Cross-regional validation of the PhenoFlex framework for flowering date prediction in apple: A study across Germany and Spain

• We tested location- vs. species-specific approaches for climate-proof predictions. • Species-specific parametrization improved apple flowering date predictions in Spain. • Location-specific parametrization enhanced model performance in Germany. • Enhanced global optimization covered more parameter space than global optimization. • Multi-site data integration improves the calibration of species-specific parameters. Modeling the dormancy period in temperate fruit trees is essential to assessing the impacts of climate change on tree phenology and spring frost events. Whereas cultivar-specific parametrization of phenology models may provide accurate predictions, it offers limited outlook at the species level. Using shared parameters for apple species alongside cultivar-specific parameters would improve phenology projections, provide insights into future risks, and inform strategies for adapting temperate orchards to climate change. In this study, we assessed the performance of PhenoFlex when used to estimate a common set of model parameters for apple trees from Germany and Spain. We used historical phenology and weather data for 5 German and 11 Spanish cultivars, testing two calibration approaches: location-specific (data grouped by country) and species-specific (combining data for all 16 cultivars), under two fitting procedures: a global optimization and an enhanced global optimization. Overall, species-specific calibration increased the PhenoFlex model accuracy, especially for the Spanish cultivars under the enhanced global optimization (RMSE of 4.6 vs. 5.4 days, respectively for ‘Clara’). On the other hand, location-specific calibration performed better for German cultivars. The enhanced global optimization reduced maximum errors from 12.0 to 7.5 days (‘Collaos’ in the species-specific fit), outperforming the global optimization procedure. Analysis of chill and heat response curves revealed variation across calibration approaches, with the species-specific calibration showing moderate responses. Our results highlight the trade-offs between generalizability and specificity in phenology modeling. Integrating multi-site data for the same cultivar could improve species-level parameter reliability and inform climate-resilient orchard management.