• Open Access

The development of MISCANFOR, a new Miscanthus crop growth model: towards more robust yield predictions under different climatic and soil conditions

Authors

  • ASTLEY HASTINGS,

    1. Institute of Biological and Environmental Sciences, School of Biological Sciences, University of Aberdeen, Cruickshank Building, St Machar Drive, Aberdeen AB24 3UU, UK,
    Search for more papers by this author
  • JOHN CLIFTON-BROWN,

    1. Biorenewables and Environmental Change Division, Institute of Biological, Environmental and Rural Sciences, Aberystwyth University, Plas Gogerddan, Aberystwyth SY23 3EB, UK,
    Search for more papers by this author
  • MARTIN WATTENBACH,

    1. Institute of Biological and Environmental Sciences, School of Biological Sciences, University of Aberdeen, Cruickshank Building, St Machar Drive, Aberdeen AB24 3UU, UK,
    Search for more papers by this author
  • C. PAUL MITCHELL,

    1. College of Physical Sciences, University of Aberdeen, Fraser Noble Building, Kings College, Aberdeen AB24 3UE, UK
    Search for more papers by this author
  • PETE SMITH

    1. Institute of Biological and Environmental Sciences, School of Biological Sciences, University of Aberdeen, Cruickshank Building, St Machar Drive, Aberdeen AB24 3UU, UK,
    Search for more papers by this author

Astley Hastings, tel. +44 1224 272702, fax +44 1224 272703, e-mail: astley.hastings@abdn.ac.uk

Abstract

Miscanthus×giganteus (Greef et Deu.), a perennial rhizomatous grass, native of SE Asia, has been trialed Europe-wide as a potential bio-energy crop. Plant growth models have been developed to match previously reported field experiments. These models have been used to extrapolate Miscanthus yields to other environments. Although the models use similar process descriptions, the parameters used to match the experimental data vary from site to site. This paper describes the development of universal process descriptions that use genotype-specific parameters to predict yields in a wide range of environments. Using these, we develop a new model, MISCANFOR, from an existing model MISCANMOD by improving process descriptions for light interception by the canopy and the impact of temperature and water stress on radiation use efficiency. Genotype-specific process descriptions for plant growth phase, photo-period sensitivity, thermal time, temperature dependant radiation-use efficiency, drought and frost kill predictions, nutrient repartition to the rhizome, and moisture content at harvest are added. Predictions made with MISCANFOR are compared with MISCANMOD for 36 experimental data sets for a wide variety of soils and climatic conditions in Europe. MISCANFOR matches field experiments with an r2=0.84 compared with 0.64 for MISCANMOD, building confidence that the new model will be better able to predict Miscanthus yields for other areas and future IPCC climate scenarios. This model has identified photoperiod sensitivity in addition to drought resistance and frost tolerance as parameters for crop improvement to extend the range of climatic conditions under which this crop can be grown economically.

Introduction

Miscanthus×giganteus (Greef et Deu), a perennial rhizomatous giant grass, native of SE Asia, has been trialled Europe wide as a potential bio-energy crop. These field trials have shown that it can be grown successfully in the wide range of European climatic conditions. Sims et al. (2006) showed that Miscanthus has the highest yield in terms of energy per hectare when compared with bio-ethanol, bio-diesel and short rotation coppice (SRC) willow biomass. Lewandowski & Schmidt (2006) demonstrated that the energy use efficiency (EUE), in terms of energy produced by a particular bio-energy crop as a ratio of the energy used in their cropping and fuel production, was highest for Miscanthus at 54 when compared with reed canary grass (26) and triticale (13). This value for Miscanthus compares favourably with other studies, for example, in Sweden (Börjesson, 1996) which gave EUE for SRC willow (21), reed canary grass (11) and wheat whole crop (7). Stampfl et al. (2007) used the MISCANMOD model, developed by Clifton-Brown et al. (2000, 2004) to show that Miscanthus has the potential to provide a high proportion of the European objectives for renewable energy. Hastings et al. (2008), using a FORTRAN version of MISCANMOD, demonstrated that Miscanthus could provide between 5% and 15% of Europe's primary energy needs by 2080 for future climate and land use scenarios if considered to be the only bio-energy crop.

Previous models have been developed using process descriptions for leaf development and radiation use efficiency (RUE) (Monteith, 1977) with parameters matched to the data gathered from a specific experimental site (Clifton-Brown et al., 2001). Various relationships have been used for the leaf expansion index counting degree days using a base temperature of above 0 °C, above 5 °C and above 10 °C and using extinction coefficients ranging from 0.2 (Price et al., 2004) to 0.61 (Greef, 1996). The maximum leaf area index (LAI) is reported to be between 6 and 8 m2 m−2 (Beale & Long, 1995), but values of 10–12 m2 m−2 have been reported (Tayot et al., 1995). The RUE for aboveground dry matter (DM) has a range of values from 1.4 g MJ−1 (Price et al., 2004) to 4.1 g MJ−1 (Tayot et al., 1995). MISCANMOD uses 2.35 g MJ−1, the value determined in the Cashel experiment in Ireland (Clifton-Brown et al., 2000). This can be compared with the maximum theoretical RUE for C4 plants of 4.6 g MJ−1 (Monteith, 1978). The physiological processes included in MISCANMOD are the start of leaf growth when the average temperature is above 10 °C and to limit the growing season for photosynthesis to 1800 degree days, using a base temperature of 10 °C. Other aspects of growth phases such as sprouting, leaf and plant senescence, nutrient repartition to the rhizome and above ground drying were not included, although these processes have been described in published experiments (Beale & Long, 1997; Lewandowski et al. 2000, 2003; Clifton-Brown et al., 2004; Heaton et al., 2004; Lewandowski & Schmidt, 2006).

In some models, Miscanthus growth has been reduced by soil water deficit (SWD) (Clifton-Brown et al., 2004), using discontinuous linear functions to achieve a match with the experimental data for both leaf expansion index and photosynthesis, but other influences, such as the availability of nutrients, photoperiod or temperature, have not previously been incorporated though such interaction has been observed in the laboratory and in field experiments (Beale et al., 1999; Farage et al., 2006; Ercoli et al., 1999).

