基于信噪分离的短期气候预测方法研究

Both the theory of dissipative structures or synergetic theory, has a closed ties between accidental process and inevitable process of mutual transformation, rather than independent. Randomness of atmospheric motion determine that climate is not deterministic motion, stochastic force will make climate change continuously generate new ordered structure and lead to non-stationary phenomena. It is very important to describe effectively random force mechanisms for improving climate predictivity. Due to the lack of understanding about relevant stochastic mechanisms, it is very difficult to simulate stochastic dynamics in the climate models existed, and to solve differential equations based on a complex model, which hinders improving the accuracy of climate prediction. The project intends to study this problem by another way: Firstly, we will separate deterministic component and random component from climate observations; secondly, we will focus on analyzing the mechanism of random factors in a deterministic process using differential strategies, and set up a series of statistical models including stochastic mechanisms. Finally, we plan to get distribution function of the random process and to study on embedding a stochastic mechanism into numerical models, comparating the advantages and disadvantages of each method. With evolutionary algorithms and Monte Carlo methods, this project will carry out a series of numerical experiments based on an ideal model to revise random errors in climate model to improve the models' prediction accuracy.

无论是耗散结构理论还是协同学理论，都证实了偶然过程与必然过程之间的密切联系及相互转化，而非相互独立。大气运动的随机性决定了气候不是确定性运动，随机力将使气候在变化中不断产生新的有序结构，并导致非平稳现象。因此，研究气候的随机强迫机制对提高预测精度十分重要。由于缺乏对气候随机力的认识，加上求解复杂模式的微分方程又十分困难，现有的气候模型均难以模拟随机动力学，这阻碍了预测精度的进一步提高。本项目拟通过另一种途径研究随机问题：首先对观测资料中的确定性分量与随机分量进行分离研究；其次采取区别对待的策略，着重分析随机因子在确定性过程中的作用机制，研究两者有机结合的有效方案。最后，建立随机强迫的分布函数，研究在气候模型中嵌入随机作用机制的可能性，对比不同方案的优劣性。项目将在理想模型的基础上开展一系列数值实验，用演化算法和蒙特卡洛方法分析随机分布规律，发展一套基于随机强迫力的预测方法，以提高预测精度。

本项目提出了基于信噪分离的短期气候预测理论，并对理论的可行性进行了研究与验证。首先，项目研究了动力系统中外强迫噪声的分离问题，发展了基于经验模态（EMD）的噪声分离方案，揭示了动力系统中的外强迫过程可以通过数学方法分离，也证明非线性系统在演化过程中受随机强迫是十分普遍的物理现象。其次，利用DFA方法对被分离的动力系统各平稳变量进行分析，研究噪声源与被分离分量之间的关系，确定信兆分离的有效阈值。然后，为构建数学模型需要，进一步发展了演化算法开展噪声的建模方法，通过典型非线性动力系统（Rössler）的外强迫试验完善了演化建模能力，并利用实际气候资料（热带气旋）验证了建模的效果。最后，根据信兆分离方案及改进的演化算法，构建了与之匹配的预测理论模型。该思路除了可以避免噪声强迫的影响，还能提高预测的针对性与稳定性。通过构建真实情形下复杂系统（ENSO）的预测模型，检验了这一预测思路是完全正确的。所建立的ENSO模型不但大大提高了预测的稳定性，而且延长了时效，实现了24个月的准确趋势预测，效果令人满意，验证了方案的科学性与可行性。