• Open Access

Assessment of variability in biomass yield and quality: what is an adequate size of sampling area for miscanthus?


Correspondence: Universität Hohenheim, Institute of Crop Science, Fruwirthstraße 23, 70599 Stuttgart, Germany, tel. + 49 711 45922516, fax + 49 711 45924344, e-mail: Iris_Lewandowski@uni-hohenheim.de


To diversify energy crop production and improve its biodiversity and sustainability, there is currently a search for alternative energy crops. Many of the newly arising alternatives are perennial species such as the C4 grass miscanthus. The assessment of reliable data is a prerequisite for understanding the performance of these crops and developing corresponding management systems. However there is great uncertainty concerning research methodology for these crops. When data are collected from small plots of perennial crops, such as miscanthus or short rotation coppice plantations, a larger variability is expected than for cereals. A square meter cut, corresponding to harvest practice in cereals, is not sufficient for perennial C4 grasses and is not recommended for these species. The aim of this research was to identify an adequate size of sampling area for miscanthus to estimate the true biomass yield or quality. For this purpose, whole plots of 10- and 14-year old miscanthus stands were divided into smaller subplots. These were used to calculate variances for various sizes of simulated plots. The variances for all traits in the experiments were rather high when the sampling area was smaller than 2 m². A cutting regime of more than 5.6 m² would be advisable, but an area of 3 m² is sufficient to eliminate approximately 90% of the variances and is therefore an adequate size of sampling area.


There is an increasing demand for biobased products and bioenergy crops. In Germany, the proportion of arable land used for the production of renewable primary products for energetic or material use increased from 5.4% in 2000 to 18% in 2010 (FNR, 2011).

The bandwidth of crops used for the production of bioenergy and biobased products is very narrow. In Germany, on approximately 74% of the arable land (1.6 million ha) used for the production of renewable primary products, there is a predominance of maize (Zea mays) grown for biogas production and oilseed rape (Brassica napus) grown for biodiesel. The dominance of particular crops is also a global trend and this has led to a decrease in the acceptance of bioenergy production. There is much criticism of the predominance of maize in the landscape and problems with intensive energy crop production (e.g. high input of agrochemicals in rapeseed production for biodiesel) and its adverse environmental and biodiversity impacts call for the diversification of energy cropping systems. The search for alternative energy and industrial crops focuses on perennials e.g. Silphium perfoliatum, Rumex OK2, Sida hermaphrodita (Borowska & Styk, 2006; Conrad, 2009; Rumpler & Reichardt, 2009) for biogas production, jatropha for oil production and lignocellulose-rich plants such as miscanthus (Miscanthus × giganteus) and switchgrass (Panicum virgatum) for use as solid fuels or in microorganism-based ethanol or oil production (Dufreche et al., 2007; Tian et al., 2011).

Perennial crops are planted once in a plantation lifetime of 10–25 years, which greatly reduces the need for soil cultivation and the danger of soil erosion (Smeets et al., 2009). They are often characterized by efficient use of nutrients, particularly nitrogen, because they recycle nutrients internally (Lewandowski & Schmidt, 2006). They are also expected to improve biodiversity, sequester carbon and be environmentally more favourable than many annual crops (see e.g. Böhmel et al., 2008).

When investigating the yield and quality performance of perennial crops, their specific characteristics need to be taken into account. In contrast to cereals or annual species, perennials are often reproduced vegetatively via underground rhizomes and therefore the canopy is less homogeneous than in drilled arable crops and the sampling area may need to be expanded.

When data are collected from small plots, it is to be expected that perennial energy crops, such as miscanthus stands or coppice plantations, will show a larger variability than cereals. The number of tillers in wheat, for example, ranges from 400 to 700 m², whereas Miscanthus × giganteus may have only 30 to 60 stems per m². When data from perennials are collected from a small area, inaccuracies can occur due to greater canopy inhomogeneity and reduced numbers of plants per m². In practice, however, field trials with perennials have simply adopted the recommended sampling area for cereals, i.e. one square meter. However, the question which has so far been neglected in the scientific discourse about dedicated bioenergy crops is: Which sample area would be required to estimate the true biomass yield or quality parameters of the field or plot? In their introduction, Swallow & Wehner (1986), Storck (2011) and Zhang et al. (1994) stress the necessity of finding an adequate sample size as well as shape using uniformity trials and geostatistical methods. An adequate size of a sampling area should be as small as possible, as field studies are labour-intensive and expensive, but as large as necessary to obtain reliable data. This study presents first insights into the determination of an adequate sampling area for biomass yield and quality analysis in two long-term miscanthus stands, one with different fertilizer treatments, the other with different genotypes. For this reason the analysis was done in two consecutive stages:

The first step was an analysis of variance for each trait in each miscanthus stand to obtain (a) the variance in each trait and (b) residuals. For traits of interest for combustion purposes of the biomass or where there was high variance, the second step was to analyse the change in variance with changing plot size following the concept of Petersen (1994) combined with concepts from geostatistial approaches.

The aim of this study was to initiate a discussion on the adequate sampling size and to contribute to a standardized sampling regime in perennial grass field trials.

Material and methods

Harvest data collection from field trials

In spring 2011, harvest data were collected from two field trials, one which had been established 10 years and the other 14 years previously at the University of Hohenheim (Germany) experimental station ‘Ihinger Hof’ in the climatic region “Upper Neckar Valley”. The station is located in southwest Germany (48°27′36″N, 8°33′36″E; 460–520 m above sea level) and has an average yearly precipitation of 690 mm and an average temperature of 7.9 °C. Field Trial 1 was conducted on soils classified as Haplic Luvisols (IUSS Working Group WRB, 2007) with moderate drainage and a stone content of around 1%. The soils have a silty clay texture (approximately 40% clay) and lie over loess. Soil Ct and soil Nt were measured in spring 2004 and ranged from 0.92% to 1.07% and 0.10% to 0.11% of soil respectively. Field Trial 2 was conducted on a soil considered a Vertic Eutrudept with soil Ct and Nt contents of 1.6% and 0.11% respectively.

Field Trial 1 was established in the year 2001 as part of an extensive bioenergy crop trial. The genotype Miscanthus × giganteus was planted in rows 70 cm apart and at a distance of 71 cm within each row. However, 10 years after planting individual plants were no longer distinguishable. The trial was designed as a split plot with four replications and three different fertilizer treatments (for a description of this trial see Böhmel et al., 2008). Plot size per treatment was 160 m². Three different fertilizer (ENTEC®) levels were applied in late spring to the respective plot: 0 kg N ha−1, 40 kg N ha−1 and 80 kg N ha−1. In the year 2010, the miscanthus was additionally fertilized with 350 kg potash ha−1 (60% K2O). Plant protection measures were not needed in the vegetation period 2010/11 as the miscanthus stand was well established and dense enough to resist weed pressure.

Field Trial 2 was established in 1997. It was arranged as a complete randomized block design with 15 different genotypes, three replications and a plot size of 25 m² (for description, see Clifton-Brown & Lewandowski, 2002). The trial was fertilized uniformly with 140 kg K ha−1 (potassium sulphate), 100 kg P ha−1 (superphosphate) and 60 kg N ha−1 (nitro-chalk) annually. Manual weeding was performed only during the years of establishment. During the vegetation period 2010/11 no crop protection measures were necessary. For this study, data from only three of the genotypes were selected for analysis: Miscanthus × giganteus (M 1 Lasei 1), Miscanthus sacchariflorus (M 11 Materec 11) and Miscanthus sinensis (88–111).

Data were collected from the two field trials during the harvest in mid-March 2011. The harvest methods and quality analysis were identical for both trials. To avoid border effects, the outermost rows were not taken into account. The harvest was performed by hand using a hedge trimmer and the miscanthus was cut at a height of 5 cm. To determine the size of the sampling area required to estimate the true biomass yield or quality parameters of the plot, a specific harvesting scheme was applied. Each plot was divided into twenty 0.5 m² subplots in Field Trial 1 and ten 0.5 m² subplots in Field Trial 2 (Fig. 1). Harvesting was begun in the southwestern part of each plot and proceeded in a northwesterly direction. For the statistical analysis, the 0.5 m² subplots were added up to 1, 1.5 m² etc. to form simulated subplots in a northerly and an easterly direction (Fig. 1). The resulting simulated subplots were thus quadratic or rectangular with a final size of up to 10 m² for Field Trial 1 and 5 m² for Field Trial 2. As many simulated plots were made up in the whole plot, as possible. Depending on the simulated plot size some subplots remained unused, e.g. for building up a simulated plot of 3 m² in Field Trial 2 six subplots are used and four remain unused (Fig. 1).