A Miscanthus growth model, MISCANMOD (Clifton-Brown et al., 2000), was designed to use a subset of available meteorological data as although a complete time series of all daily data were available at the experimental sites for the development of the model, only a limited subset of monthly mean temperature, precipitation, global radiation (GlobRad) (Hulme et al., 1995) were available for the period 1960–1990 on a 0.5° grid for extrapolation to other sites. The Penman–Monteith formula (Monteith, 1965) was used to calculate potential evapo-transpiration (PET) as vapour pressure (Vpd) and wind speed were available in this data set. Recently, more meteorological variables have become available at a high resolution (Mitchell et al., 2004) so the model can be updated to use input variables such as maximum and minimum temperature and Vpd. However, as wind speed is not available for University of East Anglia Climate Research Unit (CRU) future scenarios (Mitchell et al., 2004), another approach is required for PET estimation. The Thornthwaite model (Thornthwaite & Mather, 1957) can be used to calculate (PET) with the empirical correction for arid climates used by the FAO (Deichmann & Lars, 1991) to match the Penman–Monteith estimation.

The correction of PET to actual evapo-transpiration (AET) in MISCANMOD used the Aslyng (1965) discontinuous linear process description, which is based upon soil moisture content. Now, both the wilting point (WP) and field capacity (FC) data are available from the FAO/IGBP (Global Soil Data Task Group, 2000), so the model can be updated to use the physically based corrections for AET used in the soil and water assessment tool (SWAT) model (Neitsch et al., 2005).

MISCANMOD only considers Miscanthus×giganteus (Greef et Deu.), a sterile triploid hybrid genotype, widely distributed in Europe but originated in East Asia, and does not limit the yield projections by considering temperature and drought conditions that would kill the shoot or rhizome. In certain areas of Europe, winter temperatures or summer rainfall are too low for Miscanthus×giganteus to survive every year. This is an important consideration for a crop which has an economic life of between 15 and 20 years as rhizome death would require reestablishment of the crop, rendering it economically unsustainable. Field and laboratory experiments have determined that Miscanthus×giganteus rhizomes, in the establishment year or first winter after planting, died at a temperature below −3.4 °C but interestingly, a high yielding Miscanthus sinensis hybrid (Sin-H6) was able to withstand −6.4 °C (Clifton-Brown & Lewandowski, 2000a). In the following years, when the stand is established, there are enough rhizome sprouts at depths below 5 cm to reestablish growth although freezing occurs. Weng (1993) and Clifton-Brown & Lewandowski (2000b), Clifton-Brown et al. (2002) also determined the physiological responses to water stress in the same genotypes, and they predicted that M.×giganteus would die if an extended drought were to occur. In 2003, an extreme drought occurred in central Europe and a stand established on a sandy site in central Germany, which was neither irrigated not fertilized, was killed (K.U. Schwarz, personal communication). A climate chamber experiment with M.×giganteus showed that 30 days without water resulted in shoot death but regrowth from the rhizome occurred on rewatering. No such regrowth was observed when watering was resumed after 60 days. M. sinensis (Sin-H6) was shown to be more drought tolerant than M.×giganteus and M. sacchariflorous (Sac-5) (Clifton-Brown & Lewandowski, 2000b; Clifton-Brown et al., 2002) and the leaves stay green even when subjected to more than 60 days of water stress, whereas the other genotypes exhibit leaf senescence. This sensitivity to water deficit can limit growth under rain-fed conditions and this limits to the rain-fed range of the crop, without irrigation, has not been previously modelled. Farrell et al. (2006) demonstrated considerable genotypic variation in low-temperature spring, regrowth and leaf frost tolerance. Using an early version of MISCANMOD, the impact of the genotypic variation was predicted to be up to 20% on peak harvestable yield in Germany (J. Clifton-Brown, unpublished results). Previous estimates of European yields used the average monthly climatic conditions for the 1960–1990 period (Clifton-Brown et al., 2000, 2004; Stampfl et al., 2007) and could not predict the inter-annual variation of yields, nor consider the occasional drought and frost events that would kill the crop or severely reduce the yield and hence limit the useful natural range of the crop without irrigation or frost protection. This requires the analysis of yield derived from a multiple year time series of climatic conditions.

This paper describes the development of a new model, MISCANFOR developed from the existing model MISCANMOD, using recently available CRU meteorological data (Mitchell et al., 2004) and soil properties (Global Soil Data Task Group, 2000) and including an improved description of response to SWD. We add genotype-specific process descriptions for plant physiological status, photoperiod, temperature dependant radiation-use efficiency, drought and frost sensitivity, nutrient partitioning to the rhizomes and moisture content at harvest. These relevant processes are parameterized for both M.×giganteus and M. sinensis and provision is made for other genotypes as data becomes available. Predictions made with the new process descriptions are compared with the previous versions for 36 experimental data sets covering the period 1990–2006, using measurements taken more than 2 years after crop establishment.

Materials and methods

The MISCANFOR model is based upon MISCANMOD (Clifton-Brown et al., 2000) with modifications to improve the process descriptions for evapo-transpiration, soil moisture content, photosynthetically active radiation (PAR), plant physiological time clock (here called ‘Physiostat’), water stress, hot and cold temperature stress, shoot and rhizome death, nutrient repartition to the rhizome and above ground DM moisture content. MISCANMOD used daily or monthly mean temperature (Meant), precipitation (Ppt), PAR, PET and irrigation time series for multiple or single years and soil data for single location or multiple locations. MISCANFOR outputs the following information: daily increment and year-to-date sum of above ground DM, LAI and other predicted variables that can be compared with plant measurements and other data available for a given experimental site.

MISCANMOD was parameterized using field observations for 1994 and 1995 on one site in Cashel, Ireland (Clifton-Brown et al., 2000). Here, data are used from two sites at Cashel, with data from 1991 to 2006 (Clifton-Brown et al., 2007) to develop the process descriptions, and used experimental data from Ireland (Clifton-Brown et al., 2000), Germany (Clifton-Brown et al., 2002; Lewandowski et al., 2003; Lewandowski & Schmidt, 2006), Denmark (Jørgensen, 1997), Portugal, Spain, the United Kingdom, the Netherlands, Belgium, Italy, Greece (Clifton-Brown et al., 2002), and the Netherlands (Vleeshower, 2001), all with monthly measurements of yields and site specific, daily meteorological data, to extend the applicability of the model for different environmental conditions (Table 1a and b). The total number of data sets that are complete enough to test the model, that had monthly measurements of peak DM yield, LAI and harvest yield or monthly height and peak autumn yield, is 11.

