Yield and spatial supply of bioenergy poplar and willow short-rotation coppice in the UK


Author for correspondence:
Gail Taylor
Tel: +44 (0)2380 592335
Fax: +44 (0)2380 594459
Email: g.taylor@soton.ac.uk


  • • Limited information on likely supply and spatial yield of bioenergy crops exists for the UK. Here, productivities are reported of poplar (Populus spp.) and willow (Salix spp.) grown as short-rotation coppice (SRC), using data from a large 49-site yield trial network.
  • • A partial least-squares regression technique was used to upscale actual field trial observations across England and Wales. Spatial productivity was then assessed under different land-use scenarios.
  • • Mean modelled yields ranged between 4.9 and 10.7 oven-dry tonnes (odt) ha−1 yr−1. Yields were generally higher in willow than in poplar, reflecting the susceptibility of older poplar genotypes to rust and their tendency for single stem dominance. Replacing 10% of arable land, 20% of improved grassland and 100% of set-aside grassland in England and Wales with the three most productive genotypes would yield 13 Modt of biomass annually (supplying 7% of UK electricity production or 48% of UK combined heat and power (CHP) production).
  • • Results show existing SRC genotypes have the immediate potential to be an important component of a mixed portfolio of renewables and that, in future, as new and improved genotypes become available, higher yields could extend this potential further.


Bioenergy from biomass has been identified as having significant potential to contribute to reductions in greenhouse gas emissions and to the maintenance of a secure and sustainable energy supply, within the UK (Skea, 2006; British Forestry Commission, 2007), Europe (EC, 2005) and more widely (Berndes et al., 2003; Sims, 2007). The term biomass can refer to any solid or nonsolid biological energy source. More specifically short-rotation coppice (SRC) describes any high-yielding woody species managed in a coppice system. Typically these crops are harvested on a 3–5 yr rotation and remain viable for 15–30 yr.

Dedicated SRC energy crops, such as poplar (Populus spp.) and willow (Salix spp.), are grown commercially for heat and power generation as a consequence of their rapid growth rate and favourable energy ratio. Provided local markets exist, SRC offers growers the chance to diversify into nonfood crops and, when planted in place of conventional arable agriculture, has secondary benefits including enhanced biological diversity (Rowe et al., 2008). However, the main importance of such crops is their intrinsic value as a renewable energy resource. Greenhouse gas emissions are abated as a consequence of reduced fossil fuel inputs and increased carbon sequestration, when compared with traditional crop systems (Lemus & Lal, 2005; Galbraith et al., 2006; Sims et al., 2006). Whilst species vary, each oven-dry tonne (odt) of energy crop converted to electricity displaces approx. 0.44 toe (tonnes of oil equivalent) (Bauen, 2001; Cannell, 2003; St Clair et al., 2007).

Less favourable energy balances are often achievable when current food crops are used in the production of first-generation liquid biofuels, such as bioethanol and biodiesel (Hill et al., 2005). In the future, SRC crops may also be grown as feedstock for second-generation liquid biofuels (Houghton, 2006). Rather than simply extracting sugars by fermentation (i.e. first generation), second-generation technology uses enzymes to separate the plant lignin and cellulose, which can then be fermented as with first-generation crops. This allows biofuel to be produced from any plant material, therefore reducing both the conflict between food and fuel and also the greenhouse gas emissions of the crop. CO2 is reduced by c. 90% compared with just 75% for first-generation biofuels, both relative to fossil petroleum (Edwards et al., 2007).

In addition, energy crops suffer little intermittency of supply, being storable for use during periods when other renewable energy sources may have reduced output. However, little is known of how these crops will perform in the UK, particularly at a spatial scale, and yield data are central to the development of the future of this industry.

Biomass currently accounts for 82% of the total UK renewables market, which itself supplies almost 5% of total energy production (MacLeay et al., 2007). However, more than 70% of this biomass is derived from municipal waste or forest residues rather than dedicated energy crops (MacLeay et al., 2007). Between 2000 and 2006, only 4600 ha of SRC and 5400 ha of the perennial bioenergy grass Miscanthus¥ giganteus were established (NFFCC, 2006; Defra, 2006a). Planting is increasing at a rapid rate with applications to the Energy Crop Scheme up 236% in 2007 from the previous year (Defra, 2006a), potentially providing an additional 10 700 ha of dedicated energy crops.

To meet bioenergy targets, between 0.7 and 5.4% of the UK's 18.5 Mha of agricultural land (Defra, 2006b) may need to be committed to energy crops. It is anticipated that 1500 MW of new electrical capacity may come from energy crop and forestry residue combustion by 2010, potentially requiring the planting of 125 000 ha of dedicated energy crops (Britt & Garstang, 2002). Additionally, estimates suggest that 740 000 ha of UK-produced arable crops would be required to meet half of the Renewable Transport Fuel Obligation target of delivering 5% (by volume) of liquid transport fuel from biofuels by 2010 (Defra, 2006a). Current UK government strategy is to encourage the conversion of set-aside land to energy crop production (Defra, 2006a). Approx. 600 000 ha (2005) of set-aside land is available in the UK (Defra, 2006b), suggesting the conversion of set-aside may be an effective short-term strategy to meeting bioenergy targets.

