通过分析中国传统农历节气同水文学之间的密切关系，论证了其在水文预报中应用的科学依据。以农历二十四节气中立冬日作为时间统计的坐标系基准点，利用经Levenberg-Marquart算法的改进BP神经网络，预报了南水北调黄河以北沿线主要城市气温稳定转负日期。首先比较新乡、安阳、邢台、石家庄、保定和北京6个城市1957年-2006年气温稳定转负日相关系数，结果显示农历坐标系下预报目标的相关系数均比公历坐标系的大。表明当以立冬日作为统计坐标的基准点时，气温稳定转负日期的相关性更强。同时将农历和公历分别作为时间坐标系，各预报16组次，对预报结果进行对比发现，农历坐标系下合格组次为13组，而公历坐标系下的合格组次为11组。显然，农历坐标系下预报结果比公历坐标系好，预报精度较高，说明农历中24节气在冬季气温转负日期中应用的探索是可行的。

The relationship between Chinese traditional twenty-four solar terms and hydrology is analyzed and the scientific basis of using solar terms in the hydrological prediction is developed. The Beginning of Winter is selected as the datum point of the statistic date in the prediction. The BP-ANNs model improved by Levenberg-Marquartalgorithm is introduced to predict the dates when temperature becomes negative in the cities of the north of Yellow River. The correlation coefficients of the date when temperature becomes negative in Xinxiang, Anyang, Xingtai, Shijiazhuang city, Baoding, and Beijing from 1957 to 2005 were compared. All of the correlation coefficients in the coordinate system of Chinese lunar calendar are greater than those in the coordinate system of solar calendar, which suggests that the correlation of the date when temperature becomes negative is stronger in the coordinate system of Chinese lunar calendar. In addition, the coordinate system of Chinese lunar calendar and solar calendar were used to predict 16 sets of dates when temperature becomes negative. The results showed that the coordinate system of Chinese lunar calendar predicts 13 sets of acceptable data while the coordinate system of solar calendar predicts 11 sets of acceptable data. Therefore, the application of the traditional twenty-four solar terms in the prediction of date when temperature becomes negative is feasible.

国家自然基金（51179209；51209233）；水利部公益性行业科研专项（201301032）