Towards Miscanthus combustion quality improvement: the role of flowering and senescence

Abstract In commercially grown Miscanthus × giganteus, despite imposing a yield penalty, postwinter harvests improve quality criteria for thermal conversion and crop sustainability through remobilization of nutrients to the underground rhizome. We examined 16 Miscanthus genotypes with different flowering and senescence times for variation in N, P, K, moisture, ash, Cl and Si contents, hypothesizing that early flowering and senescence could result in improved biomass quality and/or enable an earlier harvest of biomass (in autumn at peak yield). Ideal crop characteristics at harvest are low N and P to reduce future fertilizer inputs, low K and Cl to reduce corrosion in boilers, low moisture to reduce spoilage and transportation costs, and low Si and ash to reduce slagging and consequent operational downtime. Stems and leaves were harvested during summer, autumn and then the following spring after overwinter ripening. In spring, stem contents of N were 30–60 mg kg−1, P were 203–1132 mg kg−1, K were 290–4098 mg kg−1, Cl were 10–23 mg kg−1 and moisture were 12–38%. Notably, late senescence resulted in increased N, P, K, Cl, moisture and ash contents, and should therefore be avoided for thermochemical conversion. Flowering and senescence led to overall improved combustion quality, where flowered genotypes tended towards lower P, K, Cl and moisture contents; marginally less, or similar, N, Si and ash contents; and a similar higher heating value, compared to those that had not flowered. Such genotypes could potentially be harvested in the autumn. However, one genotype that did not flower in our trial exhibited sufficiently low N and K content in autumn to meet the EN plus wood pellet standards for those traits, and some of the lowest P, moisture and ash contents in our trial, and is thus a target for future research and breeding.


Introduction
Miscanthus is a perennial energy crop with C4 metabolism. It can produce high yields from low inputs across multiple environments in temperate regions as well as in the tropics (Mccalmont et al., 2015). The most cultivated Miscanthus species for biomass production in Europe and North America is M. 9 giganteus, which exhibits rapid growth, low mineral content and high yield (Linde-Laursen, 1993;Lewandowski & Kicherer, 1997). Miscanthus 9 giganteus resulted from a cross in the wild between diploid M. sinensis (2n = 2x = 38) and tetraploid M. sacchariflorus (2n = 4x = 76) (Greef & Deuter, 1993;Linde-Laursen, 1993;Rayburn et al., 2009) and is sterile, prohibiting improvement through breeding. However, the Miscanthus genus comprises 13 or so species (Greef & Deuter, 1993;Hodkinson et al., 1997) of high diversity, providing considerable genetic and phenotypic resources to improve Miscanthus both in terms of quality and quantity of harvested biomass. Currently the crop is grown for heat and power, although in the wider bioeconomy there is also interest in its use for green chemistry (Parveen et al., 2011), biomaterials (Uihlein et al., 2008) and transport fuels (Brosse et al., 2012).
During combustion, the physical and chemical properties of the biomass itself, such as lignin, cellulose, hemicellulose, moisture content and elemental composition, strongly influence factors such as calorific value, ash and its behaviour, heat exchange and emissions (Robbins et al., 2012;Shao et al., 2012), and the fuel characteristics of biomass are very different to those of fossil fuels (Shao et al., 2012). Efforts have therefore been made to understand the different fuel properties in order to optimize combustion efficiency and reduce operational problems. All biomass species will include varying degrees of N, P, K, S, Ca, Mg, Na and Si, but concentrations of K are usually considerably higher in biomass fuels compared to fossil fuels and, in grasses such as Miscanthus, high Cl concentrations are also common (Shao et al., 2012). Compared to coal, Miscanthus generates lower concentrations of SOx, NOx and HCl, but higher concentrations of KCl (Khodier et al., 2012) because of the higher concentrations of these elements, which are mainly in the form of water soluble inorganic salts that are easily volatilized during combustion. These chloride ions have a low melting temperature (<700°C), and so form sticky layers on heat exchangers or heat transfer surfaces on boiler surfaces (Shao et al., 2012) and thus cause fouling, which can then inactivate catalysts (Lewandowski & Kicherer, 1997;Obernberger et al., 2006). Si can also be high in some Miscanthus genotypes (Monti et al., 2008), and in association with alkaline metals it can melt or sinter at 800-900°C (Jenkins et al., 1998;Baxter et al., 2014). These alkali silicates and mixed alkali and/or calcium chlorides/sulphates tend to deposit on reactor walls or heat exchanger surfaces inside the boiler causing fouling and corrosion, even at low fusion temperatures (Wornat et al., 1995;Jenkins et al., 1998;Wang et al., 2012).
Various technologies have been explored to reduce the ash deposition and corrosion associated with firing/cofiring high-alkali biomass fuels. Combustion and flue gas temperatures can be reduced during biomass thermochemical conversion (Xue et al., 2014), but this is energetically less efficient. Combustion systems can be designed, to some extent, to ameliorate the negative effects of elemental composition, but the accompanying increase in production and maintenance costs (slagging must be manually removed) extend operation downtime. Feedstock can be pretreated to reduce the alkali metals, and the addition of Al-Si-based, S-based, Cabased and P-rich substances has also been explored. For example, lime and limestone have shown some effectiveness in ameliorating ash deposition and corrosion, most probably by diluting the ash, or altering the adsorption of the porous surfaces of the alkali salts when they are calcinated (rather than chemically reacted with alkali metals or alkali containing compounds). The addition of Ca-based additives (i.e. CaO, CaCO 3 and Ca(OH) 2 ) may be effective for P-and Krich biomass fuels, by helping to convert gaseous K species into high-melting-point potassium silicates/phosphates, but for fuels containing high contents of K, Si and Ca, P-based additives might be useful to reduce ash sintering and bed agglomeration (Shao et al., 2012). Thus, the choice of additive is dependent on the fuel type.
Another strategy, which is the subject of this study, is to optimize the biomass composition, in this case through variety selection by identifying plants with desirable flowering and/or senescence characteristics.
Despite originating from a natural cross in the wild, M. 9 giganteus has good combustion traits compared to other lignocellulosic crops, such as giant reed, switchgrass and sorghum (Monti et al., 2008;Cadoux et al., 2014;Iqbal & Lewandowski, 2014), and recent studies have indicated that the quality of M. 9 giganteus biomass can compete with wood pellets (Iqbal & Lewandowski, 2014). Miscanthus plants remobilize nutrients from above-ground vegetation to storage rhizomes below-ground at the end of the growing season. Delaying harvest from autumn, when biomass accumulation ends, to the following spring therefore results in improved quality through reduction in moisture, N, K and Cl contents. This also reduces or negates the need for fertilizer application (Beale & Long, 1997;Lewandowski & Kicherer, 1997). A negative consequence of delayed harvest is yield loss of up to 30% due to leaf drop and broken stems over the winter (Clifton-Brown & Lewandowski, 2002;Strullu et al., 2011), although this helps to build up soil carbon (Mccalmont et al., 2015), and leaf loss improves quality as ash content is higher in this tissue type (Monti et al., 2008).
Plant developmental processes, such as flowering and senescence, have been postulated as important in promoting nutrient remobilization and thus improving biomass quality (Clifton-Brown et al., 2001;Clifton-Brown & Lewandowski, 2002), with senescence shown to negatively correlate with moisture content across diverse Miscanthus genotypes (Robson et al., 2012). The selection of genotypes that flower and senesce promptly and efficiently at the end of the growing season could therefore provide a strategy to create a crop: (i) with more desirable quality characteristics or (ii) that can be harvested prewinter, permitting the optimization of both yield and sustainability, and a better phasing of biomass supply to end users.
In this study, 16 genotypes (including M. 9 giganteus) were chosen from a population of 244 Miscanthus plants to represent the greatest diversity in the timing of flowering and senescence and used to understand the relationship between variation in the timing of these plant phenologies and biomass quality for N, P, K, S, Na, Cl, Si and ash contents, and heating value.