In the longer term (2010 and beyond), if the UK is to meet its ambitions, it is likely arable land will also be converted to bioenergy crops. This is a logical assumption, particularly in light of the reform to the Common Agricultural Policy, which may result in a reduction of cereal production and make more land available for alternative uses (Defra, 2007). The Biomass Task Force concluded that by 2020, up to 1 Mha of agricultural land could be available for dedicated energy crops (Gill et al., 2005). However, the Renewables Innovation Review estimated that by 2020 only 350 000 ha may realistically be available for energy crop production (BERR & Carbon Trust, 2004). On a wider scale, it has been suggested that 20–30% of all agricultural land in Europe could be used for energy crop production by 2030 (EEA, 2006), potentially producing 1.7 PW h−1 of energy, equivalent to 15% of EU25 primary energy demand.

Much of the uncertainty on future land requirements arises because of limited information on likely supply and spatial yield data of these crops and how this will evolve in future. To address issues of uncertainty over potential uptake, it is fundamentally important to provide clear and concise information regarding the spatial characteristics of biomass productivity and supply.

The first objective of this paper was to report the findings of a large UK study – a 49-site yield trial for SRC poplar and willow. The second objective was to use the data recorded in the field to spatially model productivity and energy production potential on a national scale, by developing a geographic information system (GIS)-embedded empirical technique. This approach allows complex output to be presented in an accessible and intuitive manner, providing a useful tool for all who are interested in bioenergy, including growers and policymakers in both the UK and the EU. This paper also documents an approach of wide significance to developing spatial models on biomass supply.

Materials and Methods

Field trials network

The SRC field trials network was established in 1995. The trials were the largest of their kind and were designed to provide an extensive and unique database for measuring productivity of SRC poplar and willow. Each trial was grown for two rotations, each of 3 yr, concluding in 2002.

Sixteen genotypes of each species from contrasting parentages were selected (Table 1), in order to assess genetic diversity in relation to yield. These represented commercially important and interspecific hybrids, which were high-yielding and tolerant to pests and disease. New genotypes that were not commercially available at the time of planting were also included to measure future yield potential (Table 1).

Table 1.  Name, sex, parentage, provenance, mean annual yield (oven-dry tonnes (odt) ha−1 yr−1) and rust score for the first and second rotation of all field trial genotypes
GenusGenotypeSexParentageProvenanceFirst rotationSecond rotation
Mean yieldMean rust scoreMean yieldMean rust score
  • *

    ‘Extensively’ grown genotypes;

  • †new genotypes (at time of planting).

  • Standard deviations are given in parentheses. Values with the same letter are not statistically different (P > 0.05) using one-way ANOVA (‘extensive’ trials only).

Populus71009/1FP. deltoides × P. trichocarpaUSA2.94 (1.98)1.89 (1.80)1.97 (1.84)3.37 (1.90)
Populus71009/2FP. deltoides × P. trichocarpaUSA4.95 (3.45)1.67 (1.86)4.15 (3.86)3.03 (2.04)
Populus71015/1FP. deltoides × P. trichocarpaUSA5.80 (2.67)1.61 (1.92)4.92 (3.39)2.96 (2.09)
PopulusBalsam SpireFP. trichocarpa × P. balsamiferaUSA7.24 (3.70)1.63 (1.73)7.03 (3.50)2.06 (1.72)
PopulusBeaupré*FP. trichocarpa × P. deltoidesUSA7.34 (2.33) b c2.00 (1.91)4.87 (2.43) a2.87 (1.90)
PopulusBoelareFP. trichocarpa × P. deltoidesUSA6.23 (3.02)1.98 (1.93)4.20 (3.57)2.93 (2.00)
PopulusColumbia RiverMP. trichocarpaUSA6.71 (3.33)1.49 (1.55)6.62 (3.04)1.77 (1.39)
PopulusFritzi PauleyFP. trichocarpaUSA8.59 (2.11)0.72 (0.97)8.24 (2.95)1.37 (1.35)
PopulusGaverMP. deltoides × P. nigraUSA × Belgium6.58 (2.72)0.92 (1.04)5.58 (3.18)1.73 (1.58)
PopulusGhoy*FP. deltoides × P. nigraUSA/Canada × Belgium6.45 (2.47) a1.24 (1.47)5.77 (2.46) a b2.04 (1.75)
PopulusGibecqMP. deltoides × P. nigraUSA × Belgium5.70 (2.39)1.21 (1.27)4.73 (3.28)2.16 (1.71)
PopulusHazendansFP. trichocarpa × P. deltoidesUSA7.23 (2.29)0.34 (1.05)7.56 (2.94)0.76 (1.55)
PopulusHoogvorstFP. trichocarpa × P. deltoidesUSA8.84 (2.42)0.39 (1.19)8.12 (3.81)0.62 (1.46)
PopulusRaspaljeFP. trichocarpa × P. deltoidesUSA6.69 (2.76)1.69 (2.02)4.66 (3.25)2.93 (1.91)
PopulusTrichobel*MP. trichocarpaUSA9.08 (2.67) d0.75 (0.92)9.59 (2.78) d1.11 (1.18)
PopulusUnalMP. trichocarpa × P. deltoidesUSA7.55 (3.57)1.67 (1.87)5.25 (3.91)2.61 (1.96)
SalixBebbianaMS. sitchensisUSA8.08 (2.99)0.00 (0.01)12.16 (4.38)0.00 (0.00)
SalixBjornMS. viminalis × S. schweriniiSweden7.59 (3.68)0.05 (0.16)11.21 (4.89)0.13 (0.44)
SalixDasycladosFS. caprea × S. cinerea × S. viminalisSweden7.17 (2.01)0.33 (0.78)8.21 (2.57)0.48 (0.90)
SalixDelamereFS. aurea × S. cinerea × S. viminalisEngland8.31 (1.10)0.33 (0.66)10.00 (3.13)0.61 (0.85)
SalixGermany*FS. burjaticaNorthern Ireland7.14 (2.94) a b1.55 (1.70)7.46 (4.00) c2.51 (1.75)
SalixJorrMS. viminalis × S. viminalisSweden10.50 (2.92)0.62 (0.90)11.01 (3.16)0.92 (1.38)
SalixJorunn*FS. viminalis × S. viminalisSweden9.09 (3.01) d0.31 (0.60)9.15 (2.70) d0.33 (0.74)
SalixOrmMS. viminalis × S. viminalisEngland8.33 (2.60)0.72 (0.91)8.47 (2.87)1.06 (1.32)
SalixQ83*FS. triandra × S. viminalisNorthern Ireland8.03 (3.23) c1.59 (1.56)10.71 (3.74) e1.99 (1.64)
SalixSpaethiiFS. spaethii − 7.30 (1.78)0.92 (1.24)9.44 (3.27)1.67 (1.73)
SalixST/2481/55FS. triandra × S. viminalisNorthern Ireland6.72 (2.39)1.40 (1.54)8.96 (3.29)1.95 (1.60)
SalixStott10FS. burjatica × S. viminalisEngland10.35 (3.57)1.25 (1.51)9.12 (4.30)1.69 (1.87)
SalixStott11FS. burjatica × S. viminalisEngland10.01 (3.49)1.11 (1.32)10.16 (4.51)1.56 (1.75)
SalixToraFS. viminalis × S. schweriniiSweden9.31 (3.52)0.02 (0.07)13.34 (4.43)0.05 (0.26)
SalixUlvMS. viminalis × S. viminalisEngland10.12 (3.91)0.30 (0.55)10.86 (2.77)0.62 (1.17)
SalixV789 − S. viminalis × S. capreaFinland4.12 (1.64)0.68 (1.35)5.07 (1.72)0.86 (1.51)

