苏畅 户雪敏 胡洋山 马丽杰 商文静 冯小军 胡小平

摘要 为提高陕西省小麦条锈病发生面积的预测准确度，以2010年-2018年陕西省小麦条锈菌冬繁区和越冬区的发生县区数、发生面积、温度和降雨量为数据集，通过Pearson相关性分析筛选病害流行的主要影响因子，利用全子集回归筛选病害流行的因子集。以筛选得到的影响病害流行的5个因子，即累计发生县区数、冬繁区条锈病发生面积、1月平均温度、1月平均降雨量和3月平均降雨量为自变量，采用全子集回归和BP神经网络算法开展小麦条锈病发生面积的預测研究。结果表明，全子集回归和BP神经网络算法对2019年-2020年的小麦条锈病发生面积预测准确度均达90%以上，预测2021年陕西省小麦条锈病发生面积分别为46.11万hm2和52.85万hm2。

关键词 小麦条锈病;发生面积;全子集回归;BP神经网络算法

中图分类号： S435.121;S431

文献标识码： A

DOI： 10.16688/j.zwbh.2021218

Abstract In order to improve the prediction accuracy of the occurrence area of wheat stripe rust in Shaanxi province， the number of counties， the occurrence area， temperature， and rainfall in the winter propagation and overwintering regions of wheat stripe rust from 2010 to 2018 were used as the data set to construct prediction model. Relevant factors were screened through Pearson correlation analysis， and factor sets were screened by full subsets regression. Taking the five screened factors， including the total number of counties with stripe rust， the occurrence area of stripe rust in winter propagation， the monthly mean temperature and mean rainfall in January， and the mean rainfall in March as independent variables， the predicted occurrence area of wheat stripe rust was carried out using full subsets regression and BP neural network. The results showed that the prediction accuracy of full subsets regression and BP neural network on the wheat stripe rust occurrence area in 2019-2020 were both over 90%， and the predicted occurrence area of wheat stripe rust in Shaanxi province in 2021 were 461 100 and 528 500 hm2， respectively.

Key words wheat stripe rust;occurrence area;full subsets regression;BP neural network

由条形柄锈菌Puccinia striiformis f.sp. tritici引起的小麦条锈病是世界上重要的流行性真菌病害[1-4]，属于我国一类农作物病害[5]，严重影响小麦的产量和品质[6]。我国是世界上条锈病流行面积最大的国家[7-8]，1950年、1964年、1990年、2002年和2017年我国先后发生5次小麦条锈病大流行，发病面积333万～667万hm2，共造成小麦减产138亿kg[9-13]。

陕西省陕南地区汉中、安康和商洛以及关中地区宝鸡、咸阳、西安和渭南等均为小麦条锈病的常发流行区，其中陕南地区是小麦条锈病菌冬繁区[13]，关中是小麦条锈菌越冬区[14-16]。建立准确率高的小麦条锈病预测模型对保障陕西小麦的丰产稳产至关重要。……