Trial conditions
A trial of divergent Miscanthus germplasm was established in 2004-2005 on a sloping field (52°26 0 N 04°01 0 W) near Aberystwyth on the west coast of Wales. The soil has been classified as a Cambic stagnogley (FAO, 1998). The stone fraction (particles >2 mm) was estimated at approximately 50% of the soil mass in the 0-40 cm layer. At the time of planting (2004/5) the soil organic carbon was ca. 100 Mg ha À1 . Climate data (temperature and radiation) were obtained from a weather station on site, whilst rainfall data were collected from a nearby weather station (52°25 0 N 04°01 0 W).
Average monthly rainfall for 2009 ( Fig. 1) was 98% of the long-term monthly average for Gogerddan (86.5 cm). The calculated balance between rainfall and evaporation showed that mild soil moisture water deficits occurred in July, September and October in 2009, where isolated monthly values were low for rainfall and high for solar radiation (Jensen et al., 2011b), although these had no detectable impact on growth (based on weekly measurements of change in height of the crop canopy; data not shown). Solar radiation in 2009 was slightly higher (105%) than the long-term average of 9.4 MJ m À2 day À1 . Monthly average maximum and minimum temperatures in 2009 were similar to the long-term mean, and soil temperatures at a depth of 5 cm did not fall below À1°C. Once growth had started, minimum air temperatures did not drop below freezing until 1 December, and the lowest temperature reached during the experiment was À7.9°C, in January 2010. Monthly average maximum and minimum temperatures were similar to the long-term mean. Soil temperatures at a depth of 5 cm did not fall below -1°C.

Plant material
Sixteen genotypes, from a larger collection of 244, were selected to represent maximum diversity with respect to flowering and senescence (Table 1). The selection was based on the evaluation of previously recorded phenology (Jensen et al., 2011a;Robson et al., 2013a,b). Categories for flowering and senescence were allocated to each genotype (mean values for each genotype were used, n = 4). Flowering category was based upon the exertion of the first flag leaf on or before 21 July 2009 (early (E)), 25 August (mid (M)), after 25 August (late (L)) or not at all (non (N)). Senescence category was based upon the loss of >80% greenness of each genotype across the four replicates before 23 October (E), 24 November (M) or later (L). A number is included where more than one example was available for a particular phenotype. The genotypes were thus categorized as EE1-3, ME1-3, MM1, LM1-3, LL1, NE1, NM2 and NL2 (there were no representatives for the EM, EL, ML or LE categories). Table 1 lists the categories, species and details of origin (where known).

