[目的]在测土配方施肥中,“测土”是实现合理施肥微观指导的关键技术手段.本研究针对高度分散经营的耕地代表性土样采集难、测试费用高、花费时间长等“测土”难题,探讨早稻氮磷钾施肥类别归属的统计模式识别技术.[方法]利用闽东南80个早稻氮磷钾田间肥效试验资料作为样本,采用欧氏距离-离差平方和法进行系统聚类分析,建立区域早稻施肥类别.在此基础上,根据田块氮磷钾施肥量及其产量结合地力信息等数量指标,应用贝叶斯判别准则探讨具体田块的施肥类别归属.[结果]在确保空白区产量和平衡施肥产量均值在两两施肥类别间具有统计显著差异的前提下,闽东南早稻最多可划分为6个氮磷钾施肥类别.多元统计检验表明,6个施肥类别的类内协方差矩阵差异不全相等,据此建立了各个施肥类别的贝叶斯判别函数.结果表明,原始试验数据标准化处理可大幅度提高判别准确性,监测样本的回代误判率和交叉误判率分别只有1.2％和1.3％,预留6个试验点的84个处理的氮、磷、钾施肥量和产量试验资料的施肥类别归属的判别正确率达到81.0％.[结论]统计模式识别原理及其贝叶斯判别准则可用于早稻氮磷钾施肥类别归属决策,判别准确性能够满足推荐施肥的精度要求.%[Objectives] In soil testing and formulated fertilization,soil testing is the key to realize rational fertilization through practical guidance.However,collection of representative soil samples is often difficult because of highly decentralized farmland management,and soil sample analysis is cost and time-consuming.Therefore,statistical pattern recognition techniques were studied in this paper to explore category attribution of nitrogen (N),phosphorus (P) and potassium (K) fertilization without soil testing for early rice.[Methods] Data from eighty field experiments in southeast of Fujian Province,China,were used in this study.Based on the response of early rice to N,P and K fertilizers,the paddy fields were divided into regional fertilization categories,using clustering analysis method of Euclideana distance-sum of squares of deviations.Then the NPK fertilization category for a field was calculated based on the statistical pattern recognition principle,the application rates of N,P and K fertilizers and the outputs.[Results] On condition of ensuring that the average yield of the blank area and that of balanced fertilization had statistical significant differences between any two fertilization categories,the 80 paddy fields were divided clearly into six fertilization categories.Multivariate statistics showed that the differences of covariance matrix of six categories were not all equal,and the Bayesian discrimination function of each category was established based on that.The standardization of the original data greatly improved the discrimination accuracy,with the back substitution misjudgment rate and cross misjudgment rate of training samples of only 1.2％ and 1.3％ respectively,and the category discrimination accuracy of the 84 treatments in the 6 reserved experimental sites reached 81.0％.[Conclusions] Statistical pattern recognition principle and its Bayesian discriminate analysis method may provide an effective technical approach for the attribution decisions of N,P and K fertilization category of early rice without soil testing,and its results could meet the accuracy requirements of fertilizer recommendation.
展开▼
