To improve the numerical weather prediction (NWP), more and more studies on data assimilation, model dynamics and physics are carried out, besides, new method different from these studies need to be developed as well. The methodology based on inverse problem to improve NWP performance, started and developed by Chinese scientists, is different from the traditional ways. And all the methods by solving inverse problem to improve NWP presented before are based on accurate past observations, which is not the case for reality. Therefore, the generalized solution of atmosphere equations or NWP model is required to be expanded further. Then new method based on the expanded solution will be developed for operational NWP models improvement. An ‘unknown term’, named ‘model error term’, will introduced in NWP model to represent the model error sources, which can be obtained by solving an inverse problem with past data. The model forecast will be improved due to the model error term. Because the mathematical form of the term is unknown, it will be separated into two parts: past model error term and future model error term. The data sets of past error term will be calculated by iteration approach. To obtain the future model error term, the spectra of model error will be analyzed, and suitable functional will be established. Finally, the future model error term will be solved by minimizing the functional and introduced into NWP model to correct model.
为了提高数值预报准确率,在持续地改进资料同化技术、模式动力框架和物理过程的同时,也需要发展有别于此的新方法。以求解反问题的方式来改进数值预报是我国科学家发展的不同于传统数值预报技术的独特方法。现有相关方法假定过去资料完全准确与实际不符,因此为了改进实际运行的业务数值模式,需要在理论上进一步拓展原有方法。新方法同时考虑了过去资料的误差和模式误差,并将模式误差相关因素假设为模式的一个“未知项”(将其称作模式误差项以区别于误差——预报减观测)。利用过去观测资料反求出该“未知项”,在预报中将该“未知项”加入模式以提高模式预报。本研究计划将该模式误差项分成过去时刻和未来时刻两部分,并利用迭代法求得过去时刻模式误差项的数据集。在基于模式误差分析所得出的时空分布和频谱分布规律基础上,构建相应的泛函极小问题,将过去时刻的模式误差项外推到未来时刻,并将其加入到模式以改进预报。
数值天气预报模式的准确率随着初始场和模式的完善而大幅提高,然而模式误差永远无法彻底消除。因此,本项目在常规模式发展思路以外,利用求解反问题的方式,研发了可在现有业务模式上实现预报改进的新方法。项目假定模式存在一个未知误差项,并将其分为过去时段模式误差项和预报时段模式误差项。项目利用迭代方法获得过去时段每个间隔的模式误差项。根据造成模式误差因素的性质,将模式误差源分解为定常部分和时变部分。针对定常误差源,将模式误差源在短时预报范围内设为常数,基于迭代获取的过去时段各个间隔误差项,使用最小二乘法获得定常模式误差项,并在模式预报时段进行在线订正。针对时变模式误差源,将模式误差源设为傅立叶展式,利用“广义解”泛函极小获得傅里叶展式的系数,并在模式预报时段进行在线订正。本方法分别在理想模式和我国自主研发的新一代多尺度通用资料同化与数值预报系统(GRAPES)中进行了测试,结果显示订正方法能够反演出模式误差源的结构特征,大大地减小了模式的定常误差和时变误差。项目进一步分析了本方法的理论基础,论证了通过在线订正的方式改进模式预报符合“广义解”理论。既保证和拓展了方法的理论基础,也为数值天气预报的改进提供了一个有效的补充方案。
{{i.achievement_title}}
数据更新时间:2023-05-31
监管的非对称性、盈余管理模式选择与证监会执法效率?
主控因素对异型头弹丸半侵彻金属靶深度的影响特性研究
宁南山区植被恢复模式对土壤主要酶活性、微生物多样性及土壤养分的影响
针灸治疗胃食管反流病的研究进展
钢筋混凝土带翼缘剪力墙破坏机理研究
数值天气预报中某些数学问题
热力学第二定律对数值天气预报模式的改进
求解Basis Pursuit问题的数值优化方法
基于代理模型的多宗量粘弹性反问题的数值求解