The importance of chill model selection — a multi-site analysis

Winter chill, which temperate trees require in order to overcome dormancy, is expected to decrease substantially in the future in most deciduous fruit tree growing areas. Several mathematical models have been developed in different regions to quantify chill requirements of tree species and cultivars. The Dynamic model has emerged as the most plausible and reliable model, yet all chill models have been found inadequate in at least some growing regions. Accurate models are crucial for the development of quantitatively appropriate climate change adaptation strategies for temperate orchards. To demonstrate the importance of model choice we compared the outputs from 13 agricultural and forest chill models using past and projected future weather data for nine sites in Chile, Tunisia and Germany. To evaluate chill risk, we used a weather generator calibrated with 45 years of temperature data to generate 100 years of synthetic temperature records per scenario for multiple climate scenarios. Chill was computed for 10 past scenarios and projected for 60 future scenarios (for 2050 and 2085 according to greenhouse gas concentration scenarios RCP4.5 and RCP8.5, using projections from 15 climate models). Results show that estimations differ substantially across chill models, even for the same sites and scenarios. The “Chilling Hours” model and the “Chilling Rate” function showed high sensitivity across regions in future scenarios. The “North Carolina”, “Utah”, “Modified Utah” and “Low Chill” models all suggest negative chill levels for past and future scenarios in Tunisia (despite the thriving fruit tree industry there). Only two models projected chill decreases in all sites. In Mediterranean climate areas (central Chile and Tunisia) the “Dynamic” and “Positive Utah” models forecasted similar chill reductions for future scenarios, whereas in temperate locations (Germany) the “Dynamic” model forecasted lower chill increase compared with the “Utah” and “Positive Utah” models. Despite the “Dynamic” and the “Positive Utah” models showing similar performance among climates, the “Dynamic” model appears to be the best current option, due its more physiologically credible structure. However, further research is needed to develop or identify models that are valid across wide climatic gradients. Our results show that a major source of variation and inaccuracy in chilling assessments is the choice of the chill model used to make the assessment.