中国北方地区旱涝的年代际预测分析研究

On Study of Interdecadal Prediction of Drought/Flood in Northern China

基于状态空间重构理论和嵌入定理,给出场时间序列预测模型的建立思路。与单点时间序列预测分析方法相比,场时间序列预测分析方法的优点在于,在寻找吸引子上某个相点的最邻近点及其映象以建立预测模型时,不局限于它自身的时间序列,而是在区域内所有相点的时间序列所构成的整个吸引子上寻找。这样,在一定程度上改进单点时间序列的“遍历性”,以提高预测精度。在此基础上,利用中国北方地区534年旱涝等级资料,对中国北方几个区域年代际尺度的旱涝变化及其极端旱(涝)出现频率进行预测试验分析。

The most of time series from the real world, especially those representing climate processes, are too short to satisfy the ideal data length requirement. Too short a history cannot give a full description for the state distribution of the dynamic system. This difficulty is usually referred to as the ＂data bottleneck＂ problem for short time series analysis. In order to solve the above problem, some atmosphere scientists have suggested the reconstruction of the dynamic system with observation data from different spatial positions. They have once applied this idea to estimate the dimension of climate attractors and achieved some successes. In this study, a technique of spatial-temporal series analysis prediction model is presented based on reconstruction the state space theory and embedding theory. By means of this technique, some prediction experiments for the decadal-scale drought and flood and their extreme frequencies are carried out in this paper according to the data of 534 years of drought/flood series in northern China. Comparing with the single-variable time series, the selection range of the nearest neighbor set is not only limited in the subtrajectory in which the current point lies, but also extended to the whole attractor. This technique can be used to a certain extent to improve the ergodicity of the single-variable time series. Northern China~ Inter-