Sites were selected based on ecological land classification studies (Pyatt, 1995) and distributed across a physically diverse range of climatic zones and soil types, to assess environmental yield limitations. Land above 250 m was rejected, on the understanding that farmers would be most attracted to growing the crop in lowland areas.

Forty-nine sites (Fig. 1) were planted with the six most promising and widely grown genotypes (‘extensive’ trials), three of each species. Of these sites seven were also planted with an additional 13 poplar and 13 willow genotypes (‘intensive’ trials). Sites were cleared using a glyphosate herbicide spray at 5 l ha−1 and the soil was broken up to reduce plough pan and compaction. Weeds were further controlled by residual herbicide application as necessary.

Figure 1.

Location of the UK short-rotation coppice (SRC) field trials network sites. Phase 1 (established in 1995; squares) and phase 2 (established in 1996; circles) intensive (closed) and extensive (open) trial sites (adapted from Armstrong, 1997).

Hardwood cuttings with a length of 0.25 m were planted in April 1995 (phase 1) and April 1996 (phase 2) in a ‘double row’ system as used in commercial SRC plantations, essential for mechanical harvesting. Alternating inter-row distances were 0.75 and 1.5 m, with a within-row spacing of 0.9 m, to yield a planting density of 10 000 cuttings ha−1. A randomized block design with 16 genotypes × three replicated plots was used according to protocols suggested by the British Forestry Commission (Armstrong, 1997). Individual monoclonal plots were 9 × 11.5 m in size, containing 10 north–south oriented rows of 10 cuttings each. Fencing was also constructed around the sites to protect shoots from predation by deer and rabbits. All plants were cut back the year after planting to a height of 0.05 m to create a coppice system. Measurements were taken on the 36 plants (i.e. 6 × 6) in the centre of each plot and for each genotype. No nutrients were added to the soil during the course of the trial.

A detailed soil survey consisting of between six and 30 auger borings and at least one soil pit, sunk to a depth of 0.8–1.2 m (according to the soil texture), was undertaken at each site. Differences in the number of measurements were related to the soil variability and size of the site. Soil series (Avery, 1980) were recorded in-field, with samples further analysed in the laboratory for texture (Gee & Bauder, 1986) and pH (BSi, 1995). From this, available water content of the soil was derived, based on soil texture (Hall et al., 1977). Measurements of slope and elevation were recorded using a clinometer (Suunto PM-5, Suunto Oy, Finland).

Climatic variables (e.g. rainfall, dry bulb air temperature and minimum/maximum air temperatures) were measured using basic weather sensors (Holtech Associates, Harwood-in-Teesdale, UK) and more comprehensive automatic weather stations (ELE International Ltd, Hemel Hempstead, UK) at the extensive and intensive trial sites, respectively. Measurements for each variable were recorded at 1 min intervals to provide hourly means.

For standing biomass estimation, the diameter at 1 m above the ground was measured using digital callipers (Masser, Savcor Group Ltd Oy, Finland) on all live shoots per plant at the end of each growing season from 1996 to 2002 (with a total above-ground harvest every 3 yr). Concomitantly, three shoots per plot were harvested from the assessment plot, spanning a representative range of the diameter distribution, to obtain a minimum of 54 shoots per genotype and per site over the 6 yr of the trials. Harvested shoots (including branches) were oven-dried at 95°C and site-specific allometric relationships between shoot diameter and shoot dry mass were computed for each genotype to estimate standing biomass at plot level. A cube root transformation was then applied to the data. Subsequently, a potential yield (odt ha−1 yr−1) for each genotype was upscaled and computed over all plots and assessed years. This approach gave residual mean square errors of 0.030 (× 0.258 (cube root transformation)) for willow and 0.052 (× 0.032) for poplar.