Figure 1.

(a) The two repetitions of the Field Trials 1 and 2. The grey areas represent the areas used for data collection. (b) A sketch showing how the simulated plots were assembled. Two examples of 1 m² plots and one example of a 2 m² plot are shown. The white rectangles represent the harvested areas.

Plant height was measured, the number of stems counted and fresh (FM) and dry matter (DM) yield determined. The fresh material was oven-dried for 24 h at a temperature of 80 °C. The samples were then ground to a particle size of 1 mm for further quality analysis. All samples were analysed using the NIRSystem 5000 (ISI-Software, USA) and a representative number of subsamples were extracted for calibration. Nitrogen (N) and carbon (C) concentrations were determined using elemental analysis (using technology from Elementar Analysensysteme GmbH). Potassium (K), calcium (Ca) and sodium (Na) were determined using the measurement equipment “eppendorf ELEX 6361” (Eppendorf AG, Hamburg, Germany). Magnesium (Mg) was determined using the SpectrAA atomic absorption spectrometer (Agilent Technologies GmbH, Böblingen, Germany) and phosphorus (P) was measured photometrically. Ash content was analysed by incinerating the samples for 4 h in a muffle type furnace at 830 °C.

Statistical analysis

For the evaluation of an adequate simulated plot size, the best method is to use a uniformity trial design. Uniformity trials are designed to determine the amount of variation resulting purely from variation in the soil or individual plants. For this purpose, a proper experimental design is developed, but the variants in all plots are identical. The field trials described above were designed for the detection of long-term effects of genotype and management system and not as uniformity trials. To use data from these trials for the evaluation of an adequate simulated plot size, the treatment and design effects of the experiment were removed to simulate conditions analogous to a uniformity trial. In this way, residuals are obtained which mainly represent the soil and plant heterogeneity.

In the following paragraphs, the term “whole plot” is used to describe the plot area of one treatment in a replication. As mentioned above, each whole plot was divided into smaller part, which were termed “subplots”. These were harvested and analysed separately and are needed to calculate variances. The term “simulated plot” refers to a number of subplots hypothetically pieced together to simulate a plot of a specific size (Fig. 1). The smallest subplot unit was 0.5 m², from which simulated plots can easily be enlarged to sizes of 1, 2, or higher numbers of m².

For the first step of the analysis, harvest data from both field trials were subjected to an analysis of variance (anova) using the MIXED procedure of the SAS System 9.2 (SAS Institute, 2009). The model for the subplot data (y = e.g. data for fresh matter yields or quality data such as ash) conformed to:

display math(1)

where μ was the overall mean. ‘Treatment’ was either fertilizer or genotype depending on the field trial. Treatment and replication were fixed effects; the plot*replication effect, which describes the variance of the different whole plots, was a random effect. Where necessary, data were transformed or heterogeneity of variance was modelled. This analysis removes the influence of the design as well as the fertilizer treatment or the different genotypes, respectively, so that the residual error solely reflects the heterogeneity of the soil and individual plants. The residuals of the subplots were then used in the following second step for the analysis of adequate simulated plot size.

The subplots within a whole plot were assembled to form so-called “simulated plots” and their new residual, the average of the residuals of the used subplots, was estimated. The variances were estimated for the selected traits and the different rectangular simulated plot sizes (between 0.5 and either 5 or 10 m² for Field Trial 1 and 2 respectively). The rectangular simulated plots were made up in both vertical and horizontal directions. Their sizes could not be smaller than a subplot or larger than a whole plot and ranged from 0.5 and either 5 or 10 m² for Field Trial 1 and 2, respectively (see Fig. 1). The calculation of the mean of the residuals of each simulated plot size was done for as many simulated plots as possible of this size within one whole plot. For Field Trial 1, the whole plots lay side by side within each replicate. Hence the three areas of ten by two subplots were treated as if they were one area of ten by six subplots. Thus possible numbers and sizes of simulated plots were, for example, 10 plots of size 2.5 m² or 12 plots of size 2.5 m² depending on their direction. For Field Trial 2, the whole plots were distinctly separated, so for example the following sizes were possible: five of size 1.0 m² or two of size 2 m².