Design
A randomized block design was used, with four replicates. Stems were sampled from the plots on each of the harvest dates. Although this study was conducted over a single growing season, our main objectives were as follows: (i) to determine the extent of variation in mineral content in diverse Miscanthus genotypes and (ii) to determine the possible influence of flowering and senescence on mineral composition. Previous comparisons of flowering and senescence across 244 genotypes showed that rank orders for both traits were similar when compared across 3 years (Jensen et al., 2011a;Robson et al., 2012). Because a subset of these genotypes was used in the present study, along with a broad categorization, we were able to compare the relative performance of these genotypes in detail across a single year.

Sample preparation and analysis
Leaf and stem material were collected on 24 June 2009 (summer), 23 October 2009 (autumn) and 11 February 2010 (spring). Stems to be harvested were chosen at random by inserting a cane marked along its length into the plant and choosing the stem closest to each mark. The number of stems chosen was based on an estimate of how much material would be required to complete elemental analyses (and therefore differed according to stem size, which differed between plants).
Moisture content was determined by taking fresh weight, and oven-drying at 50°C until a constant weight was reached. Residual dry matter was determined gravimetrically as the residue remaining after drying 1 g sample in a convection laboratory oven (Carbolite, Hope Valley, Hope, UK) at 102 AE 2°C for at least 16 h. Subsequently, ash content was determined  gravimetrically as the residue remaining after ignition in a muffle furnace (Carbolite) at 550°C for at least 16 h. N was analysed by a rapid combustion method using a LECO FP-428 analyser (LECO Corp., St. Joseph, MI, USA) according to ISO 17025 standards. For the determination of K, Na and P, samples were prepared in accordance with AOAC (1995), MAFF (1986) and Undersander et al. (1993) and analyses were performed on dried and milled material. One gram of sample was weighed into 100 ml Kjeldahl tubes, 15 ml aqua regia was added and allowed to soak overnight. Samples were digested on a heating block at 120°C for 3 h. Extracts were analysed using a Varian Liberty ICP-AES (Agilent Technologies, Santa Clara, CA, USA).
Due to budget limitations and the expense of these types of analyses a subset of eight, from the original 16, genotypes were selected for HHV, Cl and Si analysis. Subset selection was based on representing each type of flowering and senescence classification (e.g. early, medium, late). The calorific, or higher heating value (HHV), was determined using the equation developed by Channiwala & Parikh (2002) as: HHV = 0.3491C + 1.1783H + 0.1005S À 0.1034O À 0.0151N À 0.021 ash MJ kg À1 .

Statistical analysis
All statistical analyses applied to data were performed using Genstat statistical software package (17th edition; VSN International Ltd., Hemel Hempstead, UK). The analyses combined the data from all harvests. As the amount of random variation differed between harvests, a meta-analysis was performed by fitting a linear mixed model, using the REML algorithm (Patterson & Thompson, 1971), in which a different residual variance was estimated for each harvest. Replicates within harvests were fitted as random.
The fixed model contained main effects and interactions of harvest, flowering, senescence and genotype. The terms were nonorthogonal due to the diversity in flowering and senescence over the combinations of levels of the four factors, thereby causing unequal replication of those phenologies. For example, there were no samples that had flowered or senesced at the summer harvest, and no samples that had not senesced at the autumn harvest. The order of fitting was thus important, when assessing the terms. To be sure that a factor was genuinely significant, the other factors were fitted first (i.e. eliminated) so that it was clear that any significant difference could not be explained by effects other than the factor of interest. Similar issues arose when assessing interaction terms. The overall effects of harvest were of less interest, so this was always fitted first. The genotype factor was always fitted last to be sure that any significance could not be explained by the varied flowering and senescence phenologies of different genotypes. There were a few genotypes at any single harvest time where some replicates had flowered and other replicates not flowered, or where some replicates had senesced and other replicates had not senesced. In these cases, there was insufficient information to be able to assess the effects of flowering or of senescence after fitting genotype. These were therefore always fitted after harvest but before genotype. Two models were assessed. In the first, flowering was fitted before senescence, and in the second senescence, it was fitted before flowering. As these developmental stages were always fitted before genotype, any Genotypes are classified by (a) flowering status (1st letter) and (b) senescence status (2nd letter), based on (a) the genotype exerting its first flag leaf on or before: 21 July 2009 (early (E)), between 22 July and 25 August (mid (M)), after 25 August (late (L)) or not at all (non (N)); and (b) the genotype reaching >80% loss of greenness before 23 October (E), 24 November (M) or later (L). Mean values for each genotype were used (n = 4). A number is included where more than one example was available for a particular phenotype.
conclusions about their effects are on the assumption that it is whether or not the samples had flowered (or senesced) that is the causal effect, not the distribution of the genotypes amongst the groups of flowered or nonflowered (or senesced or nonsenesced) at each of the harvests. The assessments of the fixed terms were made using F statistics, with the denominator degrees of freedom estimated using the method of Kenward & Roger (1997). Genstat provides two tables. The first assesses the terms sequentially in the order of fitting, whilst the second assesses the effect of dropping terms from the full model. However, a term can be dropped only if there are no higher-order terms that contain it. For example, the main effect of a factor cannot be dropped if the model contains an interaction involving that factor. Consequently, the first table was used for most of the tests reported below. The tables of predicted means use the effects from the full model, and these eliminate the effects of other terms. Consequently, they may not display the same significances as those detected using the F-tests in the sequential table. Standard correlations were calculated using the predicted means from the REML analyses of the stem data.
Plots of the residuals from the analyses identified that transformations were required for some of the variables: logarithms for N, Na, K, P, Cl and Si, and logit transformations for the percentages of ash and dry matter.

