Abstract: The rice yield is composed of panicles, the filled grains per panicle and the weight of 1000 filled grains. The theoretical rice yield is defined by their product. However, sometimes there are considerable discrepancies between theoretical and observed yields. One general model is proposed to approximate yield structure: yield=theoretical yield + linear adjustment + random error. Based on Bayes estimates, yield estimation in one area is obtained by combining estimates for different varieties. Similarly yield estimation covering different areas is given by combining estimates over these areas. Also, the accuracy of yield estimation for one variety with small sample size is expecting to be improved by borrowing strength from other varieties in the same area.The corresponding variation of these estimates is given. Data sets obtained from Shaoxing and Fuyang in Zhejiang province are analyzed as an illustration of the proposed methods.

Key words: theoretical yield, linear adjustment, empirical Bayes estimation, borrow strength, forecasting confidence interval

摘要： 水稻产量是由有效穗数、每穗实粒数和千粒重构成。三者相乘得到的值定义为理论产量，但理论产量和实际产量有时相差甚远。本文提出了拟合产量结构的一般模型：产量＝理论产量＋线性纠正＋随机误差。同时利用经验贝叶斯估计提出广综合一个地区不同品种数据的地区产量估计和综合不同地区的产址估计，对小样本的品种的产量估计可借用同一地区的其它品种资料而得到改进。相应的估计方差也同时给出。作为实例，文章分析了从浙江绍兴和富阳得到的数据。

关键词: 理论产量, 线性纠正, 经验贝叶斯估计, 借用信息, 预测区间