基于贝叶斯网络的小麦条锈病预测研究

目的]在地理信息系统GIS的平台上,将不确定性推理方法——贝叶斯网络引入病害预测,基于关键气象因子（温度、降水、湿度、日照）构建一个用于预测小麦条锈病发生概率的贝叶斯网络模型.方法]采用预测日前7d的气象数据预测自预测日起7d内的条锈病发病概率,并对我国小麦条锈病重要流行区域——甘肃省东南部地区2010 ～ 2012年病害发生情况进行预测.结果]模型在返青期至乳熟期输出的病害发生概率与实际调查结果吻合度分别为62.92％、63.18％、79.48％、94.75％,能够较客观地反映病害发生的时间规律和空间分布特点.结论]该研究表明将贝叶斯网络和GIS分析结合在较大的空间范围内利用关键气象因子进行小麦条锈病短期预测是一种可行的途径.

A Bayesian Network Model for Prediction of Stripe Rust in Winter Wheat

Objective] Based on the geographic information system （GIS） platform,Bayesian network,an uncertainty reasoning method,was introduced for disease prediction.In this study,a Bayesian network was constructed based on key meteorological factors （temperature,precipitation,relative humidity and sunshine duration） to predict the occurrence probability of wheat stripe rust.Method] Meteorological data during the recent seven days were used to analyze the occurrence probability of wheat stripe rust and to predict the occurrence probability of wheat stripe rust in the southeast part of Gansu Province during 2010-2012.Result] According to the prediction and actual survey results of the occurrence probability of wheat stripe rust from seedling stage to milk-ripe stage,the accuracy rate of the constructed prediction model was 62.92％,63.18％,79.48％ and 94.75％ respectively,which objectively revealed the temporal and spatial distribution characteristics of wheat stripe rust.Conclusion] Combining Bayesian network and GIS analysis could provide a reasonable means for short-term prediction of wheat stripe rust based on key meteorological factors within a relatively large spatial scope.