Results
Biomass samples were collected and characterized from stems and leaves, separately, of up to 16 genotypes at three harvest dates in spring, summer and autumn. We found extensive variation in all traits analysed, even in the spring harvest (February), which is when the crop would be harvested commercially for combustion. N, P, K, Si and ash contents were higher in leaves than in stems, but Cl and moisture content were higher in stems than in leaves (Tables 2-4).

Effect of harvest time and genotype on stem composition
In stems, N and K contents declined over the three harvests in 14 of the genotypes. In NL1 and 2 and LL, however, N and K declined between summer and autumn but then increased in spring, albeit to a concentration which was still lower than the summer value. P declined over the three harvests in 12 of the genotypes. In four genotypes (NL1 and 2, LL and NE), P declined between summer and autumn, but then increased in spring, again to a concentration which was still lower than in summer. Moisture content and ash declined over the three harvests in all 16 genotypes. At final harvest the stems of late-senescing genotypes contained higher N, P, K, Si, moisture content and ash, whilst early-senescing genotypes contained lower Cl. For example, the early-flowering and early-senescing genotype EE1 contained 45% of the Cl and 7% of the K contents of the genotype NL1 and contained 45% of the Cl and 19% of the K contents of LM1 (M. 9 giganteus).
Compositional traits exhibited varying degrees of association. Figure 2a, b shows the relationship between N, P, K, ash, MC and Cl using the predicted means for (i) flowering or (ii) senescence, genotype and harvest from the REML analyses of the stem data. The highest correlation (R = 0.96/0.95) was between Cl and K contents for flowering/senescence. Cl and K also showed correlations of >0.7 with MC (Table S2), whilst no obvious relationship was observed between Na and other variables (Table S2). Similar relationships and correlations were found using means for flowering and senescence, genotype and harvest, where the R values were either the same or slightly higher for flowering.
The earliest flowering occurred on 30 June, and ten genotypes initiated flowering on all four replicates, with the remainder either showing no sign of flowering or only flowered on one replicate. Compositional trait values at autumn and spring harvests (no flowering was observed at the time of the summer harvest) tended to be lower in stems that had flowered and/or senesced ( Fig. 2a, b), with the exception of genotype NM1, which exhibited consistently low concentrations of N, P, K and ash ( Fig. 2a, b; Table 2), despite never flowering.
In leaves, N content declined over the three harvests in four of the late (or non)-flowering genotypes. In the other 12 genotypes, N declined between summer and autumn but then stayed constant or increased in spring, to a concentration which was still lower than in summer. K contents declined over the three harvests in five of the late (or non)-flowering genotypes. In the other 11 genotypes, K declined between summer and autumn but then stayed constant or increased in spring, to a concentration which was still lower than in summer. P declined over the three harvests in 11 of the genotypes. In five of the genotypes, P declined between summer and autumn but then increased in spring, to a concentration which was still lower than in summer. Moisture content declined over the three harvest time points in all 16 genotypes. Ash declined over the three harvests in four genotypes. In 12 of the genotypes, ash increased or remained constant between summer and autumn, but then decreased in spring. Leaf elemental composition varied between harvest and genotype, and latesenescing genotypes contained higher N, but the overall pattern observed in stems, where concentrations generally fell over time, was not observed in leaves (Table 2).
In both leaves and stems, significant interactions between harvest and genotype, and harvest and flowering were identified for N, P, K, Na, ash and moisture content, with some exceptions: no interaction between harvest and flowering was observed for leaf N, stem Na     or leaf ash (Table 5a), and the interaction between harvest and genotype was only significant when fitted sequentially (when individual terms were dropped from the fixed model, the interaction was not significant) (Table 5a). Significant interactions were also observed between harvest and genotype for leaf Si, and leaf and stem HHV (Table 5b). Significant interactions between flowering and senescence were observed for leaf N (P = 0.002) and stem P (P = 0.005). No interactions were observed between flowering and genotype or senescence and genotype, or between flowering, harvest and Fig. 2 (a) Correlations between predicted means for nitrogen, phosphorus, potassium, ash, chlorine, silica and moisture content across three harvests in stems from 16 (eight for Cl and eight for Si) genotypes of Miscanthus. Symbols show whether the data are from the summer (o), autumn (+) or spring (9) harvests (1, 2 or 3 in key) and whether they had flowered by that harvest (in bold if flowered). (b) Correlations between predicted means for nitrogen, phosphorus, potassium, ash, chlorine, silica and moisture content across three harvests in stems from 16 (eight for Cl and eight for Si) genotypes of Miscanthus. Symbols show whether the data are from the summer (o), autumn (+) or spring (9) harvests (1, 2 or 3 in key). The combination of senesced and nonsenesced values was present at harvest 2 only, where senesced is indicated in bold.
genotype. However, a three-way table of predicted means is presented for N, P, K, Na, ash and moisture contents due to the significant interactions involving combinations of harvest with genotype and harvest with flowering (Table S1). Figure S1 illustrates the genotypic trends and interactions with harvests for stem N, P and K.
Only 10 samples (comprising five genotypes) exhibited detectable (>0.1%) concentrations of S. These were almost all (9/10) in spring harvested leaf material and half of these were from a single genotype (NL2).
As the majority of the harvested Miscanthus biomass is stem, we present more detailed findings for this fraction only. A closer examination of the key traits follows, and includes an analysis of the effects of flowering and senescence.

