Impact of calibration strategy and data on wheat simulation with the DSSAT-Nwheat model

Abstract

Cropping system models (CSMs) are valuable tools for analyzing genotype, environment, and management (G × E × M) interactions in crop production. To apply a CSM in a new region with specific soils, climate, and cultivars, proper calibration and evaluation are required. However, calibration methods vary widely, often depending on modelers' expertise and approach. This study compares three calibration strategies for the DSSAT-Nwheat model using two datasets: one including yield components (1000-kernel mass, ears per m², grain number per m²) alongside phenology and grain yield, and another excluding yield components. The datasets cover ∼100 site-years of winter wheat (Triticum aestivum) data from German pre-registration trials and field experiments. The calibration approaches were (1) stepwise calibration of phenology, biomass, and yield, (2) simultaneous calibration of multiple genetic coefficients, and (3) a hybrid approach combining elements of both. The Time-Series cultivar coefficient estimator tool was used for implementation. Including yield component data improved model accuracy, reducing root mean square error (RMSE) by up to 10% for key variables such as phenology (3.4–5.5 days). Future wheat yield projections under selected climate scenarios varied by strategy and dataset, ranging from 6376 to 7473 kg ha⁻¹ in fertile, wet soils and 6108 to 6757 kg ha⁻¹ in poorer, dry soils. These results highlight the impact of calibration strategy and dataset choice on model performance. Transparent calibration practices are essential for improving CSM reliability in regional agricultural analysis under diverse environmental conditions.