• A novel crop model calibration framework links phenology phases with N levels • Phenology-phase selection dominates uncertainty in single-phase calibration, while N-level selection dominates in combined-phase calibration • Findings can guide proper crop model calibration practices for precise N management Crop models have been widely used to optimize nitrogen (N) applications for agronomic decision-making, but uncertainties in model calibration under varying N levels, particularly the effects of phenology choice for calibration, remain underexplored. This study employed the ORYZA v3 model, coupled with a global optimization algorithm, to assess how different calibration strategies affected predictions of leaf area index (LAI) and biomass in two rice varieties under four N levels (0, 90, 180, and 270 kg ha -1 ). The results indicate when the model is calibrated separately for each N level, predictive accuracy varies considerably for both LAI and biomass, reflecting the difference in crop response to N availability. When calibrating the model simultaneously with multiple N levels from a single phenology phase, variability from different selected phenology phases becomes the dominant source of model uncertainty, rather than N levels. Specifically, calibrations using measured data from the stem elongation to anthesis (SA) and anthesis to maturity (AM) phases across N levels provide the most accurate predictions for LAI (RMSE: 0.51–1.92 m² m - ²; R ²≥0.88) and biomass (RMSE: 551–2619 kg ha -1 ; R ²≥0.96), respectively. In contrast, calibrations using measured data from early-season (transplanting to stem elongation) result in the least reliable predictions. Combined-phase calibrations using SA and AM phases result in the best predictions for both LAI and biomass, owing to their balanced representation of pre- and post-anthesis growth dynamics. This approach significantly reduces uncertainty from phase selection. However, variability in N application rates emerges as the primary uncertainty source in model simulations, emphasizing the importance of careful selection of N levels in calibration datasets, particularly when measured data span two-thirds of the growing season. These findings offer valuable insights into improved calibration practice for precise N management, highlighting the critical role of both phenology phase and N treatment selection.
Yang et al. (Sun,) studied this question.