Nitrogen
Mean stem N concentrations in the spring were 0.37% w/w (minimum 0.2% w/w), which was less than half the concentrations at the summer harvest (mean 0.93% w/w, minimum 0.6% w/w) ( Table 2). The seds for the genotype x harvest interaction and predicted mean values (Tables 2 and S1) indicated that these differences were highly significant from summer to autumn in every genotype, but from autumn to the following spring were not significant in seven of the genotypes (EE1, 2 and 3, LM1 and 2, LL, NL1 and 2). The difference in (transformed) N concentrations between flowered and nonflowered stem material was highly significant in the spring, but not in the autumn ( Table 6). The difference between (back-transformed) senesced and nonsenesced stem N concentrations was 0.27%, which was also highly significant (Table 6).
In three cases (NL1 and 2, LL), N concentrations were higher in the spring compared with autumn, but these were amongst the genotypes whose spring N content was not significantly different from that in autumn (Table S1). No relationship between starting concentration of N and subsequent concentrations was identified: genotypes with the highest initial concentrations did not necessarily have the highest concentrations throughout the harvests.

Phosphorous
Mean stem P concentrations in the spring were 416 mg kg À1 (minimum 41 mg kg À1 ), which was, again, less than half the concentration in the previous summer (mean 1215 mg kg À1 , minimum 190 mg kg À1 ). Five exceptions to the reduction of P concentrations   throughout the harvests were identified (Table 2), and there were nine genotypes (ME1, 2 and 3, LM1, 2 and 3, LL, and NL1 and 2) whose spring P content was not significantly different from that in the autumn (Tables 2, 6 and S1). In one case (ME1), P concentrations were well below the average at each harvest point. The differences between P concentrations in nonflowered and flowered stems in the autumn and spring were not significant (Table 6).

Potassium
Across 16 genotypes analysed across three harvest points we found the greatest variance to be in stem K concentrations between the summer and spring harvests, with a minimum and maximum value of 290 and 17 197 mg kg À1 , respectively (EE1 in spring and NL1 in summer, respectively (Table 2). In contrast to N, where concentrations were consistently highest in the summer, for seven genotypes K concentrations were highest in the autumn (Table 2). K concentrations fell again by the following spring in all but three genotypes (LL, NL1 and NL2) which also exhibited anomalous patterns of N concentration. With the exception of the EE phenotypes, the same genotypes that did not show significantly different N content in spring compared with autumn also did not show significantly different K content (LM1 and 2, LL, NL1 and 2). The difference in K concentrations between flowered and nonflowered stems in the spring was 670 mg kg À1 (untransformed), which was highly significant (Table 6). Similarly, the difference between senesced and nonsenesced stems was 1397 mg kg À1 (untransformed), which again was highly significant (Table 6).

Moisture content
Stem moisture content fell significantly with each successive harvest in every genotype (Table 3). The lowest moisture contents in both the autumn (mean 57%, minimum 46%) and spring (mean 22%, minimum 12%) were in the earliest flowering and senescing genotypes (Table 3), and differences between (untransformed) nonflowered and flowered stem moisture contents were 6.2% in the spring and 9.3% in the autumn, which were both highly significant (P < 0.001 with 88.2 df; Table 6).

Ash
Stem ash contents also fell with successive harvests (Table 3) and although these were significant in each genotype from summer to autumn, ash contents in spring were not significantly different from autumn content in the same five genotypes (LM1 and 2, LL, NL1 and 2) that had nonsignificant N and K concentrations in spring compared to autumn. Ash contents in nonflowered stems in the autumn were lower than in flowered stems, and vice versa in the spring, although neither of these differences were significant (Table 6).

Chlorine
Stem Cl concentrations fell fivefold from summer to autumn (1.05 and 0.21% w/w, respectively, P < 0.001) and again from autumn to spring (0.21-0.14% w/w, respectively, P < 0.01). Stems that had not flowered contained more than twice (P < 0.001; Table 7) the Cl of those which had flowered (0.36 and 0.15% w/w, respectively), and stems that had not senesced contained more than three times (P < 0.001; Table 7) the Cl of those stems which had senesced (0.56 and 0.15% w/w, respectively). There were also highly significant differences between genotypes, and the most extreme differences are reported in Table 7 for illustration purposes.
The lowest concentrations of Cl in the autumn and spring were 0.14 (genotype NE) and 0.10 (genotype NE) % w/w, respectively, compared to the highest concentrations of 0.39 (genotype NL1) and 0.23 (genotype LM2) % w/w, respectively (Table 4).

Silica
Stem Si concentrations were much less variable than other compositional traits analysed and only varied significantly from autumn to spring in genotypes EE1 and NM1. Neither were they significantly affected by senescence (Tables 5b and 7). However, flowering appeared to slightly increase (P = 0.05) stem Si (Tables 5b and 7). There were also highly significant differences between genotypes: the lowest concentrations of Si in the autumn and spring were 0.13 (genotype NE) and 0.10 (genotype NM1) %, respectively, compared to the highest concentrations of 0.43 (genotype NL1) and 0.40 (genotype NL1) %, respectively (Table 4). Initial correlation analyses showed little relationship between Si content and any of the other variables. However, the pattern of Si correlations shows a shift to the right for the summer harvest data with every variable (see Fig. 2a, b). When the correlations were performed separately for summer, autumn and spring, Si presented correlations with ash of 0.741, 0.897 and 0.921, respectively, the latter being the highest correlation of the entire dataset.

