关键词: 苹果, 始花期, 随机森林, 预报模型, 冷/热积量

Abstract: Three major production areas of Fuji apple, Linyi (in Shanxi province), Luochuan (in Shaanxi province) and Qixia (in Shandong province) were selected as the study region, based on 1 km gridded meteorological data, actual first flowering date data, and Chilling Hour Model (CHM) and Growing Degree Hour (GDH) data of the survey sample points in 2019−2020, the optimal chill/heat requirement at the first flowering date of apple was obtained using the gridded search method. Then, the daily air temperature characteristic values (Tmax, Tmin and Tavg) were divided into seven daily air temperature characteristic factor combinations, including single factor, double factors and triple factors, and the random forest algorithm (RF) was used to establish three regional daily chill/heat accumulation models with different daily air temperature characteristic factor combinations to select the optimal daily air temperature characteristic factor. On the basis of which, forecasting model of apple first flowering date was established based on the optimal daily air temperature characteristic factor by using RF algorithm, and the accuracy of the forecasting model was evaluated by independent actual first flowering date data. The results showed that: (1) the optimal chill/heat requirement at the first flowering date for apple in three regions were 730CH and 7350GDH in Linyi, 345CH and 4950GDH in Luochuan, and 520CH and 4450GDH in Qixia. (2) Among the seven combinations of daily air temperature characteristics, the three regional daily chill/heat accumulation models with the combination of Tmax, Tmin and Tavg had high accuracy in estimating daily chill/heat accumulation, and the RMSE between the estimated daily chill accumulation and the daily chill accumulation obtained from the CHM model was 0.97−2.50CH, and the RMSE between the estimated daily heat accumulation and the daily heat accumulation obtained from the GDH model was 1.73−15.76GDH. (3) When the daily chill/heat accumulation was estimated by forecast model of apple first flowering date, the RMSE between the estimated daily chill accumulation and the daily chill accumulation based on the CHM model ranged from 1.08 to 1.14CH, and the RMSE between the estimated daily heat accumulation and the daily heat accumulation based on the GDH model ranged from 2.03 to 3.74GDH. When the model was used to forecast first flowering date of apple, R2 between the predicted and actual first flowering date was 0.92, and RMSE was 3.44d. The accuracy of the predicted first flowering date based on daily air temperature characteristic values was in overall agreement with that based on real hourly air temperature data, it indicated that the forecast model of apple flowering date established in this paper could effectively convert the input air temperature data from hourly scale to daily scale, which will have good application value and potential in the subsequent work on apple first flowering date forecasting.

Key words: Apple, First flowering date, Random forest, Forecast model, Chill/heat accumulation