Dedicated biomass crops such as switchgrass (Panicum virgatum) and Miscanthus (Miscanthus x giganteus) are being proposed as feedstocks for a promising cellulosic ethanol industry. These crops could reduce the dependency on fossil fuels as well as reduce the net greenhouse gas emissions. These crops, however, have not been grown commercially at large scales and their success depends on their performance in a wide range of environments as well as the improvement in breeding as well as agronomic management practices. Mathematical crop models can be useful for knowledge synthesis, crop system decision management and policy analysis. In the context of crop models, complexity can be considered to be a function of the number of parameters. One disadvantage of complex crop models (i.e. many parameters) is that usually there is not enough experimental data to estimate these parameters accurately.  A simple example involving parameter estimation in a C₄ photosynthesis model and a more complex one which involves estimating dry biomass partitioning coefficients will be presented. Although there has been a great effort in developing models for other crops there is a need to integrate the current understanding of the mathematical description of leaf photosynthesis, canopy architecture, and phenological development for the application to new biomass crops like switchgrass and Miscanthus. The model proposed (BioCro, an R package) should be the first step toward developing a model which will allow to integrate current knowledge in crop models with statistical parameter estimation, which will allow for testing management practices as well as the impact of these cropping systems at a regional and/or global scale.

Results/Conclusions

Different methodologies for parameter estimation are more appropriate depending on the question and the model being used. Normally, for simpler non-linear models general algorithms are sufficient. However, the Bayesian approach (or the simpler bootstrapping) might be able to reveal subtleties in the posterior distribution which are not evident with the traditional approach based on normal approximations even for simple models (i.e. 3-5 parameters). More difficult problems with many parameters and linear restrictions required tailored numerical approaches such as Markov chain Monte Carlo.