The incidence of rust (Melampsora spp.) on the leaf was recorded by selecting one branch from each plant within the three central rows of the assessment plot. Three leaves were randomly selected from the top, middle and bottom third of the branch. This gave a total of 54 leaves per genotype for each plot and for each year of the trials. Each leaf was then visually assessed using a 0–5 scoring system, as follows: class 0, no incidence (0–5% leaf area lost); class 1, very light incidence (5–10% leaf area lost); class 2, light incidence (10–20% leaf area lost); class 3, moderate incidence (20–40% leaf area lost); class 4, severe incidence (40–65% leaf area lost); class 5, very severe incidence (65–100% leaf area lost).

Empirical model analysis

Selected measurements from the field trials network, which were spatially available as high-resolution, gridded national datasets for upscaling purposes were built into a relational database. Data from almost 150 plots were then used for each extensively grown genotype (intensively grown genotypes were not assessed because of lack of replicates). Various variables were considered and used in earlier versions of the models, including soil chemical composition, diurnal temperature range and seasonal climatic variations. The variables used to construct the final empirical model were those that best fitted the dataset, as follows:

  • • Soil – available water (mm), pH (BSi, 1995), clay, silt and sand content (%) and soil series (Avery classification, expanded into sets of ‘dummy’ Boolean values (i.e. 1/0) to overcome nonlinear distribution)
  • • Topographical – elevation (m) and slope (°)
  • • Climatic – annual and monthly mean precipitation (mm), summer maximum daily and monthly mean dry-bulb temperature (°C), summer growing day degrees and days of ground frost.

Plot data for each genotype were statistically modelled using partial least-squares (PLS) regression (Simca-P version 11.5, Umetrics, Umeå, Sweden). PLS is similar to principal-components analysis but additionally allows the user to define an observation which variables are used to predict (i.e. yield). Running the PLS algorithm creates a series of principal components. Only statistically significant components contributing to a net increase in the predictive ability of the model were used in the upscaling process (i.e. cumulative Q2 > 0.4 and < 0.1 from r2 score, with P < 0.05 for each component). The relative importance of a variable to yield was given by a variable importance plot (VIP), with the positive or negative correlation of that variable to yield described by a coefficient loading score. Outliers derived from normal probability analysis of component residuals were removed (if outlier P > 0.05), but accounted for no greater than 2% of the total data.

GIS data for England and Wales

Geographic information system software (ArcMap version 9.2, ESRI, Aylesbury, UK) was used to upscale and visualize the outputs of the empirical model across England and Wales. Scotland and Northern Ireland were excluded from this element of the study as the GIS data were unavailable, but corresponding suitability maps for Scotland have been published (Andersen et al., 2005).

The GIS datasets used for the upscaling process were as follows:

  • • 1 : 200 000 scale vector boundary data maps of Great Britain (Great Britain boundary data, Collins Bartholomew, UK), used to define urban areas and water features.
  • • 1 : 250 000 scale map of soil series in England and Wales (NATMAPvector, Cranfield University, UK). Soil texture, water content and pH were provided for each series.
  • • 1 : 50 000 scale digital terrain model (Land-Form Panorama DTM, Ordnance Survey, UK). From this model, slope was interpreted using the slope extrapolation tool in ArcMap.
  • • 5 km2 resolution climatic data, providing monthly mean temperatures and precipitation averaged from 1990 to 2000 (Perry & Hollis, 2005). More biologically meaningful data from the same dataset were also utilized relating to summer mean daily maximum temperature, summer growing day degrees and days of ground frost.

In order to upscale field measurements to a national scale, each raster cell of the existing spatial dataset was assigned a new value according to the model output scores for that variable, using the weighted overlay tool in ArcMap. Resultant layers were merged together by means of a weighted linear combination of land indices, to create a series of geocoded productivity maps of crop yield, with a resolution of 1 km2.

A 25 m2 resolution land classification map (LCM2000, Centre of Ecology and Hydrology, UK) was used to extrapolate modelled SRC yield totals under five different land-use scenarios (Table 2). These scenarios were intended to provide a better understanding of the complex interaction between yield potential and land use in the UK. Using the modelled data, it was also possible to take a case study example at a smaller scale in order to assess the practical deployment of bioenergy crops. Drax, the UK's largest power station, currently co-fires coal with small quantities of biomass. By 2009, 10% of its electricity will be generated from biomass (Drax, 2006), making it an ideal benchmark case. Using an overlaid combination of geocoded yield and land classification data, it was possible to extract yield totals under different land uses within a 25 km radius of the power station (based on the stipulation of the Energy Crop Scheme that no feedstock should be sourced more than 25 km from the power station it supplies (Defra, 2007)).

