First and second authors: The New Zealand Institute for Plant & Food Research Limited, Mt Albert Research Centre, Auckland Mail Centre, Auckland 1142, New Zealand; and third author: Tasmanian Institute of Agriculture, University of Tasmania, Private Bag 98, Hobart, Tasmania 7001, Australia.
Phytopathology. 2019 Jan;109(1):84-95. doi: 10.1094/PHYTO-10-17-0357-R. Epub 2018 Nov 19.
Botrytis bunch rot (BBR), caused by Botrytis cinerea, results in serious losses to wine-grape production in some seasons during the preharvest period. In order to predict seasons that are at risk from BBR, datasets consisting of 25 disease, weather and vine phenology variables were aggregated from 101 SiteYears across seven regions and nine growing seasons. Automated analyses were used to compare a range of statistical methods for their ability to predict BBR epidemics, including the Kruskal-Wallis test, logistic regression, receiver operating characteristic analysis, and skill-scores. Variables based on relative humidity and surface-wetness duration were significant and consistent predictors of BBR epidemics across the range of analyses applied. Variables integrating temperature and wetness duration, including the Bacchus and Broome models, also demonstrated high predictive ability; however, they did not outperform their constituent components in all analyses. Automation of data analyses was an effective way to compare a wide range of statistical methods and a large number of variables with minimal user input, following initial code development. Significant time was needed to check input data and software code, but a greater return on investment would occur should the analytical process be applied to new datasets, including those from other pathosystems.
葡萄孢菌果穗腐烂病（Botrytis bunch rot，BBR）由葡萄孢菌引起，在收获前的某些季节会对酿酒葡萄的生产造成严重损失。为了预测易发生 BBR 的季节，从七个地区的 101 个地点年份中汇总了包含 25 个疾病、天气和葡萄藤物候变量的数据集，这些数据集跨越了 9 个生长季节。自动化分析用于比较一系列统计方法预测 BBR 流行的能力，包括 Kruskal-Wallis 检验、逻辑回归、接收者操作特征分析和技能评分。在应用的各种分析中，基于相对湿度和表面湿润持续时间的变量是 BBR 流行的重要且一致的预测因子。综合温度和湿润持续时间的变量，包括 Bacchus 和 Broome 模型，也表现出较高的预测能力；然而，在所有分析中，它们并没有比其组成部分表现更好。数据分析的自动化是一种有效的方法，可以在最小的用户输入下比较广泛的统计方法和大量变量，这是在初始代码开发之后实现的。需要花费大量时间来检查输入数据和软件代码，但如果将分析过程应用于新数据集，包括来自其他病原体系统的数据集，那么投资回报率将会更高。
