由旱至灾的过程存在一个临界态, 为辨识干旱致灾临界态, 提出了一种基于主成分分析和支持向量机的干旱 致灾临界态的辨识方法。通过主成分分析对反映干旱持续时间、严重程度和极值特征的多个指标进行降维处理, 剔除指标间相关性引起的冗余信息, 再进一步结合当地的历史旱灾记录, 基于支持向量机的分类原理寻求最优分类平面, 对潜在的干旱样本是否致灾进行分类, 由此确定的分类平面就是干旱致灾临界面。该临界面可以直观地揭示干旱的发展过程及趋势, 便于在干旱预警中推广应用。以汉江石泉以上流域为例, 按照一定原则共选取 107 个潜在干旱样本, 并结合历史旱灾记录, 采用上述方法对潜在干旱样本中的致灾样本进行了辨识, 率定期和验证期的准确率分别达到 88.6% 和 78.6% , 识别精度较高。

There is a critical stage in the process from drought to disaster. In order to identify the critical state of drought disaster, a metho d based on principal component a nalysis and support vector machine is proposed. The principal component analysis method is used tor educe the dimensionality of severa lindexes that can reflect the duration, severity , and extremum of drought, and to remove the r edundant information of the correlated multiple indicators. The support vector machine is used to find the optimal classification plane based on the historical drought records. The potential drought samples are classified into drought samples a nd disaster samples based o n this classification plane. This classification plane can be phenom enologically defined as the critical state plane of drought-disa stert ransition. The development process and trend of drought can berevea led more intuitively , which is convenient to popularize and a pply to assess drought warnings. The sub-basin above Shiquan in the Hanjiang River basin is selected as a case study , and 1 07 po tential drought samples are selected according to certain principles. The above-proposed method is used to identify the disaster samples according to the historical drought records, with accuracies of 8 8.6% and 78.6 % in calibration and validation periods, respectively .

国家重点研发计划项目( 2016YFC0402708) ;中央高校基本科研业务费资助( HUST : 2017KFYXJJ195)