Table 2.  Examples of computed yield totals (Modt yr−1), mean yields (oven-dry tonnes (odt) ha−1 yr−1) and energy values (TW h−1) for poplar (Populus) genotype Trichobel and willow (Salix) genotypes Jorunn and Q83, assuming a 100% conversion of land under five different land-use scenarios across England and Wales
(a) Yield from 100% conversion of cereals (2.1 Mha)18.318.617.9
 Mean yield9.49.59.1
Electrical energy value37.638.236.8
Combined heat and power energy value66.167.264.6
(b) Yield from 100% conversion of horticulture/non cereal/unknown agriculture (3.0 Mha)25.726.325.2
 Mean yield9.29.49.1
Electrical energy value52.854.151.8
Combined heat and power energy value92.895.091.0
(c) Yield from 100% conversion of nonannual crops (0.1 Mha)
 Mean yield9.69.49.2
Electrical energy value1.21.21.1
Combined heat and power energy value2.22.22.1
(d) Yield from 100% conversion of improved grasslands (including set-aside grassland not listed under the set-aside grassland subcategory) (4.2 Mha)32.831.234.1
 Mean yield9.59.09.9
Electrical energy value67.464.170.1
Combined heat and power energy value118.4112.7123.1
(e) Yield from 100% conversion of Set-aside grasslands (0.2 Mha)
 Mean yield9.49.68.9
Electrical energy value3.23.33.1
Combined heat and power energy value5.75.85.4

Rather than supplementing the co-firing market, in the future it will also be necessary to build additional dedicated biomass power stations to meet UK energy demand. From the geocoded maps of the three most productive genotypes, we can also analyse hot spots to identify the most suitable locations for the deployment of new bioenergy power stations. Hot spot analysis, using Getis-Ord Gi* statistical rendering, was performed on the data in ArcMap and took points within a 25 km radius around each grid square.


Observed genotypic variability in yield

Field trial results shown that observed SRC yield varied significantly between genotype and rotation (Table 1). The highest yields were recorded in willow over the two rotations, with the 16 genotypes averaging 9.0 odt ha−1 yr−1 compared with 6.3 odt ha−1 yr−1 for the poplar genotypes. The highest-yielding parental line was the Swedish S. vimanlis × S. schwerinii, which displayed consistently high yields over both rotations and a high resistance to rust. This parent line included the highest-yielding single genotype, Tora, with an average yield across both rotations of 11.3 odt ha−1 yr−1. The lowest-yielding parental line was P. deltoides × P. trichocarpa, which also contained the lowest-yielding single genotype 71009/1, producing just 2.5 odt ha−1 yr−1 over the two rotations of the study (reaching 2.0 odt ha−1 yr−1 in its second rotation, the lowest of any in the field trials). Root system maturation in the developing plant may be expected to give rise to increasing yields in successive rotations, which was the case with willow but not with the poplar genotypes grown in these trials. All but two of the poplar genotypes had reduced yields in the second rotation; conversely, all but one of the willow genotypes had improved measured yields in the second rotation (Table 1). It is important to note that yields would likely decline if these genotypes were grown at true commercial scale as a result of less stringent and scientific management methods (Hansen, 1991).

The incidence of infection by rust (Melampsora spp.) varied between genotypes and rotation (Table 1). Heavy rust infection preceded crop death on several plots, including plots of previously resistant genotypes Hoogvorst and Hazendans. In the trials, 5–30% of the poplar genotypes (variable between sites) recorded > 20% leaf area lost, and 6–22% of willow genotypes, as a direct result of rust. Rust was generally highest in southern England.

Predicted yield

Computed mean yield values ranged between 4.9 and 10.7 odt ha−1 yr−1, for genotypes Beaupré and Q83, both observed in the second rotation (Table 3). Genotype Q83 reached the highest annual total yield coverage over the 6 yr of the field trials (113.9 Modt yr−1 across England and Wales). Other genotypes also produced high national yield coverage, including willow genotypes Jorunn and Trichobel (111.4 and 113.7 Modt yr−1, respectively). Total computed yield coverage is the annual odt yield on all available land which is not classified as urban or water and is below 250 m.

Table 3.  Computed mean yields (oven-dry tonnes (odt) ha−1 yr−1), r2 scores and three highest-ranking variable importance plot (VIP) scores for poplar (Populus) genotypes Beaupré, Ghoy and Trichobel and willow (Salix) genotypes Germany, Jorunn and Q83 (P < 0.05)
GenusGenotypeRotationMean yieldr2VIP scores
  1. Root mean standard errors for mean yields and relative percentile contributions for VIP scores given in parentheses temp., temperature; prec., precipitation.

PopulusBeaupréFirst7.42 (1.25)0.70Elevation (4.2), Feb. temp. (3.5) and Oct. prec. (3.1)
PopulusGhoyFirst6.50 (1.38)0.69Slope (4.1), elevation (3.3) and Feb. temp. (3.2)
PopulusTrichobelFirst9.31 (1.37)0.68Feb. temp. (3.5), slope (3.4) and Jun. prec. (3.3)
SalixGermanyFirst7.05 (1.83)0.55Mar. prec. (3.7), slope (3.6) and Feb. prec. (3.6)
SalixJorunnFirst9.29 (2.09)0.51Soil pH 25–50 cm (4.6), soil pH 0–25 cm (3.9) and elevation (3.6)
SalixQ83First8.21 (2.09)0.58Mar. prec. (3.5), slope (3.3) and Sep. prec. (3.2)
PopulusBeaupréSecond4.90 (1.38)0.69Soil silt % (4.7), soil sand % (3.9) and Jan. temp. (3.7)
PopulusGhoySecond5.85 (1.24)0.74Elevation (4.1), soil pH 0–25 cm (3.9) and annual prec. (3.7)
PopulusTrichobelSecond9.70 (1.38)0.75Jan. temp. (3.9), Oct. temp. (3.8) and Aug. temp. (3.2)
SalixGermanySecond7.49 (2.46)0.61Feb. temp. (3.9), frost days (3.8) and Jul. prec. (3.8)
SalixJorunnSecond9.30 (1.77)0.61Elevation (4.8), Avery soil class 8 (4.0) and Jan. temp. (3.7)
SalixQ83Second10.72 (1.38)0.58Avery soil class 5 (4.7), Dec. temp. (4.0) and Avery soil class 8 (3.9)

