Single-year experiments with broad sowing window are more critical than multi-year experiments with narrow sowing windows for phenology modelling and genotype classification

• Vernalization-photothermal model classifies canola genotypes and phenology. • Broad sowing windows are essential for genotype classification and modelling. • Narrow sowing windows limit generalisation and fail genotype classification. • Multi-year narrow-window trials give little improvement over single-year data. • Single-year broad windows are more informative than multi-year narrow windows. Phenology model calibration often relies on multi-year trials, yet the limited environmental contrast from narrow sowing windows may hinder accurate genotype classification and model generalisation. Here we compared the influence of sowing date range versus multi-year replication on parameterisation of a vernalisation–photothermal model for canola ( Brassica napus L.). Two field experiments were conducted at Wagga Wagga, New South Wales: (i) a multiple sowing experiment in 2024 with six sowing dates spanning 101 days, and (ii) multi-year trials in 2023 and 2024 with four sowing dates spanning 42 days each year. Observed flowering differed by up to 50 days between early and late sowings, providing strong environmental contrast for distinguishing genotypic vernalisation responses. Model fitting showed high predictive skill (R² = 0.83–0.99; RMSE = 1.1–3.1 days) and reliably separated spring and winter genotypes when parameters were derived from the whole sowing window (WSW). In contrast, parameters based on early (ESW) or late (LSW) sowings reproduced flowering only within their calibration period and often misclassified genotypes. Year-based calibrations (2023, 2024, or combined) produced broadly consistent results, with multi-year replication providing only minor improvements compared with single years. Long-term simulations further demonstrated that WSW-derived parameters were robust across environments, while restricted sowing windows led to biased extrapolations. We conclude that sowing date range is more influential than multi-year replication for robust phenology model calibration and genotype classification. These findings emphasise the importance of including broad sowing windows in experimental design to ensure reliable predictions for crop breeding, simulation modelling, and climate adaptation studies.