The selected traits were DM, ash, N and K (Fig. 3) and an analysis was carried out for the data from each field trial separately. The traits DM and K were chosen on account of their importance in estimating the energy output and combustion quality. Ash and N were chosen as representative for traits which have either a comparatively high (ash) or comparatively low (N) variance.

Variance against subplot size was plotted on a graph, an example of which is shown in Fig. 2. The graphs for each trait and field trial show a decreasing variance with increasing plot size (Fig. 3). Therefore for all selected biomass and quality traits, a hyperbola was fitted per trait and trial according to the following equation:

display math(2)
Figure 2.

Illustration of the highest measured variance, limiting value and border range for the determination of the suitable simulated plot size.

Figure 3.

Variances for the different traits (DM yield, ash content, N and K concentrations) plotted against the sampling area for Field Trial 1 (left) and Field Trial 2 (right). A monotonic hyperbola was fitted to calculate the border range and adequate sampling area.

To determine a suitable size of simulated plot for each trait, a 95% reduction in the variance compared to the variance of a subplot (0.5 m²) was set as the aim. Our approach is akin to the determination of the practical range in geostatistics (Isaaks & Srivastava, 1989; Schabenberger & Pierce, 2002). Hundred per cent was set as the distance between the variance found within the simulated plot of 0.5 m², which always had the highest value, and the variance value of the asymptote of the fitted hyperbola (Fig. 2). After a reduction in variance by 95%, it does not change substantially with further increase in simulated plot size.


Biomass yield and quality

The results for the biomass yield, the chemical composition and the ash content for the entire harvested area from Field Trial 1 are shown in Table 1; the results from Field Trial 2 are shown in Table 2. Field Trial 1 had unfertilized plots (N0) and treated plots, which received 40 (N40) and 80 kg N ha−1 treatment (N80). For fresh matter (FM), dry matter (DM) and plant height (PH), N0 was significantly different from the other two fertilizing regimes. Values for FM were between 19 and 25 t ha−1 and DM between 13 and 18 t ha−1. The nutrients C, N, P and Ca showed significant differences between all three fertilizing regimes, whereas for K there were significant differences only between N0 and N80. For ash, the difference between N40 and N80 was significant, but the difference between N0 and both N40 and N80 was not significant. The ash content was between 2.97% and 3.13% of dry matter. No differences were found with Na, Mg, stem number and DM content (%).

Table 1. Biomass yield, mineral concentrations and ash content for the three fertilizer treatments in Field Trial 1. Results shown are from sample collections of 10 m². Values with the same superscript are not significantly different at α = 0.05 according to a t-test
 Treatment: Fertilizer
0 kg N ha−140 kg N ha−180 kg N ha−1
  1. a

    for statistical analysis, the data needed to be log transformed. Values shown here are on the original scale.

  2. b

    Due to transformation, the variance shown is a mean variance.

FM (t ha−1)a19a24b25b1.57b
DM (t ha−1)13a17b18b34.7
DM content (%)66a69a68a31.3
Stem number (m²)31a32a31a84.1
Plant height (cm)214a245b255b192
C (% of DM)47.9a47.7b47.5c0.14
N (% of DM)0.30a0.36b0.48c0.004
K (% of DM)0.49a0.53ab0.59b0.008
P (% of DM)0.08a0.09b0.1c0.0001
Na (% of DM)0.004a0.004a0.005a9 × 10−7
Ca (% of DM)0.19a0.15b0.13c0.001
Mg (% of DM)0.05a0.04a0.04a0.00003
Ash content (% of DM)3.05ab2.97a3.13b0.178
Table 2. Biomass yield, mineral concentrations and ash content for the three genotypes in Field Trial 2. Results shown are from sample collections from 5 m². Values with the same superscript are not significantly different at α = 0.05 according to a t-test
 Treatment: Genotype
Miscanthus × giganteus Miscanthus sacchariflorus Miscanthus sinensis
  1. a

    for statistical analysis, the data needed to be log transformed. Values shown here are on the original scale.

  2. b

    anova was performed with heterogeneous variances for each genotype. For this reason the variance shown is a mean variance. Heterogeneity was extreme: varmin: 17.9, varmax: 1028.8

  3. c

    anova was performed with heterogeneous variances for each genotype. For this reason the variance shown here is a mean variance.