HHV
The increase in mean HHV between autumn (17.5 MJ kg À1 ) and spring (17.69 MJ kg À1 ) was highly significant (P < 0.001), as were the differences between genotypes (P < 0.001). The table of fixed effects, where flowering is fitted before genotype, suggests that flowering significantly impacts HHV, but when the full model is taken into account the difference between flowered and nonflowered stems is only 0.08 (SED of 0.04978, df 14.5), which is not significant. The minimum HHV exceeded 17 MJ kg À1 in each genotype and in both autumn and spring harvests.

Discussion
Biomass quality improvement through delayed harvest is genotype-specific Despite yield losses of up to 30%, standard agronomic practice for commercial Miscanthus biomass production is to delay harvest until the spring following growth in order to allow improvement to combustion quality through moisture loss, nutrient remobilization and leaching of soluble minerals, like K and Cl, through rainfall (Jorgensen, 1997;Lewandowski & Kicherer, 1997). This practice also helps ensure crop sustainability, as nutrients will not need replacing for subsequent growth. However, the composition of M. 9 giganteus in our study (genotype LM1), which was comparable to values reported elsewhere (Clifton-Brown & Lewandowski, 2002;Lewandowski & Heinz, 2003;Monti et al., 2008;Meehan et al., 2013;Yu et al., 2014), did not significantly change between October and the following February for a number of key parameters. Indeed, our REML analysis of 16 genotypes showed that five of these (including M. 9 giganteus) did not exhibit significant reduction in N, P, K and ash from autumn (October) to winter (February), and two of these (including M. 9 giganteus) also did not show significantly reduced Cl and Si content from autumn to spring. A sixth genotype (NM1), despite showing improved quality from autumn to spring, exhibited sufficiently low N and K content in autumn to meet the ENplus wood pellet standards for those traits (European Pellet Council, 2015). All genotypes exceeded the ≥16.5 MJ kg À1 requirement for ENplus wood pellet certification, even in the autumn (Table 4), and HHV increased significantly from autumn to spring in only half the genotypes tested. Moreover, stem Si concentrations were consistent across the year, with minimums/maximums of 0.11%/0.59%, 0.13%/0.43% and 0.10%/0.40% in summer, autumn and spring, respectively (Table 4). On the other hand average leaf Si concentrations increased significantly from summer (0.46%) to autumn (0.65%), with a lesser increase again from autumn to spring (0.75%). There was a fivefold difference between the minimum (0.23%) and maximum (1.15%) concentrations of Si in the autumn. With respect to Si, therefore, there was no benefit, and even some detriment, in delaying harvest until spring. Average leaf and stem N concentrations in the autumn were 43% and 46%, respectively, of summer values, but surprisingly, from autumn to spring the overall average leaf N content increased very slightly, whilst stem N content decreased by only 13% of autumn values (Table 2), and the decrease in stem N from autumn to spring was significant in only half the genotypes tested. This lack of significant N reduction over winter was consistent with findings by Jorgensen (1997), who suggested that N must therefore be fixed in nonsoluble organic substances. However, others have reported significant N content reduction over winter months Baxter et al., 2014), which may be due to differences in N fertilizer regimes: there was no fertilizer treatment in our trial.
For N and ash content in stem material, we identified four genotypes (ME1, ME2, NM1 and NM2) that met the ENplus A2 pellet classification criteria when harvested in spring, two of which were close to reaching A1 standards (European Pellet Council, 2015). Furthermore, two of the aforementioned genotypes, NM1 and NM2, were within N concentration limits for the A2 classification during the autumn harvest in October, and only 0.1% over the ash limit. This is promising for an autumn harvest at peak yield, although the leaf fraction would need to be minimized.
Consistent with previous studies (Jorgensen, 1997;Baxter et al., 2014) leaf N concentrations were much higher than those of stem, and during the spring, they were comparable (even slightly higher) than those of stems in the summer. This illustrates the importance of leaf loss over the winter, which varies greatly between Miscanthus genotypes , and should be considered a key trait for the optimization of biomass quality because, with the sole exception of spring Cl content, mineral concentrations were overall lower in stem compared with leaf material at each harvest (Tables 2-4), consistent with previous reports Monti et al., 2008;Baxter et al., 2014).

