Bayesian comparison of six different temperature-based budburst models for four temperate tree species

Abstract

Phenology models are tools to analyze changes in the timing and duration of the growing season. During the past three decades different budburst models have been developed, but, so far, no consensus model has been found to accurately predict budburst date across different tree species. The aim of this study was to estimate the performance of six different temperature-driven models of leaf budburst (thermal time, thermal period fixed, sequential, parallel, alternating, unified) for four temperate tree species in Belgium (birch, chestnut, oak, beech). The models were parameterized using a Bayesian approach. The performance of these models was compared using Bayesian model comparison (BMC) and root mean square error (RMSE).Model comparison showed that the two models that do not include a calculation of chilling requirement were the best for the studied four tree species. The Sequential model (SM) was the third most plausible model for predicting budburst, having a higher probability to be correct than the other two-phase models combining a chilling phase with a forcing phase. This suggested that in our budburst observation dataset, the chilling requirement was probably always fulfilled, making the date of budburst controlled by forcing temperature. We cannot rule out that in warmer regions or future warmer conditions, chilling may become insufficient and a sequential pattern of chilling and forcing may become most appropriate to simulate budburst date. Parameter analysis showed that the last month prior to budburst had the greatest impact on determining the date of bud opening in the case of birch and chestnut, whereas the last 3 months were the main determinants for oak and beech, the two later flushing species. Validation showed that the models that fitted the parameterization data well had much poorer performance when tested with independent data. This indicates that other factors (e.g. photoperiod) might affect the budburst process and/or model parameterization (determining the sensitivity of budburst to temperature) substantially change between different localities.