Abstract: Vegetation phenology is a powerful indicator of the response of terrestrial ecosystems to climate change and plays an important role in water, carbon exchange and energy balance. Under the background of global warming, the changes of vegetation phenology and simulating the phenological phase of plants have been paid much attention. Up to date, the performance of most phenolgy models on spring phenolgy is better, but the simulated autumn phenology has been less accurate. In this study, meteorological data and surface phenological data (leaf bud opening date, first leaf date, first flowering date and leaf coloring date) at 10 sites were selected as the input of the SW, Unichill and DNGDD models, which were used to simulate the phenological phase of Robinia pseudoacacia(L.). The objectives were to examine the performace of SW, Unichill and DNGDD models simulating spring and autumn phenology, and give a guide on optimum the parameters of phenology models. The odd-year data of the sites were used as the internal test, and the even-year data were used as the cross test. The simulation values of three models were compared with the observed ones for analysis to find the best one to predict vegetation phenology. The simulated annealing algorithm was applied to optimum the parameters of SW, Unichill and DNGDD models. The simulated spring and autumn phenology was compared with the measured one, and evaluate the modelling to select the best one to predict vegetation phenology.The results showed that the leaf bud opening, the first leaf, the first flowering dates of Robinia pseudoacacia (L.) were significantly negatively correlated with average temperature in the same period. Average temperature was the main factor affecting phenology. The phenological period in spring had the strongest correlation with mean air temperature, but not significant correlation with mean daily minimum temperature and mean daily maximum temperature. Therefore, compared with Unichill and DNGDD models, SW performed better in simulating spring phenology. The variance interpretation (R2) of cross test of SW model simulating the leaf bud opening, the first leaf and the first flowering dates were 0.807, 0.876 and 0.907 separately and the root mean square error (RMSE) 6.0, 4.6 and 4.4 days. Mean temperature in the daytime and night mean temperature in autumn had different influences on phenology. There was a large deviation when leaf coloring date was simulated only using average temperature. Average temperature was replaced by daily minmum and maximum temperature in DNGDD model. DNGDD model performed well in comparison with SW model when the leaf coloring date in autumn was simulated, and R2 of cross check was 0.580. Therefore, SW model is the best one to simulat the spring phenological phases, while DNGDD model perforemed well in simulating the autumn phenological phase.

Key words: Robinia pseudoacacia, Phenology model, Spring phenology, Autumn phenology, Integrated temperature, Simulated annealing