Flowering and senescence as factors affecting Miscanthus biomass quality
The relationship between flowering and senescence in Miscanthus has not been well defined but it has been proposed that both potentially promote nutrient remobilization, and hence combustion quality and sustainability. We did not identify an unambiguous correlation (R = 0.52) between flag leaf emergence (the first visible sign of flowering) and senescence; however, the order of fitting these terms in our model influenced the level of significance for some traits (Table 5a and b). Genotypes had been selected in our study that exhibited variation in flowering time and senescence to explore the effect of these phenologies. We did not physically control flowering and senescence in this experiment. Nevertheless, for N, P, K, Cl, ash and moisture there was a clear reduction in contents and therefore improvement in quality associated with senescence, and a less pronounced but still significant improvement to N, K, Cl (for spring harvest only) and moisture content (for autumn and spring harvests) associated with flowering (Table 5a and b). However, despite these overall trends, the very late-flowering (did not flower at all in our trial) and medium-senescing genotype NM1 exhibited an exceptional phenotype for chemical combustion, with consistently low concentrations for all compositional traits, and in the case of N, P and Si, the lowest concentrations of all genotypes examined. This genotype also exhibited the highest heating value and we therefore suggest would make an excellent genotype for Miscanthus variety improvement.
We speculated that stems that had flowered and were senescing by autumn were more prone to Cl (and K) leaching and thus had improved quality by spring (Jorgensen, 1997). We also expected that leaching might be promoted in thinner-stemmed genotypes. However, genotype NM1 confounded both of these theories as it did not flower at all in our trial, but had the lowest overall mineral content, and it also had the thickest stems of all the genotypes in our trial (15.4 cm; Table S3). We also compared the progression of senescence in this medium-senescing genotype with those of early-senescing types (Fig. S2), expecting to see a prompt completion of senescence before the autumn harvest that explained its performance. However, this genotype had reached <60% loss of greenness by the October harvest and yet had lower N and K contents than some of the early-senescing genotypes (Table 2) and the lowest Cl and ash contents of all genotypes at that harvest.
In the case of LM1, LM2 and NM1, stem Cl concentrations barely declined between autumn and spring, whereas contents decreased consistently and significantly in the remaining five genotypes analysed (Table 4). Jorgensen (1997) identified reductions of about 50% in Cl content of M. 9 giganteus over winter, but 85-95% in two clones of M. sinensis, which could not, therefore, be attributed to weather patterns. Jorgensen (1997) highlights the role of frost damage in promoting leaching, and that cuticle integrity, leaf age and tissue wettability, as well as membrane malfunction, are known to affect leaching (Charley & Richards, 1983). Jorgensen (1997) also highlights the range in frost sensitivity of (i) different plants, where membrane function may be retained or lost only transiently after frost in a frost-hardy plant, and any ions diffused from the cell would be reabsorbed after thawing; and (ii) different tissues, where studies in sugar cane show that lethal temperatures range from À2.8°C for leaves of young plants, and À5.0 to À5.6°C for lower stems. M. 9 giganteus and late-ripening selections of M. sinensis were observed as apparently alive during part of the winter, even in Denmark (Jorgensen, 1997), where some tissues are not killed until severe frosts appear. In our study, the coldest temperature was À7.94°C, on 8 January. Substantial leaching may therefore not occur in some genotypes until lethal frost temperatures occur, and certainly many Miscanthus genotypes are far more frosthardy than sugar cane. In our trial, the three stay green M. sinensis genotypes (LL, NL1/2) showed much higher levels of K (as well as N, P, ash and MC) in the spring than other genotypes, and similar Cl content to M. 9 giganteus. One of these, NL2, also contained the highest concentrations of Si and S and is thus highly unsuitable for breeding, but could make a valuable parent in a cross to help dissect senescence and composition traits.
A further explanation for confounding mineral content could be due to influx and efflux of different elements. In cereal leaves, chloroplasts contain 75% of the reduced N in the cell (Peoples & Dalling, 1978). The remobilization of minerals affected by senescence has been less well researched, but studies in trees (Walter et al., 1988) demonstrated the translocation of minerals, both in and out of leaf blades, during autumn senescence. The complex interactions of senescence processes may explain why some mineral elements declined rapidly, for example N and K from summer to autumn, and others, such as Cl, were maintained at a more steady level in our study. The large translocations of nutrients no longer required in above-ground tissues, particularly with elements such as N, result in a net decrease in biomass within those tissues; as a consequence other minerals, when measured as a proportion of biomass, they may remain at relatively high levels despite active translocation occurring, because their efflux is masked by larger changes in other minerals and biomass yield.
Such complex translocations may also affect N levels. Detailed comparisons of the visual progression of senescence showed that the majority of early-senescing genotypes were above an 80% loss of greenness before the autumn harvest, and some had even completed senescence by then (Fig. S2). Therefore, the major translocation of N would be expected to be completed in these genotypes. Where N is appearing to increase slightly in leaf tissue between autumn and winter harvests is possibly the result of an efflux of other minerals from leaf tissues and a general overall decline in biomass, resulting in the proportion of N in the remaining biomass appearing to increase.

