慢特征分析法在气象上的应用进展

Progress of Slow Feature Analysis Method Applied on Meteorological Research

气候系统具有非平稳特征,根本原因在于其外强迫随时间发生改变,因此外部驱动力的分析对于理解气候系统的动力学特征至关重要,而如何有效提取系统外部驱动信息是一个亟待解决的前沿科学问题。最近几年,在生物神经学领域中应用的一种提取非平稳信号中外强迫信息的方法——慢特征分析法(Slow Feature Analysis,SFA),在气象领域中也得到了初步成功的尝试,结果显示出此方法对气候系统的外强迫信息分析及有关动力学机制的探究有较好的应用前景。本文主要介绍SFA方法的理论思想及实施步骤,并通过一个理想的非平稳时间序列检验其提取外强迫信息的能力,结果证明在衰减的Logistic模型中,可利用SFA算法提取出模型中的外强迫,且与真实外强迫的相关系数可达0.99;此外,还介绍将该方法应用于Arosa臭氧时间序列,分析其提取的外强迫信息的动力学特征;并介绍了在气候时间序列建模中引入外强迫因子的预测效果。

Climate system has non-stationary characteristics,because its external force changing with time. So detecting the external driving force is important to reveal the dynamic mechanism of climate system. However,how to extract these external driving factors effectively is a challenge. In recent years,a technique for extracting external driving force in non-stationary system used in biological neural field,called Slow Feature Analysis（ SFA）,is used on meteorological research and makes some advances. The results show that： it has good application prospect in analyzing the external driving force of climate system and relevant dynamic mechanism. In this paper,we introduce the theory SFA and its process. Furthermore,an ideal non-stationary time series is built to test the extracting external driving force ability of SFA. The testing result indicates that： the external driving force can be extracted and its correlation coefficient with the real one is 0. 99 in decreasing Logistic model. In addition,we review the progress in meteorological application： one is in Arosa ozone time series to analyze dynamic characteristics of the extracted external driving force,the other shows the forecasting effect in an improved climatic prediction model through SFA.