Table 1a.    Miscanthus experimental data sets used in this study with geneotypes used and nitrogen and irrigation treatments
CountrySite nameDailyLocationEstablishmentStartStopIrrigated (mm)Nitrogen (kg/ha)Genotypes
MeteoLongLatSource
AustriaAtzenbrug 15.648.2Schwarz et al. (1994)198919901991No0/60/90/120/180Mxg
AustriaAtzenbrug 15.648.2Schwarz (1993)198919901991No60Mxg
AustriaMarkgrafneusiedl 16.448.2Schwarz (1993)198919901991No60Mxg
AustriaSteinbrunn 16.347.5Schwarz (1993)198919901991No60Mxg
AustriaIlz 15.647.0Schwarz (1993)198919901991No60Mxg
AustriaSt Florian 14.248.1Schwarz (1993)198919901991No60Mxg
AustriaKirchberg 14.248.1Schwarz (1993)198919901991No60Mxg
BelgiumStreeDaily5.350.5Clifton-Brown et al. (2001)199319931994No0/60/120Mxg
DenmarkMosegaard 9.656.5Vargas et al. (2002)199219961996No0/75Ms Goliath
DenmarkFoulumDaily9.656.5Jørgensen (1997)199019931995No0/75/140Mxg/Ms
DenmarkFoulumDaily9.656.5Jørgensen et al. (2003)199719971999No60Mxg(4), Msa(1), Msh(5), Ms(5)
DenmarkFoulumDaily9.656.5Clifton-Brown et al. (2001)199719971999No60Mxg(4), Msa(1), Msh(5), Ms(5)
FranceGrignon 2.048.0Tayot et al. (1995)199019911994Yes120Mxg
FranceLusignan 1.044.0Tayot et al. (1995)199019921992Yes120Mxg
FranceEstrees-MonsDaily3.152.9Private communication200420052006No0Mxg
GermanyBoitzenhagen 10.852.6Kahle et al. (2001)199019941997No0/50/100Mxg
GermanyKlein Markow 12.653.9Kahle et al. (2001)198919941997No0/50/101Mxg
GermanyGuntersleben 9.949.9Kahle et al. (2001)198819941997No0/50/102Mxg
GermanyBraunschweigDaily10.452.3Clifton-Brown et al. (2001)199019901995No0/60/120Mxg
GermanyDurmersheim 8.149.0Lewandowski et al. (2003)199219941996No50Mxg
GermanyGutenzell 10.048.0Lewandowski et al. (2003)199319941996No0Mxg
GermanyIhinger HofDaily9.048.7Lewandowski et al. (2003)199219941996No50Mxg
GermanyRhine valley 6.751.5Himken et al. (1997)198919921992No0/90/180Mxg
GermanyIhinger HofDaily9.048.7Clifton-Brown et al. (2002)199719971999No60Mxg(4), Msa(1), Msh(5), Ms(5)
GreeceCefalonia 20.538.2Clifton-Brown et al. (2001)199019911993Yes0/60/120Mxg
GreeceCRES – Kopias 23.138.4Clifton-Brown et al. (2001)199319941994Yes0/60/120Mxg
GreeceVolos 22.739.4Danalatos et al. (2007)199820012002600/40050/100Mxg
IrelandCashelDaily−7.852.7Clifton-Brown et al. (2001)199019901994No0/60/120Mxg
IrelandCashel-1Daily−7.852.7Clifton-Brown et al. (2000199019941995No120Mxg
IrelandCashel-1Daily−7.852.7Clifton-Brown et al. (2007)199019962005No0/120Mxg
IrelandCashel-2Daily−7.852.7Clifton-Brown et al. (2007)199319962005No120Mxg
Table 1b.    Miscanthus experimental data sets used in this study with geneotypes used and nitrogen and irrigation treatments
CountrySite nameDailyLocationEstablishmentStartStopIrrigated (mm)Nitrogen (kg/ha)Genotypes
MeteoLongLatSource
ItalyCatania 15.037.4Clifton-Brown et al. (2001)199319931995Yes0/60/120Mxg
ItalyCatania 15.537.4Cosentino et al. (2007)19931994199525%/50%/100%0/60/120Mxg
ItalyENEA – Trisaia 11.144.1Clifton-Brown et al. (2001)199319931995No0/60/120Mxg
ItalyENEA – Brasimone 16.640.2Clifton-Brown et al. (2001)199319931995Yes0/60/121Mxg
NetherlandsLelystad-BTGDaily5.950.9Vleeshower (2001)199319971997No60Mxg
NetherlandsTer Apel (NE) 7.152.9van der Werf et al. (1993)198819911991No30 mm/ha pig slurryMxg
NetherlandsValthermondDaily7.052.9Clifton-Brown et al. (2001)199319931995No0/60/118Mxg
NetherlandsWijnandsradeDaily5.452.5Clifton-Brown et al. (2001)199319931995No0/60/119Mxg
PortugalUNINOVA −9.038.7Clifton-Brown et al. (2001)199019911995Yes0/60/120Mxg
PortugalÉvoraDaily−9.238.7Clifton-Brown et al. (2001)199719971999300/382/50660Mxg(4), Msa(1), Msh(5), Ms(5)
SpainSantiago −8.442.9Clifton-Brown et al. (2001)199319931995No0/60/119Mxg
SwedenSvalöf WeibullDaily14.056.0Clifton-Brown et al. (2001)199719971999No60Mxg(4), Msa(1), Msh(5), Ms(5)
TurkeyKonya 32.538.0Acarğlu & Aksoy (2005)199920002001Yes0/50/100/150/200Mxg
UKArthur Rickwood 0.152.4Clifton-Brown et al. (2001)199219941998No60Mxg
UKEssex Writtle 0.451.7Beale et al. (1996)199219921993133120Msg
UKEssex Writtle 0.451.7Beale et al. (1999)199219941995200120Msg
UKEssex Writtle 0.451.7Beale & Long (1997)199219941994200120Mxg
UKRothamstedDaily−0.451.8Clifton-Brown et al. (2001)199319951995No0/60/120Mxg
UKRothamstedDaily−0.451.8Powlson et al. (2005)199319932002No0Mxg
UKAurther Rickwood 0.152.4MAFF NF0403 (2001)199219941998No60Msa/Mxg
UKRosemaund −2.652.1MAFF NF0403 (2001)199219941998No60Msa
UKBuckfast Abbey −3.850.5MAFF NF0403 (2001)199219941998No60Msa
UKBoxworth 1.552.3MAFF NF0403 (2001)199419961998No60Msa
UKGleadthorpe −1.153.1MAFF NF0403 (2001)199419961998No60Msa
UKHigh Mowthorpe −0.653.1MAFF NF0403 (2001)199419961998No60Msa
UKAurther Rickwood 0.152.4Price et al. (2004)199219942000No60Mxg
UKRosemaund −2.652.1Price et al. (2004)199219942000No60Mxg
UKBuckfast Abbey −3.850.5Price et al. (2004)199219942000No60Mxg
UKBoxworth 1.552.3Price et al. (2004)199419962000No60Mxg
UKGleadthorpe −1.153.1Price et al. (2004)199419962000No60Mxg
UKHigh Mowthorpe −0.653.1Price et al. (2004)199419962000No60Mxg
UKBridgets −1.351.1Price et al. (2004)199419962000No60Mxg
UKRothamstedDaily−0.451.8Clifton-Brown et al. (2001)199719971999No60Mxg(4), Msa(1), Msh(5), Ms(5)
USAShabbona, IL −88.941.9Heaton et al. (2004)200220022004est only25Mxg
USAUrbana, IL −88.240.1Heaton et al. (2004)200220022004est only25Mxg
USASimpson, IL −88.737.5Heaton et al. (2004)200220022004est only25Mxg

