'Breeding' genetic coefficients in plant growth simulation models

Crop models in the DSSAT [Design Support System for Agrotechnology Transfer] suits include cultivar-specific coefficients that represent the genotype-environment interactions of the cultivar. These coefficients are thermal and photoperiod quantities used for simulating the reproductive development and for quantifying the vegetative and reproductive growth characteristics of the cultivar. The current techniques employed to calculate these coefficient are either by trial-and-error or by the use of GENCALC [Genetic Coefficient Calculator]. Trial-and-error is tedious to implement with uncertain results, while GENCALC adjusts the coefficients within the plant's realistic physiological ranges with non-optimal values. This paper presents a computational strategy that combines the benefits of trial-and-error and GENCLAC using selective breeding as a metaphor. A simulated breeding program is performed on a population of abstract entities whose genetic codes are the cultivar coefficients. If an entity's coefficients can provide simulated outputs closer to the experimental data, then the entity is said to be 'better'. The better entities, during one generation, will undergo a series of operations called selective mating and random mutation. These operations when applied over several generations to selected entities, the later generations' offsprings will encode coefficients with much better values than that of the earlier. This computational strategy was applied to 'breed' the cultivar coefficients of the CROPGRO-soybean model using the DSSAT's default experimental data involving the cultivar Bragg. MRE [Mean Relative Error] was used as a measure of closeness between the simulated and observed growth variables. The MRE of the tops weight, seed weight, leaf weight, stem weight, and SLA are lower with the simulations using the coefficients found by selective breeding than with the simulations using the coefficients found by GENCALC. Thus, selective breeding as a procedural metaphor for computing the cultivar coefficients of plant models can minimize the error between the simulated output and the observed data.