基于偏最小二乘法的前汛期月降水量预测方法

采用偏最小二乘回归建立了前汛期(4-6月)月降水量的预测模型,其中模型的输入因子是通过对3个前期月平均物理量场(海温场、500 hPa温度场和200 hPa高度场)大量的场相关因子采用系统降维的处理方法获得.为实现同时对多个站点的月降水量预测,将多站点的月降水量预测转换成多站点气候场的主分量预测,进一步利用气候场特征向量的近似不变性进行回算,从而得到多站点的逐站月降水量预测结果.对广西37个基本站的前汛期月降水量进行了6年独立样本检验,其预报结果显示该模型具有较好的预报能力.

PLS Based Prediction of Monthly Precipitation in Early Rainy Season

A monthly early rainy season precipitation prediction model is devised by means of the partial least squares regression (PLS) method. The input factors of the model are selected from a large quantity of preceding period high correlation factors by using the empirical orthogonal function (EOF) method. The model converts the prediction of multi site monthly precipitation to that of the principal component of that field. According to the approximate invariability of eigenvectors of climate fields, the return computation is conducted to get the monthly precipitation forecasts of more than one site, together with the principal component predicted by the PLS model. A 6 year independent sample test is carried out on the monthly early rainy season precipitation prediction for 37 basic stations in Guangxi. The results show that the model has good forecasting ability.