Another 25 data sets with published annual yields (Table 1a and b) from crop experiments in Sweden and Portugal (Clifton-Brown et al., 2001), Germany (Kahle et al., 2001), Greece (Danalatos et al., 2007), Italy (Ercoli et al., 1999; Cosentino et al., 2007), the United States (Heaton et al., 2004) and England (Beale et al., 1996; Price et al., 2004), using either monthly meteorological time series from the (1901–2000) University of East Anglia CRU 0.5° data base (Mitchell et al., 2004) or from actual daily time series, were used to validate the performance of the new process descriptions and parameters at independent sites.

The new process descriptions were developed in the sequence they are used in the model using data from the available test sites. The parameters in these process descriptions were adjusted to give the best RMS degree of fit between the predicted and observed variables from the trial sites (parameter optimisation). For example, formulas to predict the RUE with leaf temperature were developed from experimental data (Farage et al., 2006) and then tested to predict field trial DM harvest yield at all available trial sites and the RMS error estimated. Equation parameters were then adjusted to minimize the RMS error of the predictions. When the best-fit was obtained for a given process description, the work progressed to the next process description in the model sequence.

Deriving meteorological input variables from available data

Meteorological data are available globally as monthly averages (Mitchell et al., 2004) for the following variables: maximum (Maxt) and minimum (Mint) temperature, precipitation (Ppt), monthly average cloud cover (Mcc), maximum (maxVpd) and minimum Vpd deficit (minVpd). PAR and PET have to be estimated from the other meteorological variables.

The plant growth process descriptions used in MISCANFOR use daily time series of Maxt, Mint, Ppt, PET and PAR. Most experimental sites measure the daily GlobRad at the site and also estimate PAR as 50% GlobRad (Jones, 1992). Measurements of PET are not generally available. MISCANFOR is designed to use either the measured values for the meteorological variable, or to estimate them from the available variables. For example, measured GlobRad is used directly when available or can be calculated using latitude, day of year, water Vpd and average cloud cover using the SWAT2000 method (Arnold & Fohrer, 2005). This method includes corrections for solar distance, daily declination and latitude. Photo-synthetically active radiation (PAR) is then calculated from GlobRad using mean Vpd to determine the partitioning of energy between long and short wave radiation, and the LAI, calculated the previous day, to calculate the albedo to determine reflected energy. The correction made for cloud cover was verified at experimental sites, where the measurements for GlobRad and PCcloud were available, to validate the algorithm used in MISCANFOR. The correlation was high (r2=0.94, P>0.001).

image(1)

We use the empirical Thornthwaite equation (Thornthwaite & Holtzman, 1939), which is based on mean temperature, to calculate PET (ThornthwaitePET). However, as the Thornthwaite equation was developed for humid temperate climates and MISCANFOR and will be used from temperate to arid climates, we apply the FAO correction (Penmanfactor) to account for variations in annual rainfall, and to produce a value of PE that closely matched the Penman–Monteith calculation (PenmanPET).

image(2)

where:

image(3)

(where Σ represents annual sums of precipitation and PET)

Soil water content and drought stress

MISCANFOR calculates the AET as a proportion of PET by using a three step process description, first considering the evaporation of rainfall intercepted by leaf and stalk and then leaf transpiration, which is related to LAI and limited by the capillary pressure of the remaining soil water, and finally evaporation from the soil by diffusion through the air filled soil space. The process description is constrained by daily precipitation and PET. The process is similar to the one used in SWAT2000 (Neitsch et al., 2005), and an example of the process for a clay soil, with soil water parameters of a FC (capillary pressure>10 kPa) of 120 mm and a WP (capillary pressure >1500 kPa) of 60 mm is shown in Fig. 1. The shape of the curve is similar for other soil water parameters between the FC, which could be up to 600 mm and WP which will vary with the clay content of the soil. Transpiration dominates at FC and decreases toward the WP. At this value, soil water evaporates in the air filled soil pore space and diffuses to the surface. When the LAI>3 the process description uses a value of 3 (Neitsch et al., 2005).

Figure 1.

 Process description relating actual evapotranspiration (AE) to potential evapotranspiration (PE) for different values of leaf area index (LAI) in m m−1 and soil moisture deficit in mm, the example shown for a clay soil with a wilting point (capillary pressure of soil water 10 kPa) of 60 mm and a field capacity (capillary pressure of soil water >1500 kPa) of 120 mm. For other soils the curves will be scaled between the field capacity (up to 600 mm) and wilting point (which depends on soil clay content). The relationship for LAI>3 is given the value of 3.

The difference between AET and precipitation is then used to determine the daily soil water change and hence the SWD. The daily SWD is used to calculate the capillary pressure threshold (CPT) of remaining soil water using a logarithmic relationship between FC of 10 kPa and WP. The CPT is then used to calculate a linear downregulation factor for photosynthesis with a value of 0 at WP and 1 at the FC (Weng, 1993). The WP used varies with genotype, 1500 kPa for M.×giganteus and 3000 kPa for M. sinensis (Clifton-Brown & Lewandowski, 2000b). The onset of drought stress on LAI is earlier (Weng, 1993 & Inman-Bamber & De Jager, 1986; Smit & Singels, 2006), so the CPT end point is set empirically at half of the WP for this factor by matching the experimentally observed slowing of leaf expansion with remaining soil water.

Physiological time clock of the model (Physiostat)

The cumulative degree days above 10 °C are used to estimate the total growing season (Clifton-Brown et al., 2002), subject to the constraints of Physiostat stage, frost and drought events. Cumulative degree days above 0 °C drive the physiological time clock of the model (Physiostat) and the phases of growth, which are dormancy, shoot emergence, leaf growth, leaf senescence, where LAI starts to reduce, and plant senescence, where the above ground DM diminishes due to leaf fall and nutrient partition to the rhizome. The conditions which trigger each phase for M.×giganteus were estimated by analysis of the best fit to the experimental data sets. These included degree days, actual temperature, soil moisture and photoperiod. Degree days are calculated as (temperature × days) when the minimum temperature is above the threshold, when part of the day is below the threshold the degree day contribution of that day is calculated as the area of the daily hourly sine wave fitted to Maxt and Mint that is above the threshold.

