Evaluation of Parameterization Strategies for Rice Modelling

Abstract

In a previous study [Confalonieri and Bocchi, 2005], the generic crop simulator CropSyst was parameterized and evaluated against biomass data collected on Japonica (early- and medium-late maturity) and Indica rice varieties but, upon further investigation, inconsistencies in the simulation of leaf area index emerged for the latter. The need for a re-parameterization of the model became apparent after carefully studying new field data made available in the meantime. In particular, some relevant parameters such as specific leaf area, stem-leaf partition, extinction coefficient and light-to-biomass conversion efficiency, are now available from reliable field measurements (respectively 27 m2 kg-1, 3.6 m2 kg-1, 0.59 and 3.2 g MJ-1) and differ considerably from those previously obtained via calibration (39 m2 kg-1, 1.5 m2 kg-1, 0.50 and 3.0 g MJ-1). In general, the agreement between simulated and observed values of crop biomass was pretty much the same with both parameters sets (average values of GSD [general standard deviation, also called relative root mean square error] were, respectively, 25% and 26% for old and new parameterization; corresponding values of EF [modelling efficiency] were 0.90 and 0.87; FAI, a fuzzy-based indicator given by the aggregation of GSD and EF assumed values of 0.09 and 0.17, 0 being the best value and 1 being the worst). Such simple comparisons figure out as the accumulation of mass and leaf area in crop models can be depicted in different ways and reasonable estimations of overall growth can be achieved by different pathways, not all of which being logically acceptable. A check on model parameters like the one performed here is worth to give some protection against spurious conclusions and indicate whether the model parameterization is conceptually consistent and related to reality.