Agronomic practices in Miscanthus quality improvement
Despite the favourable N and ash contents reported above that met current wood pellet standards, the lowest Cl concentration we identified (0.1%) was approximately threefold higher than the acceptable ENplus limit of ≤0.03 for A2 classification wood pellets. Cl content in grasses is known to be higher than in wood, although Miscanthus Cl content is favourable to those of cereal straw (0.4%) and reed canary grass (0.6%) (Obernberger et al., 2006). Cl concentrations in the present study were comparable to those of Jorgensen (1997) and Meehan et al. (2013), but exceeded those of Baxter et al. (2014). However, Jorgensen (1997) found interannual differences of 0.32% Cl in M. 9 giganteus, and it is possible that Miscanthus grown at coastal locations, as in the present study, may have elevated Cl due to salt spray. Importantly however, 80-90% of Cl (as well as K) has been shown to leach in mown barley straw exposed to 100 mm precipitation (Sander, 1997), and more recently, it was found that cutting M. 9 giganteus in January, but leaving it in the field prior to collection in April, led to a fall of~0.4% in Cl content compared to that cut and harvested in April. This was considered a likely consequence of leaching due to rainfall (Meehan et al., 2013). Leaving the crop on the field postharvest could therefore vastly improve quality, especially in climates like Wales where the greatest rainfall (~200 mm) tends to be in November (e.g. as in Fig. 1).
Although ENplus is the benchmark for wood pellets, in recent years standards have been emerging for straw and grass fuel pellets (Carroll & Finnan, 2012;Miranda et al., 2015). As markets develop, greater value will be placed on quality, and meeting these standards will require management of quality throughout the supply chain from grower through to end user. This will include variety selection and agronomy over the feedstock production cycle to optimize combustion efficiency, minimize slagging and fouling and meet emissions targets. Plant breeding is thus an important aspect of achieving quality targets as an alternative, or in addition, to pretreatment processes such as pellet washing/leaching with water (Jenkins et al., 1998;Yu et al., 2014), surfactants (Banks et al., 2014), acid or other solutions (Saddawi et al., 2012). Indeed, the advantages of early and/or rapid senescence is likely to be that resources are remobilized before an early harvest to the rhizome and that winter rainfall will promote the leaching of elements such as K and Cl from the standing stems. Additional leaching could be achieved by cutting stems and leaving flat or in swaths in the field (Meehan et al., 2013). This may be particularly important in achieving potentially rigorous Cl standards and will also help reduce MC % (Meehan et al., 2013).
Quantitative trait loci (QTL) investigations aimed at identifying regions of the Miscanthus genome that impact on combustion traits have been reported (all in M. sinensis). These were conducted on a population of limited size (n = 89), and an incomplete genetic map (comprising 28 linkage groups, whilst M. sinensis has 19 chromosomes). The linkage groups were identified using RAPD markers, which have limited reproducibility (Atienza et al., 2002(Atienza et al., , 2003a. We propose the development of new mapping families using material similar to that identified as NM1 in the present study for the purposes of improving combustion traits and promoting sustainability. There also may be a need to develop mapping families for the identification of traits relating to biomass quality for varied end uses, that is combustion, pyrolysis and gasification or fermentation.

Conclusions and recommendations
• Significant variations in Miscanthus mineral content between genotypes and harvest times were found under the environmental conditions at a site near Aberystwyth, Wales, UK. These are being used to guide parental selections needed to make breeding improvements in the compositional characteristics that influence biomass quality.
• Not all quality relevant traits were significantly improved by delaying harvest until spring (February), when harvestable yields determined in plot trials are typically one-third less than in late autumn (October). Five (including M. 9 giganteus) of sixteen genotypes did not exhibit the expected significant overwinter reductions in N, P and K. In two of these genotypes (including M. 9 giganteus), there was also no significant overwinter reduction in Cl.
• Early flowering and senescence were not closely correlated with each other, but both were shown to impact mineral content, thus improving biomass quality attributes generally. Exceptionally, one lateflowering genotype that does not flower or senesce thoroughly in Aberystwyth during winter (NM1) exhibited excellent overall quality traits, with sufficiently low N and K content in autumn to meet the ENplus wood pellet standards (European Pellet Council, 2015).
• Whilst breeding should select parents with optimized quality traits, agronomic practices should also be optimized, such as avoiding overfertilization, identifying the most appropriate harvest time, and cutting plants before collection to allow maximum moisture loss and leaching of water soluble minerals.
• Few genotypes in our trial had <50% MC at the October harvest. However, Lewandowski & Heinz (2003) concluded that for CO 2 equivalents, harvesting Miscanthus in December, before winter losses, resulted in higher hectare-related CO 2 equivalent saving potentials than a February or March harvest, despite the increased energy demand for technical drying.
• Additional studies in changes to the chemical composition of Miscanthus biomass over the growing season across diverse locations and years are recommended to determine consistencies in nutrient dynamics and enable valuable modelling for varied soils and environments, and thus end uses.

Supporting Information
Additional Supporting Information may be found online in the supporting information tab for this article: Table S1. Predicted means for N, P, K, Na, ash, and MC from REML Genotype and Harvest analysis at available flowering states (flowered vs. non-flowered) for diverse Miscanthus gentoypes from trait trial in Aberystwyth (Wales, UK). Genotype LM1 is Miscanthus 9 giganteus. Table S2. Correlation matrices between nitrogen, sodium, potassium, phosphorus, chlorine, silica, moisture content, ash and higher heating value (HHV) based on flowering (a) and senescence (b), genotype and harvest from the REML analyses of stem data for diverse Miscanthus genotypes across three harvest points (summer, autumn and spring) in 2009-2010 as part of trait trial in Aberystwyth (Wales, UK). Table S3. Plant characteristics of the 16 Miscanthus genotypes (two species and their hybrids) studied, including: plant basal diameter (mm), transect count (a relative measure of stem number), height of tallest stem (mm), canopy height (mm) and dry matter single plant biomass (g). Figure S1. Genotypic trends and interactions with harvest for stem N, P, and K in diverse Miscanthus genotypes. Figure S2. Senescence progression (recorded as loss of greenness, where 10 = 100%) in early senescing genotypes, and genotype NM1, of Miscanthus grown in 2009 trait trial in Aberystwyth (Wales, UK).