Leaf expansion and photosynthesis model

After shoot emergence, a leaf expansion and photosynthesis process description similar to that used by Clifton-Brown et al. (2000) is used to calculate the LAI. This considers the overall effective LAI to which a single extinction coefficient can be applied and will be different for each Miscanthus genotype as the stem density, height and leaf number will vary. In this model these variables are implicitly included in the relationship between degree days and LAI and the extinction factor for each genotype. The Cashel data set (Clifton-Brown et al., 2000) was used to determine the relationship for leaf expansion based upon degree days above 0 °C rather than the 10 °C for M.×giganteus. This is then combined with solar radiation and the photosynthesis rate to calculate the above ground DM production for each day. At the beginning of leaf senescence, leaf area is allowed to decay from the maximum at the rate of 0.03 m m−1 day−1, but at the same time photosynthesis is allowed to continue. This decline is derived from published observations at Cashel, Ireland (Clifton-Brown et al., 2000), Wageningen, the Netherlands (Vleeshower, 2001), Jutland, Denmark (Jørgensen et al., 1997, 2003), Writtle, UK (Beale & Long, 1997) and Catania, Sicily (Cosentino et al., 2007). At the onset of plant senescence, photosynthesis stops, and the DM yield is allowed to decay at a constant rate to 0.66 of peak DM yield at harvest time on March 1 the following year (Clifton-Brown et al., 2007). Although the actual harvest time varies considerably over Europe this is the last practical harvest time so the field is cleared before first shoot emergence. There was not enough field data to parameterize an optimum harvest date for different locations, although this was considered, for example, when the ground was hard with frost or the soil water fell below FC to allow heavy machinery on to the field. The key issue to model was the difference between peak yield and harvest yield, which was captured with a straight line fitting the reduction of peak to harvest data and assuming a fixed harvest date.

To address the differing values of RUE found in experiments, a process description was developed to calculate a variable RUE. This works by reducing the maximum possible RUE for the production of above ground DM for Miscanthus of 3.9 g MJ−1 which will occur when the leaves have developed above 25 °C and the actual temperature at the time of photosynthesis is above 25 °C, with no water or nutrient constraint. This value is reduced by three factors based upon temperature, nutrient and water stress. The temperature variation factor (TVF) is based upon the findings of Farage et al. (2006), who demonstrated that RUE depended upon both the temperature at which the leaf was formed and on the temperature at which photosynthesis takes place. The TVF effect is based on a two dimensional exponential function using input variables of the average temperature over the period of leaf formation and the daily temperature at the time of photosynthesis to produce a continuous variable of TVF. The process description is shown graphically in Fig. 2 for Miscanthus×giganteus. The resulting RUE is subject to the previously described drought stress factor.

Figure 2.

 Variation of RUE (g MJ−1) with the average temperature (°C) during the leaf formation and the actual temperature (°C) on the day photosynthesis occurs. This curve is for Miscanthus × giganteus and follows the experimental work by Farage et al. (2006).

Other calculations for statistical analysis and yield predictions

In addition to calculating the rain-fed DM yield from the meteorological conditions, we also calculate the yield that would be achieved without any water stress, and the amount and timing of water required for irrigation to achieve this yield. For all output variables, over a period of years, the mean and standard deviation (SD) are calculated to look at year-to-year yield variation for the period of interest. In addition to rainfall, the irrigation quantity time series can be added to the water balance equation. As the DM accumulated and the cumulative transpiration during the growing period are calculated the water use efficiency (WUE) of the crop can be estimated.

Frost and drought kill events

In addition to limiting the LAI and yield based on climatic conditions, conditions that will result in either the shoot or rhizome death are also calculated. Shoot death means that in a given year, there will be limited yield but a recovery the following year. Rhizome kill means that the crop needs to be replanted. The geographical range of Miscanthus is limited by climatic conditions, due to frost and extreme drought killing the shoot and its rhizomes, so for each year we calculate kill flags for drought or frost events that kill the shoot or rhizome. For drought conditions, we calculate the time below the wilting point: if this exceeds 30 days, then the shoot is killed for that year, if it exceeds 60 days for M.×giganteus the rhizome is killed and the crop destroyed. This was based upon a growing chamber water stress experiment with M. × giganteus (Clifton-Brown and Hastings, unpublished data). This is extended to 60 and 120 days for M. sinensis. Rhizome kill flags for frost conditions for the M. ×giganteus, (at a soil temperature of −3.4 °C at 10 cm depth) and M. sinensis genotypes (at a soil temperature of −6.4 °C at 10 cm depth) are calculated. This models frost kill during the establishment year and indicates a location that can experience crop establishment difficulties, even though they would not occur each year. In both genotypes, the plant stops photosynthesis at the first frost. These kill events should enable the geographic range of two Miscanthus genotypes to be determined for each climate scenario and the appropriate genotype parameters to be selected to the location. For mean annual yield calculation purposes, the yield for the year following a kill event is set to zero to reflect reestablishment of the crop.

DM repartition to the rhizome and nitrogen stress

The daily DM production is summed at the end of the growing season to give peak yield and then the DM is reduced linearly to 0.66 of the peak value by the harvest time on 1 March the following year to give the harvest yield. This accounts for the repartition of nutrients to the rhizome and leaf fall (Clifton-Brown et al., 2007). The moisture content at harvest is also calculated using a relationship developed on drying experiments from the Cashel site.

The reduction of RUE due to nutrient deficits is based upon parameters suggested by Ercoli et al. (1999). For nitrogen a stress factor is assigned a value of 1 for levels of 200 kg ha−1 and 0.5 for 0 kg ha−1 and is implemented in the model for cases where the crop is harvested at the end of the growing season before nutrients are allowed to repartition to the rhizomes. For the standard crop management of replacing nitrogen removed in the spring harvest of the field dried material, this model feature is not used.

Model evaluation

To evaluate the model, the predicted variables were compared with field experimental data. MISCANFOR was run using the site specific meteorological data with the measured soil input variables, and the daily incremental crop yield was compared with incremental harvests made during the crop experiments. MISCANFOR was then run using the CRU 0.5° grid meteorological data for the year of the experiment and the FAO soil data (Global Soil Data Task Group, 2000) to calculate the harvestable yield in each of the grid blocks that contained an experimental site for which meteorological data were not available. The model predictions for the crop growth and the kill parameters were validated for Miscanthus×giganteus (Greef & Deu.) and where appropriate data were available from crop experiments (Clifton-Brown et al., 2001), M. sinensis (Sin-H6 and others) and Miscanthus sacchariflrus (Sac-5 and others).

