基于三维对数线性模型的气象干旱等级预测研究

Meteorological drought grade prediction using three-dimensional log-linear models

采用滦河潘家口水库控制流域21个雨量站1958—2009年逐月降水数据,通过计算各站12个月时间尺度的标准化降水指数(SPI),得到实测干旱等级序列。根据1958—2008年干旱等级转换的观测频数,提出基于三维列联表的对数线性干旱预测模型,并计算干旱等级转换间的优势及其置信区间,以估算干旱等级转换的概率,实现了滦河潘家口水库控制流域预见期为1个月和2个月的气象干旱等级预测。选取2009年的干旱演变过程对预测模型进行了验证。结果表明:干旱等级转换的三维列联表呈现一种强对角线趋势,各站三维对数线性模型的检验显著性水平p均大于检验水准α(0.05),观测频数与期望频数吻合程度高,且预测干旱等级能反映实测干旱等级,模型具有较高的预测精度,可用于滦河潘家口水库控制流域短期干旱预警。

Based on the monthly precipitation data from 21 rain gauges during the period of 1958-2009 in the Panjiakou Reservoir catchment of the Luanhe River Basin, drought class time series were derived from Standardized Precipitation Index （SPI） time series computed in a 12-month time scale. Log-linear models for three-dimensional contingency tables were fitted to the observed frequencies of drought class transitions during the period of 1958-2008. Odds and respective confidence intervals for drought class transitions were also calculated to estimate the drought class transition probabilities. Thus the drought classes with 1 and 2 months lead were predicted and the short term meteorological drought class prediction was achieved in the studied area. The validation of the predictions was performed for the 2009 drought. The results show that the contingency tables of drought class transitions present a strong diagonal tendency and results for all sites show a similar agreement between observed and expected frequencies with all the models presenting a test p-value exceeding the chosen significance level of α=0.05. And results of three-dimensional log-linear modeling present good results when comparing predicted and observed drought classes for those 21 stations. It could be concluded that log-linear prediction of drought class transitions is useful for short term drought warning in the Panjiakou Reservoir catchment.