FM (t ha−1)16a16a11a9.3
DM (t ha−1)13.2a13.7a8.3b6.0
Stem number (m²)a27a27a110b196.2b
Plant height (cm)214a234a160b196.3
C (% of DM)47.4a47.5a48.3b0.26c
N (% of DM)0.35a0.40a0.45a0.003
K (% of DM)0.54a0.54a0.33b0.015
P (% of DM)0.08a0.09b0.09b0.03c
Na (% of DM)0.004a0.004a0.004a1.2 × 10−7
Ca (% of DM)0.22a0.14b0.22a0.012
Mg (% of DM)0.04a0.04a0.05a5.3 × 10−5
Ash content (% of DM)2.9a2.4b3.0a0.16

It is well known that besides fertilization regime and management system, the genotypic effect has a strong impact on yield performance and chemical composition of miscanthus. Hence, in Field Trial 2, three different genotypes were compared: Miscanthus × giganteus (M. gig.), Miscanthus sacchariflorus (M. sacch.) and Miscanthus sinensis (M. sin.). In our analysis no significant differences between the three genotypes were found for FM (11–16 t ha−1), N, Na and Mg. For DM, stem number, plant height, C and K there were no significant differences between M. gig. and M. sacch., but both were significantly different from M. sin. For P, M. sacch. and M. sin. showed significant differences from M. gig. For Ca and ash content, M. gig. and M. sin. differed significantly from M. sacch.

Suitable size of simulated plot

The evaluation of an adequate sampling area can be deduced from the graphs either by calculation (Tables 3 and 4) or by visual deduction (Fig. 3). The adjusted monotonic function (hyperbola) for the variances plotted against the sampling area showed a steep decline within a sampling area of 0.5–2 m² for all analysed traits. According to the calculated values or the border range (95%), the adequate sampling area varied between 3.1 and 7.2 m². On average, a sampling area of 5.6 m² would be necessary to eliminate 95% of the variances of the whole plot (mean of values in Tables 3 and 4 without value for biomass). For the parameter “biomass yield” a plot size of more than 50 m² would be needed to eliminate 95% of the variance. This plot size is not suitable for field trialling.

Table 3. Elimination of variance for different sampling areas for Field Trial 1 (fertilizer treatment) using the variance of the 0.5 m² plot as 100%
 Elimination of variance (%)
Sampling areaBiomass yieldN concentrationAsh contentK concentration
1 m²57.758.468.158.9
2 m²76.879.784.680.8
3 m²83.186.890.088.0
4 m²86.390.392.891.7
5 m²88.292.494.493.8
6 m²89.593.995.595.3
7 m²90.494.996.396.3
8 m²91.095.696.997.1
9 m²91.696.297.497.7
10 m²92.096.797.798.2
Adequate sampling area at border range (95%)>50 m²7.2 m²5.5 m²5.8 m²
Table 4. Elimination of variance for different sampling areas for Field Trial 2 (genotypes) using the variance of the 0.5 m² plot as 100%
 Elimination of variances (%)
Sampling areaBiomass yieldN concentrationAsh contentK concentration
1 m²73.473.072.677.6
2 m²83.386.186.190.5
3 m²86.790.590.694.7
4 m²88.392.792.896.9
5 m²89.394.094.198.2
6 m²90.094.895.099.0
7 m²90.595.595.799.6
8 m²90.895.996.2100
9 m²91.196.396.5100
10 m²91.396.696.8100
Adequate sampling area at border range (95%)≫50 m²6.2 m²6.0 m²3.1 m²

Data analysis of the harvest from Field Trial 1 (20*0.5 m² subplots) showed a mean coefficient of determination (R²) of 0.75 between the calculated variances and the adjusted hyperbola and a mean required sampling area of 6.2 m². Data analysis of the harvest from Field Trial 2 (10*0.5 m² subplots) showed a mean coefficient of determination (R²) of 0.28 and a mean adequate sampling area of 4.8 m². Field trial 2 had smaller and separated whole plots within a replicate, so the shapes and sizes were limited. In addition, from this experiment different genotypes were used. Even though we tried to correct for this via anova it seems to be most probable that the differences between genotypes were responsible for the large variation in Field Trial 2.

The range of variances differed both between traits and between field trials (see Fig. 3).