Root-mean-square error (RMSE) describes the range of uncertainty associated with the primary computed observation, in this case yield. RMSE of the empirical models developed in this study is variable and reflects the large difference in yields between and within sites (Table 3). The models predictive abilities are assessed by their r2 values, which compare observed and predicted yield scores (Fig. 4). Results show that the variables inputted into the empirical model were able to account for between 51 and 75% of the total variation in yield.

Figure 4.

Examples of relationship between observed and predicted yield values (oven dry tonnes (odt) ha−1 yr−1) at plot level for poplar (Populus) genotype Trichobel (a, d) and willow (Salix) genotypes Jorunn (b, e) and Q83 (c, f). (a–c) first rotation; (d–f) second rotation (P < 0.05). RMSE, root-mean-square error.

Yield maps derived from the empirical models of the three most productive genotypes show the spatial difference in productivity (Fig. 2); yields were generally highest in the northwest and lowest in the southeast, although some of the yield variation is lost as a result of upscaling. Yields for poplar genotype Trichobel were highest in southern and western Wales and eastern and northwestern England, and lowest in the northeast of England. Willow genotype Jorunn similarly showed high yields in the east of England and in the Midlands, decreasing in the northeast and southwest of England. The yields of willow genotype Q83 were highest in the west and lowest in the southeast of England.

Figure 2.

Example of spatial productivity maps (25 × 25 km resolution) for poplar (Populus) genotype Trichobel (a, d) and willow (Salix) genotypes Jorunn (b, e) and Q83 (c, f). (a–c) first rotation; (d–f) second rotation (oven-dry tonnes (odt) ha−1 yr−1).

The dominant factors affecting the spatial distribution of the extensively grown genotypes were hydrological. PLS principal-components scatterplots identify the high association between hydrology (particularly spring and summer rainfall) and SRC yield. Willow genotypes generally featured a higher number of hydrological factors, scoring > 1 in VIP scores when compared with poplar (6.0 compared with 5.0), particularly in summer months. Willow also had a higher cumulative count for hydrological factors, including annual rainfall, which was the highest ranking variable for willow. Poplar genotype Trichobel was the most hydrologically sensitive of all the extensively grown genotypes, with a rotational average of nine hydrological factors scoring > 1. By contrast, only five hydrological factors scored > 1 for poplar genotype Ghoy per rotation, making it the least hydrologically sensitive genotype.

Models showed poplar had higher VIP scores for temperature compared with willow, particularly in the winter months, with February temperature ranking the highest across all three poplar genotypes. By contrast, willow genotypes featured more soil-specific factors in their high ranking scores than poplar. The relative significances of the highest-ranking three variables for each extensively grown genotype are given in Table 3.

Results also show SRC can be planted on a wide range of soil types, from heavy clay to sand, but yields were generally higher on loamy and clayey soils with naturally high water tables and lowest on pelosols and shallow lime-rich soils over chalk or limestone.

It is also important to look at the variation within each factor, as variance describes the correlation between the changing value of the variable against yield rather than simply the importance of a variable's presence to crop yield. The greatest positive variation occurred in summer precipitation, available water content, slope and frost days for willow genotypes and a combination of summer precipitation, summer temperature and frost days for the poplar genotypes, meaning that as these factors increase, so too does yield. Winter temperatures showed negative correlation in all but one genotype; therefore in areas of high winter temperatures you find low yields. By contrast, all genotypes showed a strong negative correlation between yield and elevation.

Land-use scenario analysis

The combined yield of the three most productive modelled genotypes (Trichobel, Jorunn and Q83) ranged from 4.7 to 15.8 odt ha−1 yr−1 across England and Wales (Fig. 3). Mean yields varied between land types but were not significantly different; yields were up to 0.5 odt ha−1 yr−1 lower than average on land under cereal production, and up to 0.4 odt ha−1 yr−1 higher on improved grasslands and set-aside. Replacing 10% of arable land (scenarios a–c in Table 2), 20% of improved grassland (scenario d) and 100% of set-aside grassland (scenario e) in England and Wales with these genotypes would yield 12.6 Modt yr−1 of biomass and require 1.3 Mha of land. Development under such a scenario could produce 25.8 TW h−1 of electricity (6.7% of total UK electricity production (MacLeay et al., 2007)) and displace 5.5 Mtoe (Cannell, 2003). Alternatively, if this biomass was used for CHP, these genotypes could produce 45.3 TW h−1 of energy (48.3% of total current CHP production in the UK (MacLeay et al., 2007)).

Figure 3.

Spatial productivity map (25 × 25 km resolution) combining the three highest-yielding extensively grown genotypes. Areas shaded white are excluded as they are above 250 m, urban or predominantly inland water. Areas shaded red are those identified through Getis-Ord Gi* hot spot analysis as the most suitable for the construction of new bioenergy power stations based on crop productivity (highest-ranking 2% based on Getis-Ord Gi* scores). Conversely, land shaded blue is the least suitable (lowest ranking 2%).

