes of M. × giganteus. After the year of establishment, both treatments restarted in the same period each spring and the growth and development process were identical.
M. sinensis established in single and double density reached similar maximum yields. These results are consistent with those of Lesur et al. (2013) and Miguez et al. (2008) who found no effect of plant density on maximum yields for M. × giganteus. Atkinson (2009) obtained results which differed from ours: they showed that increasing the establishing density from 10,000 to 40,000 plants/ha increased yields. This difference with our results could be explained by the larger difference between the two densities than the one we tested, and by the use of a plant density (10,000 plants/ha) that is below the recommended plant density (around 15,000 plants/ ha). The slightly faster yield increase of double density in the first years of growth for M. sinensis established from plantlets could be linked to the higher density of plants. The crop therefore covered the area faster and received more light, which enabled more biomass production per area unit. However, after the first few years of growth, there would be greater competition for light, but crops would also be at the maximum of their LAI, and therefore at the maximum of their production (Bullard, Heath, & Nixon, 1995;Lewandowski et al., 2000).
The successful results of establishment from transplanted plants obtained for both species open the way to additional possibilities for miscanthus establishment.
Establishing miscanthus from transplanted plants could be an interesting approach, as it would reduce losses during the establishment phase. However, the resulting profitability of this crop management route needs to be assessed, by taking into account establishment costs and resulting yields. Xue et al. (2015) estimated the cost of establishment from rhizomes at ~1,900-3,400 €/ha and the cost of establishment from plantlets at ~4,200-4,400 €/ha. Establishment from rhizomes is therefore less expensive. However, the multiplication power of rhizomes (×10) is low compared to the multiplication power of transplanted plants (×30), which could create a potential gap between supply and demand for establishment from rhizomes in case of large-scale plantation (Xue et al., 2015). The high cost of establishment from plantlets is related to the cost of labour and energy for pre-growing. According to Xue et al. (2015), by improving the multiplication rate of establishment from transplanted plants, the cost of establishment could be reduced to ~2,600 €/ha which would be lower than or equivalent to the cost of establishment from rhizomes. Establishment from seedlings or direct seeding are the least expensive establishment modes (Hastings, 2017), but so far M. sinensis seeds are not sterile. Research is required on the breeding of M. sinensis sterile seeds to avoid risks of invasiveness . Lastly, a study could be carried out to assess the technical feasibility of establishing miscanthus from transplanted plants in farmers' fields, by mechanizing this establishment mode.