Abstract:

Accurate prediction of the initial flowering date (IFD) of cherry blossoms holds significant importance for tourism planning and park management. It serves as a scientific basis for park authorities to implement protective measures in advance. Thus, the Gui'an Cherry Blossom Garden was considered to analyze the IFD of cherry blossoms in combination with the meteorological data from 2013 to 2024. Pearson correlation coefficients were employed to evaluate the relationships between meteorological factors and the IFD, for identification of significant factors from 18 meteorological variables for model construction. Based on the extracted key factors, a BP neural network model and a multiple linear stepwise regression (MLSR) model were developed. The BP neural network underwent six−fold cross−validation, and its results were compared with those of the MLSR model. The IFD results of cherry blossoms in the Gui'an Cherry Blossom Garden exhibited a general trend of "advancement followed by a slight delay," with significant interannual fluctuations. The IFD mainly occurred in the first half of March and was  highly influenced by climate fluctuations, particularly in mid−January and late February. Accumulated temperature (especially low−threshold temperatures, such as ≥3°C) played a critical role in advancing the IFD and was designated as a key reference indicator for prediction. Validation of the established models using RMSE and R² revealed that the BP neural network model significantly outperformed the MLSR model in both accuracy and stability. The BP model achieved an average root mean square error (RMSE) of 0.2636 days, markedly lower than the 2.92 days of the MLSR model. Furthermore, the BP model's coefficient of determination (R²) reached as high as 0.9961 during cross-validation, whereas the R² of the MLSR model was much lower. Overall, the integration of BP neural network model with multiple meteorological factors, represented better nonlinear relationships, thereby providing a robust scientific foundation for park management and tourism planning. Additionally, it offered an effective technical approach for the accurate prediction of the IFD of cherry blossoms.

Key words: Cherry blossom, Initial flowering date prediction, BP neural network, K-fold cross-validation, Multiple linear stepwise regression