Interannual yield variation

To quantify the benefits of running the model for individual years for a given period compared with using the mean meteorological conditions over the same period, MISCANFOR was run using the 12 month CRU meteorological time series, representing the mean climatic conditions for 1960 to 1990 (known here as ‘mean meteo run’), the base case for climate change studies, and compared with the mean yields obtained from a run using the actual 30 year time series for the same period extracted from the CRU 1900–2002 time series (known here as the ‘mean yield run’). The resulting yields for Miscanthus×giganteus were displayed using ArcGis and the difference between the runs calculated and mapped.

Results

Limitations of the original MISCANMOD model

A linear regression of the original MISCANMOD model predictions with a number of experimental sites measurements had an R2=0.54 (n=11) with a linear coefficient of 1.03 (Clifton-Brown et al., 2004). This model was rewritten in FORTRAN (MISCANFOR) as the starting point of the development of the new model. A linear regression of this FORTRAN version of MISCANMOD with MISCANMOD had a R2=0.98 (n=11) with a linear coefficient of 1.00 demonstrating the model implementations gave similar results. When MISCANMOD was run on the entire time series of field experimental results in Cashel plots 1 and 3 between 1994 and 2005, the difference between the mean observed peak yield and the model prediction for this period was 0.323 Mg ha−1 with a SD of 3.17 Mg ha−1. This shows that while the process descriptions used describe well the mean plant growth of Miscanthus, there is much residual error that is not described by MISCANMOD.

Deriving meteorological input variables from available data

Meteorological conditions drive plant growth, and this is implicit in plant growth models. For this reason, it is important to minimize sources of error in the input meteorological variables. MISCANMOD was used to map the above ground DM for Europe using the average CRU climate data the period 1960–1990 using evapo-transpiration calculated outside the model using a Penman-Monteith formula (Arnell, 1999). Here, we ran the same model using the Thornthwaite method (Thornthwaite & Holtzman, 1939) described above and achieved similar results.

The SWAT method for the calculation of PAR was compared with data from those sites with a measurement of PAR, and in other sites the calculated cloud-cover-corrected GlobRad was compared with measured data. In each case a regression produced a unity relationship with an r2>0.97.

Soil water content and drought stress

A comparison between the Aslyng (1965) and the SWAT method (Arnold & Fohrer, 2005) for estimating AET from PET indicated that the Aslyng (1965) method underestimated AET in wet years and overestimated it in dry years. The SWAT method was in better agreement with the observations of the onset of drought conditions at the experimental sites, which was recognized by either the progression of leaf index and DM yield time series where available, or the plant height time series as a proxy where yield was unavailable.

Figure 3 shows the time series for the Cashel experiment for 1994 and 1995. The soil does not reach wilting point in 1994, but there is some indication of the onset of drought stress as the leaf index increment reduces at that point compared with the thermal time progression. In 1995, there is clear drought stress, and the plant stops growing abruptly at around 215 days for that year (580 days in Fig. 3), but continues growing later in the season after significant rainfall. This experimental observation is reflected in the drought stress computed by the model.

Figure 3.

 Time series of modelled soil water content in mm (SW) (mm) plotted for the Cashel site 1 for the period 1994 and 1995 in the top panel with the wilting point (WP) (mm) and the field capacity (FC) (mm). The middle panel is a plot of the modelled drought stress factor for radiation use efficiency (RUE) (g MJ−1) and leaf expansion index (LAI) (m m−1) derived form the estimated capillary pressure of the soil available water. The lower panel plots the resulting RUE from the model and the observed results. RUEmax (g MJ−1) is calculated from the average temperature during leaf development (Farage et al., 2006), RUEtemp (g MJ−1) is a function of RUEmax and the average temperature in daylight hours during photosynthesis. RUEeffect (g MJ−1) is the product of RUEtemp and the RUE drought stress factor. RUEfield (g MJ−1) is the observed value calculated from a linear regression through the observed weekly measurements. This example is for Miscanthus × giganteus.

By matching to the experimental data, it is observed that leaf expansion stops at a higher soil CPT than photosynthesis. The best match for Miscanthus×giganteus was obtained with a CPT endpoint of 600 kPa for LAI and 1500 kPa for RUE. The time series of water stress factors in the model are shown in Fig. 3 for the Cashel site for 1994–1995.

Physiological time clock of the model (Physiostat)

Thermal time is expressed as degree days (DD0) above 0 °C which drives the Physiostat clock along with additional criteria that were derived from observations on the experimental sites. For Miscanthus×giganteus, the onset of shoots starts when the DD0>650, the max temperature goes above 10 °C and the photoperiod is >12 h. A late frost will kill the shoots and the process will restart. Analysis of all of the trial data set showed that the sensitivity of the onset of shooting to photoperiod was significant P>0.002. None of the observations recorded shoots before 81 journal days, even in the case of adequate soil water and temperatures above 10 °C for the entire winter period. However, the days of the year and days after harvest were colinear in the data set so either could be the predictive variable.

Leaves start to develop at >850DD0, subject to the same constraints. Leaf expansion continues up to a maximum of 8 m2 m−2 or until DD0=2200 or 3 days below 10 °C or 3 days below the wilting point. At this time leaf area declines but photosynthesis continues with the remaining leaves until DD0=3200. This is equivalent to DD10=1800 which is used in MISCANMOD to define the growing season length. Plant senescence begins at this time, or earlier if there are 6 days below 10 °C, one frost day, or 30 wilting days for M.×giganteus and 60 days for M. sinensis. This is the time at which peak harvest is calculated. The plant is then allowed to dry and transfer nutrients from the stem to the rhizome. The progression of the physiostat is shown in Fig. 4 for Cashel 1994–19945.

Figure 4.

 The MISCANFOR modelled and observed leaf area index (LAI) (m m−1) and leaf and stem dry matter yield (DM) (Mg ha−1) for Miscanthus × giganteus grown at the Cashel trial in Ireland plotted for the 1994–1995 period. The top panel plots LAI. The second panel plots the DM. The third panel plots the thermal time above 0°C. The bottom panel plots the physiological plant growth stage (Physiostat), where 0 is dormant, 1 is shooting, 2 is leaf expansion, 3 is leaf senescence, 4 is plant senescence and 5 is plant drying.

Leaf expansion and photosynthesis model

After shoot emergence, using the Cashel (Clifton-Brown et al., 2000), Wadingen (Vleeshower, 2001) and Catania (Cosentino et al., 2007) data sets it was determined that the LAI expands at the rate of 0.006 m2 m−2 per DD0. The DD0 increment is subject to the drought stress factor for leaf expansion, derived from the remaining soil water capillary pressure CP. At the start of leaf senescence, LAI declines at the rate of 0.03 m2 m−2 day−1, determined by fitting experimental observations made at sites in Ireland and the Netherlands. The match of the calculated LAI compared with the Cashel time series for 1994–1945 is shown in Fig. 4.

