Evaluation of the CropSyst model for simulating the potential yield of cotton

Cotton produced in Uzbekistan has a low water and fertilizer use efficiency and yield is below its potential. To introduce improved production methods, knowledge is required on how the agro-ecosystem would respond to these alternatives. For this assessment, dynamic simulation models such as the crop-soil simulation model CropSyst are useful tools. CropSyst had never been applied to cotton, so it first was calibrated to the cotton variety Khorezm-127 grown under researcher-managed optimal conditions in the Khorezm region of Uzbekistan in 2005. The model performance was evaluated with a data set obtained in 2004 on two farmer-managed sites. Both data sets comprised in-situ measurements of leaf area index and aboveground biomass. In addition, the 2004 data set included the normalized difference vegetation index derived from satellite imagery of the two cotton fields, which provided estimations of leaf area index with a high temporal resolution. The calibrated optimum mean daily temperature for cotton growth was 25 °C., the specific leaf area 13.0 m2 kg−1, the leaf/stem partition coefficient 3.0, the biomass/transpiration coefficient 8.1 kg m−2 kPa m−2 and the radiation use efficiency 2.0 g MJ−1. Simulations matched 2005 data, achieving a root mean square error between simulated and observed leaf area index and aboveground biomass of 0.36 m2 m−2 and 0.97 Mg ha−1, respectively. The evaluation showed that early cotton growth and leaf area index development could be simulated with sufficient accuracy using CropSyst. However, final aboveground biomass was slightly overestimated by CropSyst, because some unaccounted plant stress at the sites diminished actual aboveground biomass, leading to a root means square error of around 2 Mg ha−1. Some characteristics of cotton, such as the indeterminate growth habit, could not be incorporated in detail in the model. However, these simplifications were compensated by various other advantages of CropSyst, such as the option to simulate crop-rotation or its generic crop growth routine that allows modelling of additional, undocumented crops. The availability of normalized difference vegetation index data with a high temporal and acceptable spatial resolution opened possibilities for a precise, in-expensive and resource-efficient way of model evaluation.