Calibration and validation of grapevine budburst models using growth-room experiments as data source

Robust calibration of phenological models requires long term field observations, which are not always available or sufficiently widespread. This has motivated the evaluation of short-term experiments using cuttings under semi-controlled conditions as an alternative data source. Single-node cuttings from two grapevine cultivars were exposed to variable chilling durations and allowed to sprout in a growth room. The observed budburst dates and temperature series were used to calibrate two budburst models, which were validated against a 39-year field observation dataset by means of a fuzzy-logic based integrated index (FI). Satisfying validation scores were obtained, ranging from 0.262 to 0.411 on a 0-1 scale (best-worst response). The experiment was then inverted, using field data for calibration and cuttings for validation, and FI scores ranging between 0.352 and 0.495 were obtained. On this occasion however, the models were not able to estimate budburst occurring after short chilling exposures, where they returned either high overestimations or failed completely. This was due to the narrow winter length variability in the field dataset, which made the optimization algorithm converge towards unrealistically high chilling requirements and artifactual descriptions of the temperature effects on dormancy. Cutting-based calibration on the other hand produced parameterizations that were more consistent with available experimental knowledge. Despite this difference between them, the two approaches proved to be equivalent under the climatic conditions present, but not when tested on projected scenarios of climate change over the period 1990–2090, where cutting-calibrated models, which are more sensitive to decreasing winter length, predicted higher variations of the budburst dates.