In order to broaden the picture of biomass potential in the UK, it is important to include Scotland in any calculations. By aggregating potential yields from this paper, land already under energy crop production in the UK (0.2 Modt yr−1 (NFFCC, 2006; Defra, 2006a)) and potential in Scotland, assuming a 5% uptake on the most suitable land (0.7 Modt yr−1 (Andersen et al., 2005)), the UK has the potential to supply 28.1 TW h−1 of electricity (7.3% of electricity production) or 49.3 TW h−1 of energy from CHP (52.6% of CHP production).

Spatial case study

To meet short-term renewable energy targets, co-firing is necessary. Applying the same land conversion as above within 25 km radius of Drax power station, 0.7 Modt yr−1 of biomass could be grown. This value would produce nearly 1.4 TW h−1 of electricity, which is approx. 5.7% of the station's electrical production. Therefore, an additional 40% more land may be required than is currently realistically available for Drax to meet its biomass targets from local sources.

In the longer term, if we are to meet wider renewable energy targets, new bioenergy power stations must also be built. Hot spot analysis found the most suitable locations were predominantly in south Wales and northwest England (Fig. 3). Spatial analysis does not take into account reduced yields at commercial scale or wider socioeconomic barriers and opportunities to uptake, but instead demonstrates that a variety of locations may be suitable for new bioenergy infrastructure.


Characterization of yield variation in SRC

Results show that SRC is generally suitable for growth across a range of land cover types, including set-aside (Table 2). This may be because of the tolerance of poplar and willow species to high soil toxicity and waterlogging (Robinson et al., 2000; Nixon et al., 2001). However, field measured yield was variable between species and genotypes, with willow in general generating higher yields than poplar, particularly in the second rotation.

Poplar and willow are largely undomesticated, and improvement with breeding is limited to the development of F1 hybrids. Rae et al. (2004) recorded yield for SRC poplar in an F2 progeny with highly differentiated stem and canopy architectures, between 0.04 and 23.7 odt ha−1 yr−1, suggesting that considerable genetic variation exists on which to improve this genus for future use. In particular, for poplar, all breeding programmes to date have focused on selecting plants for single straight stems with high apical dominance, including the genotypes used in this study. Lower second-rotation yields for poplar therefore suggest that this tendency for single stems with high apical dominance may detrimentally affect yield.

Rotational yields may also have been affected by irregular climatic conditions; for example, during the first rotation (1995–1999) rainfall was 3% above the 1961–1990 baseline, whilst during the second rotation (1999–2002) rainfall was 19% higher than the baseline (Perry & Hollis, 2005). The spatial pattern of climatic variability was also inconsistent between rotations and could explain changes in the geographic spread of yield over time. Alternatively, changes in temporal geographic distribution may be attributed to root system maturation, resulting in roots being able to take water and nutrients from greater depths.

Results suggest variation in rotational yield is also linked to rust (Table 1). It is known that infections causing leaf blemishes, such as rust, may be correlated with decreased photosynthetic activity and yield (Erickson et al., 2004). Infection decreases the energy available for processes involving the development of frost hardiness and influencing the successive year's growth (Christersson, 2006). Rust scores from the trials were generally higher in the second rotation, suggesting a breakdown in resistance. The pathogen can also remain in the leaf litter, on the stool or in a secondary host (i.e. larch) between rotations.

A parallel study conducted in Belgium, showed similar trends in genotypic mean annual yields, ranging from 2.0 odt ha−1 yr−1 for poplar genotype Gibecq to 10.4 odt ha−1 yr−1 for poplar genotype Hazendans (Laureysens et al., 2004). Genotypes with the highest first rotation yields again typically performed poorly in the second, a finding was generally attributed to poor rooting capacity and rust.

Other European trials have shown yields comparable to those achieved in the field trials network. For example, in southern Sweden 12 willow genotypes were grown (4 yr rotation, 20 000 plants ha−1), with yields ranging between 6.2 and 9.5 odt ha−1 yr−1 (Nordh & Verwijst, 2004).

A single-site trial in Washington state (US) recorded poplar yields of 14–35 odt ha−1 yr−1 (4 yr rotation, 10 000 cuttings ha−1) (Scarasscia-Mugnozza et al., 1997). High productivity was attributed to rich alluvial soils and length of the frost-free period. More recently a Canadian study (Labrecque & Teodorescu, 2005) showed poplar genotypes grown in southern Quebec could yield between 16.6 and 18.1 odt ha−1 yr−1 without the application of fertilizer (4 yr rotation, 18 000 cuttings ha−1). Willow biomass productivity from the same trial was between 9.0 and 16.9 odt ha−1 yr−1. It was hypothesized that the high yields found in these trials could be the result of high soil quality, good drainage conditions and a lack of disease infestation.

Factors limiting modelled SRC yield

Results indicate a strong correlation between water availability and SRC yield. Water use during summer months by mature poplar and willow exceeds most other vegetation (Hall & Allen, 1997; Lindroth & Båth, 1999) and studies have shown these fast-growing species tend to show poor resistance to drought stress (Lindroth & Båth, 1999; Wikberg & Ogren, 2004; Cochard et al., 2007), although a recent study on the effect of water availability on willow found that the species was able to cope with short-term reduced water availability by improving water-use efficiency (Linderson et al., 2007); however, stem growth was reduced. Despite drought not being a significant issue during the course of the trials, future predictions for lowland England suggest increased soil moisture deficit, leading to increased water stress, is likely during summer months (Hulme et al., 2002) and will remain an issue of critical future concern.