A sampling area of 5.6 m² or more would be difficult to handle on an experimental scale. Figure 2 shows that, due to the steepness of the slope, the reduction in variance declines considerably as the sampling area increases. For this reason a sampling area of 2–3 m² seems to be advisable. Considering the data from both field trials, a 2 m² sampling area gave a mean elimination of variance of 83%; a 3 m² sampling area gave 89% (see Tables 3 and 4).


Ramsey (2004) stated that sampling procedure is a very important part of the process of determining the concentration of any chemical component of the environment. Besides sampling accuracy, sampling method and – in our case – sampling area may have a great influence on the results. It is impossible to determine the true value of the analyte concentration due to the uncertainty in the chemical analysis as well as the sampling procedures (Boon et al., 2007). Nevertheless, the total variance of a measurement can be reduced by optimizing the parameters, which determine the variance. According to Ramsey et al. (1992) and Boon et al. (2007) the total variance of a measurement can be defined as the total variance of a measurement as the sum of the geochemical variance, the sampling variance and the analytical variance. To be able to apply this formula to the analysis of samples from contaminated land, they proposed adapting this formula by replacing ‘geochemical variance’ with ‘soil variance’. This adapted formula can also be applied to biomass analysis. By optimizing the sampling procedures and standardizing the analyses, the results can approximate the true value. Optimizing or standardizing the sampling area for biomass analysis of dedicated perennial energy crops such as miscanthus is therefore important to obtain reliable data.

Official figures for sampling area have only been published for plant breeding programmes and plant protection trials. For plant breeding, the BSA (Bundessortenamt, 2000) regulates the trial procedure in detail with particular regard to sampling area. Core sections for most cereals should be 10 m², for beets and leguminous plants it can range between 12 and 15 m². For plant protection trials in cereals the EPPO standards (Biologische Bundesanstalt für Land- und Forstwirtschaft Bundesrepublik Deutschland (BBA), 2000/2001) stipulate a minimum net sampling area of 20 m2. For most other crops they specify at least 40 m2. In arable crops for food production, the current practice is to sample data from 1 m² cuts. However, this cannot necessarily be applied to dedicated perennial energy crops and innovative management systems such as precision farming (Leithold & Traphan, 2006) or grassland/grazing production scenarios (Schröpel, 2008). For these, new guidelines need to be created.

Up until now various sampling areas for perennial grasses have been reported in the literature, for example: “core section harvest in 5*5 m² plots” (Clifton-Brown & Lewandowski, 2002), “tillers were counted in a 1.5 m² area” (Hodgson et al., 2010), “inner 6*6 m² harvest” (Christian et al., 2008), “sampling area of 9 m²” (Monti et al., 2008), “a representative sample of 5–30 stems” (Burner et al., 2009) and “2 m² cuts within central plot area” (Lewandowski et al., 2003). Most of these are smaller than the sampling area size of 5.6 m2 determined in our study as on average necessary to reduce variance by 95%. However, these harvests were done to determine biomass yield. For quality, yield parameter or mineral composition analyses, subsamples were used. The underlying sampling areas are rarely mentioned. It can be concluded that the square meter cut applied in cereal harvest practice is not sufficient for perennial C4 grasses such as miscanthus. A sampling area larger than 5.6 m² would be advisable, but an area of 3 m² would be sufficient to eliminate approximately 90% of the variances.

In the year this study was conducted the miscanthus stands were well established and so the biomass yield data can be considered representative for the economic lifetime of miscanthus. Reliable data are necessary for further recommendations with respect to crop management over time. Therefore, consistent data analysis and sampling methods are required to make the best use of the plant's potential for bioenergy application. The specification of a sampling area of 2–3 m² for miscanthus could be a first step towards standardized and statistically confirmed data evaluation.

The analysis presented here gives a first insight into the required size of sampling area for crops under research. It is obvious from the R² of Field Trail 2 that further research is necessary regarding size and shape of plots, especially for different genotypes. To determine more precise values, it would be necessary to perform uniformity trials. These could for example be integrated into commercially used fields.


The authors thank Laura Mack, Nora Zöhrens, Sören Wansel and Stefan Nölke for their assignment, their diligence and their assistance with field, laboratory and preparatory work. The manuscript was edited by Nicole Gaudet.