年降水量统计马尔可夫预测模型及其应用

论文基于大气降水过程存在着大量的不确定性和不精确性的特点,将简单的统计计算与马尔可夫链理论有机结合起来预测降水量,方法的物理概念清晰,计算简便。以杭州市1956—2008年的降水量资料为例,应用统计模型进行逐年降水量预测,从前40 a的降水量序列资料(1956—1995年)开始,预测1996年的降水量,然后剔除1956年的降水资料,将1996年的实际资料加入到序列中,再按照降水量预测的基本步骤预测1997年的降水量,依此类推进行逐年降水量预测,结果表明,13 a中误差小于±10%、±15和±20%的分别占30.77%、53.85%和69.23%,预测误差最大值为-24.03%;应用统计马尔可夫预测模型进行逐年降水量预测,8 a中误差小于±1%和±5%的分别占37.50%和62.50%,预测误差最大值为8.77%,其降水量预测结果的精度有较大的提高,该法为提高大气降水量预测的精度提供了一条值得探索的途径。

Statistic Markovian Model for Predicting of Annual Precipitation

Based on the speciality of uncertainty and inaccuracy of precipitation, both simple statistical calculation and Markov chain theory were used together for predicting the precipitation in this paper. The method was characteristic of clear physical concept and simple calculation. Data of precipitation in Hangzhou, from 1956 to 2008, was used as an example. The precipitation can be predicted year by year using the statistical models;starting from the precipitation sequence data (1956—1995) in the last 40 years, the precipitation in 1996 was predicted, then the precipitation data in 1956 was removed and the actual data in 1996 was added to the sequence, and then in accordance with the basic steps of precipitation forecasts, precipitation in 1997 was predicted, and so on for each year of precipitation forecasts. Results show that the error value smaller than ±10％, ±15% and ±20％ was 30.77%, 53.85% and 69.23% respectively in the 13 years of precipitation prediction, and the maximum error value of the prediction was -24.03％. The precipitation can be predicted year by year using the statistical Markonian models;the error value smaller than ±1％ and ±5％ was 37.50% and 62.50% respectively in the 8 years of precipitation prediction, and the maximum error value of the prediction was 8.77％. The accuracy of precipitation prediction of the statistical Markonian models was improved obviously, hence a practicable method for predicting future precipitation was put forward.