中文摘要：

随着全球气候变化,人类活动干涉,干旱发生的频率逐年增加,影响范围也不断扩大。对干旱进行有效的预测以提前采取应对措施减少极端天气对社会的影响是极为必要的。由于影响干旱发生的如气候、水文等因素十分复杂,应采用合适的方法预报具有非线性特征的干旱。以气象干旱评判标准SPI划分干旱为三等级,并以前期12个月的降雨与从74项大气环流因子中初步筛选出30项因子共372个因子作为初步筛选集,通过Incnodepurity指数挑选出重要性排在前30的因子作为模型解释变量,采用RF模型对淮河流域21个代表站的1962-2012年各月干旱等级进行分析。以1962-2006年作为模型检验期,2007-2012年作为模型预测期,整体预测平均准确率为73.0%,高于气候系统的天气预报准确率,可在不同区域进行推广应用。

英文摘要：

With the global climate change and human activity intervention, the frequency of drought is increasing year after year, and the scope of influence is also expanding. It is very necessary to find an effective way to predict droughts to adapt measures to reducing the damage to society. Because the mechanisms of drought are complicated, we should use appropriate ways to forecast droughts with nonlinear characteristics. Based on the meteorological drought evaluation standard-SPI, droughts are divided into three levels.Every month drought grade of 21 stations in the Huaihe River Basin of 1962-2012 year are analyzed by using important RF model in the top 30 factors as explanatory variables that depends on Incnodepurity index chosen from a preliminary screening set： previous 12 months＇ rainfall and 30 factors are preliminarily se- lected from 74 hydrological-meteorological circulation factors. This paper uses 1962 to 2006 as a model detection period, 2007 to 2012 as the model prediction period, the overall average prediction accuracy is 73.0%, higher than that of the climate system of weather forecast accuracy, and the model will be applied in different areas in the future.