Results showed that hydrology affected species differently. VIP scores for the intensively grown genotypes identified willow as being more dependent on hydrological factors for growth than poplar. This is in accordance with the findings of Cochard et al. (2007), who established a strong correlation between high biomass in four willow genotypes and xylem vulnerability, when compared with five poplar genotypes. Other studies have found that differences in water use between poplar and willow are negligible (Hall et al., 1998). Thus caution is necessary in generalization on how these two species cope with drought. The dependence of poplar yield on hydrology was variable amongst genotypes, with small-leaved Ghoy being less vulnerable to reduced water availability than large-leaved Trichobel. This result is in accordance with the findings of Souch & Stephens (1998) and may be attributed to increased surface area for evapotranspiration in larger leaves and the inability of P. trichocarpa species to close their stomata during periods of drought. The genes determining drought tolerance in poplar have recently been identified using a transcriptomics approach (Street et al., 2006) and it is likely that in future such information will be used to develop genotypes with improved water-use efficiency within poplar and willow breeding programmes.

Model limitations and uncertainty

Not all of the variation in yield could be accounted for by the variables used in the model; therefore, results suggest yield is governed by complex interactions amongst a large number of site variables that cannot be fully quantified using this approach. The field trials network covers a contextually small number of sites and a large number of associated variables – leading to confounding relationships between correlated variables. Model outputs should be considered a guide for upscaling rather than a rule, as multicollinearity can and does show incongruous results (i.e. if increasing rainfall is strongly correlated to rising yield, then decreasing temperature may also be linked to increasing yields as areas of high rainfall and low temperature in the UK can occupy the same area). Models would have been improved by the provision of more site data across a wider geographic range.

It should also be noted that the results provided have a static temporal context. The genotypes used in the trials are no longer extensively utilized on a commercial scale, planting densities have increased by 50% and SRC is now planted in mixtures of genotypes to improve disease resistance. Despite these limitations, the models developed in this study represent the best predictive tools for measuring UK SRC productivity currently available to growers and planners. It is likely that the spatial context of the outputs will remain valid for genotypes from similar genetic backgrounds and serve as a useful predictor for areas of the UK more suited than others for SRC bioenergy crops.

Future productivity of SRC

In the future, breeding programmes and genomic tools should enable the rapid deployment of plants with improved productivity (Taylor et al., 2001; Sims et al., 2006). Rook (1991) suggested improved breeding could theoretically produce yields up to 30 odt ha−1 yr−1.

Production physiology has identified architectural traits associated with high poplar and willow biomass growth, including the production of large leaves, leaves with many small cells and late-season branching (Casella & Sinoquet, 2003; Rae et al., 2004; Robinson et al., 2004; Marron & Ceulemans, 2006). The availability of the full poplar DNA sequence should enable the identification of underlying genes that control these traits. Areas of the poplar genome determining yield have already been identified (Wullschleger et al., 2005; Rae et al., 2008), suggesting large-step improvements in yield are likely in the future.

Management regimes should also be considered, since increasing the plant spacing within rows of coppice crops, such as poplar, is likely to increase their light interception efficiency, while increasing planting densities should benefit less productive ones because of their weak potential in canopy closure dynamic (Casella & Sinoquet, 2007). Planting genotypes with narrow leaves and small petioles may similarly increase productivity of high-density coppice poplar crop systems by improving light interception (Casella & Sinoquet, 2007).

Furthermore, studies on climate change impacts on crop productivity have suggested an atmosphere of elevated CO2 could contribute to a rise in poplar yields of up to 27% by 2050 (Calfapietra et al., 2003). Future work to develop a process-based productivity model for SRC should enable linkages between climate change scenarios and productivity to be investigated.


Both poplar and willow may be suitable for biomass production across a wide range of sites in the UK, with recorded yields varying between 1.97 and 13.34 odt ha−1 yr−1. Upscaling the yields of the three highest-yielding extensively grown genotypes showed that the UK has the potential to exceed its short- and long-term energy crop targets to plant 125 000 ha (Britt & Garstang, 2002) and 1 Mha (Gill et al., 2005), respectively. If these genotypes (or similar) were grown across the UK on the land proposed by this paper, 1.3 Mha could be grown to produce 12.6 Modt yr−1.

The UK has a largely rain-fed agricultural system and its is unsurprising that low precipitation was identified as the principal limiting factor to crop yield and should be taken into account during site selection and for future breeding and improvement programmes. Recorded and modelled yields were in general less than 50% of potential yields, suggesting step-change improvements are likely over the coming decades in these largely undomesticated plants. In the future temperate landscape, it is likely we will see an increase in value and production of these crops as feedstocks for heat, power and liquid transportation fuels.


This research was funded by NERC as part of the ‘Towards a Sustainable Energy Economy’ initiative (http://www.tsec-biosys.ac.uk) and through a PhD studentship to MJA (NER/S/J/2005/13986). GT was supported by UKERC as part of the ‘Future sources of energy’ theme, led by the University of Edinburgh. We thank Forest Research for the provision of the site data, and particularly Tim Randle for help and advice in developing the modelling approach.