As explained in the Physiostat section it was observed that shoots did not appear before 81 days in all of the experimental data sets available even though the degree day criteria was met and there was no drought limitation. In order to fit the LAI and yield data from Greece, Italy and Portugal the degree day count had to be started at 81 journal days. This enabled the same relationship to be used for both temperate and Mediterranean sites.

The photosynthesis process description uses an extinction factor of 0.68 (Clifton-Brown et al., 2000) and a maximum RUE of 3.9, which is reduced by the average leaf formation temperature and temperature during photosynthesis (Fig. 2). The effective RUE is then the product of the temperature derived optimum RUE and the RUE drought stress factor. An example of the model prediction for these is shown in the lower panel of Fig. 3 where the maximum, optimum and effective RUE are shown for the years 1994 and 1995 for Cashel, and are compared with the experimental value obtained for the year by a linear regression between absorbed radiation and incremental surface DM production, for each growing season. Owing to the lack of data sets with measured subterrain accumulation of plant material in the root and rhizome this could not be included in the model although it would increase the total plant RUE.

Surface crop drying

The model for Miscanthus drying up to the time of harvest and bailing was derived from an experiment in Cashel Ireland in 2002 where, after the first frost, three serial cuts were made of the crop and the moisture content was determined for each cut after the harvest until 4 March 2003, at this time the rest of the samples were harvested and allowed to dry in stooped bundles to mimic drying standing in the field (this was to avoid interfering with the 2003 growth) and sampled at three further time intervals. The ARCTAN function that gave the best fit (Fig. 5) to the data points is

image(4)
Figure 5.

 The stem moisture content in (%) from the first frost in Autumn on 21 October 2000 (day 1) until June the following year. The stand was harvested on 4 March the following year (day 134) and after harvest the bundled cut stems were left in stoops in the field. Five replicates of moisture content were made at each time sample (dots). The arc-tangent function model [Eq. (4)] response fits the experimental measurements with r2=0.94.

Frost and drought kill events

The frost and drought kill criteria used in MISCANFOR were tested using meteorological data from experimental sites in Europe. Frost kill of Miscanthus×giganteus at the Denmark, Sweden and N. Germany sites for the winter of (1997–1998) was correctly predicted as the soil temperature dropped below −3.4 °C at 20 cm (Clifton-Brown & Lewandowski, 2000a), and the model predicted that M. sinensis genotype would survive as the soil does not reach −6.4 °C at 10 cm. The drought kill criteria correctly predict the need for irrigation on a sandy site in Germany (Braunschweig) as the soil water was below the wilting point for 70 days. The model predicted the drought stress observed in the Writtle college farm site in 1994 (Beale et al., 1999) by matching the observed reduction in plant height increase with the model predictions of DM increase. The model also predicts that M.×giganteus does not grow without irrigation at the test sites in Portugal, Sicily and Greece, which were irrigated for the trials (Clifton-Brown et al., 2001; Danalatos et al., 2007; Cosentino et al., 2007).

Match of Cashel yield time series

MISCANFOR was run using the meteorological time series variables for the two Cashel sites from 1994 to 2005. One, classified as marginal land, had an estimated plant available soil water of 160 mm and the other required 180 mm to obtain a better match to the higher yields. MISCANMOD and MISCANFOR predictions of peak DM yield were compared with the observed peak harvest yields of Miscanthus×giganteus for each site/year and the difference calculated. The statistical description of these differences shows that the mean values are similar, however, the MISCANMOD SD is 3.18 Mg ha−1 and the MISCANFOR SD is 1.74 Mg ha−1.

Match of predictions for all experimental sites

The comparison between MISCANFOR peak DM predictions and measured yields was made for all of the available European site/years for Miscanthus×giganteus using both daily site specific and the CRU monthly meteorological variables (Fig. 7). A linear regression gives a unity relationship with an r2=0.84 (n=36) for the Miscanthus×giganteus results. The model was initially parameterized for Miscanthus×giganteus and only eight sites had data from M. sinensis and six from Miscanthus sacchariflorus. Insufficient data were available to fully parameterize the model for these varieties but matched yields were obtained by adjusting the maximum RUE. M. sinensis requires a maximum RUE of between 3.9 and 3.2 depending on the hybrid and 2.3 for M. sacchariflorus. In addition the frost tolerance of M. sinensis (Sin-H6) was increased to 6.4 °C and the drought tolerance was increased to 120 days at the wilting point. Data on frost and drought tolerance was not available for M. sacchariflorus.

Figure 7.

 Modelled and observed peak Autumnal yields of Miscanthus × giganteus dry matter (Mg ha−1), n=36, for all available European sites from Ireland, UK, Sweden, Denmark, Germany, the Netherlands, France, Portugal, Spain, Italy and Greece and the United States.

Comparison of MISCANMOD and MISCANFOR

The overall comparison of peak DM measured at the European sites with predictions by the MISCANFOR model shows a better fit (r2=0.67, n=11) than the MISCANMOD model (r2=0.54, n=11). When new sites are included using the published yields and the CRU meteorological and FAO soil data the degree of fit improves further (r2=0.84, n=36).

Modelling interannual yield variation 1960–1990

The mean yield for Europe derived from the mean value for each of the years 1960–1990 (mean yield run) was 15.0 Mg ha−1 yr−1 compared with 15.5 Mg ha−1 yr−1 for calculated using the mean meteorological conditions for the same period (mean meteo run). This small difference in the mean European DM value masked the large spatial variation in the difference. For most of temperate Europe, the mean meteo run yield is 20% less (0–3 Mg ha−1 yr−1) than the mean yield run, but larger, by up to 26 Mg ha−1 yr−1, in more arid areas such as continental and Mediterranean regions. The M×giganteus peak yields for the mean yields run for the period 1960–1990 (Fig. 8a) and mean meteo runs (Fig. 8c) with the difference between them (Fig. 8d) show this spatial variation in yield without considering frost and drought kill by setting the yield to zero for the year after a kill event to mimic reestablishment of the crop. The SD of the yield for this 30 year period is close to 20% of the mean yield at each grid point (Fig. 8b). These results indicated the need to use a meteorological time series rather than mean climatic conditions for current and future climate scenario projections of energy yield as well as for identifying potential kill events due to the extremes of interannual variation of temperature and rainfall.

Figure 8.

  Miscanthus × giganteus mean peak dry matter (DM) yields in Mg ha−1 calculated using MISCANFOR for the period 1960–1990 using the CRU mean meteorological conditions, (mean meteo run), (c) are compared with the mean of the DM yield in Mg ha−1 calculated with the meteorological time series for each year for 1960–1990, ‘mean year run’, (a) and the standard deviation (SD) in Mg ha−1 of the yield for these years (b). The difference between the mean meteo run and the mean year run in Mg ha−1 is displayed (mm-my) (d).

