首页分享浙江水蜜桃成熟期集合预报模型

浙江水蜜桃成熟期集合预报模型

来源：花匠小妙招时间：2024-12-10 00:51

摘要： 物候期预报多基于单一性模型，预报结果的精度和稳定度较差，难以实现业务应用。本文利用2005?2017年宁波水蜜桃主产区收集的物候期和气象资料，构建不同时间尺度（小时、日、候、旬、月）和不同时间起点（固定日期、物候期）下的成熟期预报模型集，探究集合模型在物候期精细化预报中的可行性。基于模型预报结果的精度和稳定度，分别利用算术平均法、回归系数法、相关系数法和绝对误差法确定集合预报模型成员的权重，构建不同预报时效的水蜜桃成熟期加权集合模型。结果表明：利用4种权重系数确定方法构建的加权集合模型均保持了较高的精度和稳定度，集合模型回代检验的绝对误差AE平均仅0.69（在0.56～0.87）d，均方根误差RMSE平均为0.90（在0.69～1.14）d，相关系数R平均达0.95（在0.92～0.98）；相比单一模型，集合模型的AE和RMSE分别缩小0.5d和0.6d，R值提高0.12。基于绝对误差法构建的加权集合模型效果最佳，回代检验AE和RMSE平均值分别为0.66d和0.88d，对宁波水蜜桃主产区成熟期预报的AE≤2d。集合模型中适当融合硬核期观测能将预报误差缩小0.2～0.3d。集合预报模型为物候期精细化预报提供了新的思路，能够满足业务应用需求。

关键词: 物候期, 物候期预报, 集合模型, 预报时效, 权重确定, 气候变化

Abstract: The phenological forecast is mostly based on the single model which is with poor accuracy and stability. It is difficult to realize the business application. Taking the mature stage of peach as an example, the feasibility of the collection model in the refined prediction of phenological period was explored. By using the phenological data and meteorological data collected from major peach producing areas in Ningbo from 2005 to 2017, the maturity prediction models of peach with different time scale (hours, days, 5-days, 10-days and month) and multiple starting point (phenological period, fixed date) were built. The weighted sum method was employed to construct the collection models with different forecast aging for maturity forecasting. The weights of ensemble forecasting members were determined by using arithmetic mean method, regression coefficient method, correlation coefficient method and absolute error method based on the accuracy and stability of the model prediction results. The results showed that: the collection models, constructed with four kinds of weight coefficient determination methods, were with high accuracy and stability. The AE (absolute error) of the collection models’ regression test was only 0.69 (0.56-0.87) days, RMSE (root mean square error) was 0.90 (0.69-1.14) days, R (correlation coefficient) was 0.95 (0.92-0.98). Compared with the single model, the AE and RMSE of the collection model were 0.5 days and 0.6 days lower, and R was 0.12 higher. The collection model based on the absolute error method was the best, the average value of AE and RMSE by back-generation test were 0.66 and 0.88 days, respectively. The AE of maturity forecasting for main peach producing area in Ningbo was less than 2 days. The prediction error could be reduced by 0.2-0.3 days by the fusion of stone hardening observation in the collection model. We suggest that collection model provide a good proxy for fine phenological prediction and can meet the needs of business applications.

Key words: Phenological, Phenoloyical forecast, Collection model, Period of validity, Weight determination, Climate change