Discussion

The key improvements in the model are drought stress function, temperature variable RUE and the inclusion of photoperiod in the plant physiology model.

Analysis of all of the site data identifies that the reasons that previous models had used different base temperatures for the leaf index calculation was that photoperiod had not been included in the model. We identified that at the experimental sites, shoot emergence did not start before the 21 March which coincides with the vernal equinox P>0.001, but could also be related to the date of harvest in early March P>0.001. This was the case even when mean temperatures were above 10 °C and there was ample soil water. In the model we used both photoperiod and a mean temperature of 10 °C to define the shoot emergence point and start to degree day count using a base temperature of 0 °C to replicate the observations. However with the data available it is not possible to determine whether it is harvest time or photoperiod that is the critical factor (due to co-linearity) and this requires further investigation. The harvesting may not physiologically induce the development of shoots but removing the straw lets the sunlight reach the soil and heat it up. In this case, the amount of leaf litter or its burning can also influence emergence time.

Photosynthesis measurements at different ambient temperatures on Miscanthus×giganteus leaves grown at different temperatures (Weng & Ueng, 1997; Naidu & Long, 2004; Farage et al., 2006; Long et al., 2006) enables us to fit a process description that could be used in a model. These improvements improved the fit at all experimental sites to give the final r2=0.84. This also reduced the SD of the residuals from the comparison of MISCANFOR predictions and field trial results at the Cashel site from 3.2 to 1.7 Mg ha−1.

Enhancement to the conditioning of the input variables, such as meteorological and soil data, were made possible by the availability of new data sets. For example, the availability of maximum and minimum temperatures in the CRU data set allowed these variables to be used, instead of the mean temperature, to calculate thermal time which is used to derive the physical status of the plant development in the model. This also enables us to use continuous functions rather than step changes to calculate the degree days when the maximum and minimum temperature fluctuated around the degree day threshold. The availability of VP and cloud cover improved the accuracy of the PAR prediction, and as this is a key variable driving photosynthesis, the improvement has a profound effect when predicting yields at sites where the daily meteorological time series for all variables is not recorded.

The lack of wind speed in the CRU future climate scenario data sets precluded use of the full Penman–Monteith formulation to calculate PET. Wind speed was available in the historical data and we were able to check the modified Thornthwaite equation against the full Penman calculation for historical data for 1960–1990 for all sites, and a regression of both predictions gave a unity relationship with an r2 of 0.97. The ability to use the improved meteorological input variables fully had little impact on predictions at the experimental sites where full daily time series of all of the input variables were available, but should improve predictions of Miscanthus growth at new locations and for future climate scenarios.

The availability of the WP and FC and bulk density from the FAO-IGBP data set enabled us to use functions to calculate the CP of remaining soil water and hence improve the water stress process descriptions. However, this physics based approach yielded little improvement over the empirical Aslyng (1965) methodology but it does produce output parameters that can be related to plant measurements such as CP to turgor pressure and WUE.

The inclusion of the ability to use annual meteorological time series rather than the average conditions of 1990–1990 has enabled annual variations in yield to be analysed and has also shown that yield calculations made on average meteorological conditions tend to overestimate the yields when compared with the average of the yields of each year over the same period. Many past studies have used this approach, especially when predicting the geographical yields, and geographical ranges of bio-energy crops such as those made by Stampfl et al. (2007) for Miscanthus. The use of annual meteorological time series has also enabled a model to be built that is able to simulate the effects of extreme drought and frost events that can kill the crop and rhizome, so that the areas within which the crop is viable without frost protection or irrigation, can be identified. This will change the criteria for Miscanthus suggested by Tuck et al. (2006).

MISCANFOR was initially fully parameterized for Miscanthus×giganteus as most of the research on all aspects of Miscanthus used as a potential bio-energy crop to date has been undertaken on this genotype. M. sinensis (Sin-H6) was partially parameterized for drought stress, frost and drought tolerance but no data were available to construct a robust temperature dependant RUE process description, and this work needs to be undertaken in the future as it is the key to improving the fit to experimental data. At present we have only shifted the maximum RUE to fit the experimental data for the other genotypes of M. sinensis & M. sacchariflorus. However, the model has been constructed so that genotype specific parameters can be added as they become available.

The comparison between running MISCANFOR using the mean meteorological conditions for the period 1960–1990 demonstrates that the previous estimations of peak yields have been underestimated in the temperate regions by 20% and grossly overestimated in the arid climates and that the 20% interannual variation needs to be considered for energy supply planning. In addition the range of each Miscanthus variety is determined by its frost and drought hardiness and this requires the annual variation in temperature and rainfall to be considered for both current and future climatic conditions if energy yields and the resulting greenhouse gas mitigation are to be forecasted with accuracy.

A complete physiological description of new hybrids that are developed in breeding programs, such as at Aberystwyth, UK, is being made to facilitate predictive modelling. This will include frost and drought tolerance, WUE, leaf expansion rates and the relationship between leaf formation temperature, ambient temperature and photosynthesis.

Conclusions

The overall comparison of peak DM measured at the European sites with predictions by the MISCANFOR model shows a better fit (r2=0.67, n=11) than the old model (r2=0.54, n=11). When new sites are included using the published yields and the CRU meteorological and FAO soil data the degree of fit improves further (r2=0.84, n=36). MISCANFOR appears to be robust enough to be used to reexamine Miscanthus×giganteus yields for European conditions, including annual variation in yields and effects of global change. We have shown that to determine yields for different climatic conditions the variation of RUE with soil water availability and temperature and drought and frost tolerance are critical factors. The photoperiod sensitivity, although used to achieve a match with observations in the Mediterranean region needs to be verified experimentally. We have begun to include other genotypes, but further experimental data are required to parameterize MISCANFOR so that the predictions are as robust as for Miscanthus×giganteus.

Acknowledgements

This work was funded by a Sixth Century Scholarship from the University of Aberdeen as a joint project between the College of Physical Sciences and the College of Life Sciences and Medicine. We thank the Irish Meteorological Service (Met Éireann, in particular Mr. Padraig Carrigan) for providing the Kilkenny daily meteorological data, Prof. Mike Jones of Trinity College Dublin for suggestions during the field work of Cashel used in the analysis and Dr. Uffe Jørgensen and Dr. Klaus Hammel for early discussions on the model. Data were derived from numerous projects, of particular note are the EU contracts FAIR3-CT96 –1392 (EMI), FAIR3-CT96-1707 (EMN) and SSPE-CT-2005-006581 (ENFA). Other trial data was taken from publications and is acknowledged in the text.

Ancillary