中文摘要：

数字滤波初始化应用的前提假设之一是不平衡重力惯性波在频率谱上与有天气意义的波动不重叠。随着水平分辨率的不断提高,数值模式能够解析的重力波具有波长更短、频率更高的性质,模式中有天气意义的波动与高频噪声的界线更加模糊,因此,数字滤波初始化在高分辨率模式中的应用性能尚未明确。以一次深对流过程为例,在3 km水平分辨率条件下采用不同滤波参数配置进行数字滤波初始化试验,试图对高水平分辨率预报初始场中高频噪声特征、如何区分具有天气意义的高频信号和虚假噪声、数字滤波初始化在数值预报模式（WRF）高水平分辨率条件下的应用性能等一系列相关问题给出初步的认识。试验结果表明,在针对高分辨率模式进行数字滤波初始化时,原有应用于低分辨率模式的初始化配置参数已不适用,必须根据高分辨率噪声的特点重新设计相应的滤波方案。即使在＂无噪声＂状态下,有天气意义的信息在模式中也具有与虚假高频噪声一致的表现形式,即大的地面气压变化倾向。数字滤波初始化的确能对特定切断频率的高频振荡进行滤波,但无法区分其性质究竟是具有天气意义的高频信号,还仅仅只是虚假的高频噪声。如果要达到滤除初始场中高频噪声的理想效果、并且消除积分过程中的高频振荡,必须延长滤波时间窗,但这是以增加额外的积分时间、以及有可能损害气象上有意义的波动为代价的。滤波效果越显著,滤波结果对初始场的改变越大,这两个方面实质上是有冲突的,因此,在应用数字滤波初始化时必须在滤波效果可接受和保证预报性能两方面之间进行妥协。最后就数字滤波初始化在高分辨率有限区域数值模式中的可能应用提出了具体的建议。

英文摘要：

The rationale to use the digital filter initialization（DFI） relies on the assumption that the fastest unbalanced inertial gravity waves propagate quasi-linearly within the balanced background flow and consequently can simply be filtered as long as there is no overlap with the meteorologically significant part of the spectral frequencies.For the numerical forecast with higher horizontal resolution,the inertial gravity waves with shorter wave length and higher frequency can be resolved by the numerical model.Therefore,it＇s difficult to identify the fake high-frequency oscillations from the meso-scale weather signals.In this context,the performance of the DFI for high-resolution forecasts has been at stake for both the research and operational implementation.In this paper,a preliminary investigation into the noise control and initialization for high resolution forecasts are conducted.It＇s revealed that even for a ＇noise-free＇ forecast,the weather-related information displays a large and unreasonable surface pressure tendency,the same as the fake high frequency oscillation.A challenging issue is the introduction by the 3DVAR of high-frequency fake oscillations.Although the DFI is an efficient noise controlling approach,it is found to be impossible to remove the fake oscillaiton introduced by the 3DVAR while ensuring the meteorologically significant signals be kept in the initial conditions.An ideal noise controlling effect can be reached by extending filtering windows with cost of more additional computation time and the possible damage to weather-related signals.Therefore,compromise should be made between the acceptable filtering effect and forecast quality when implementing DFI.Finally,some suggestions relating to the implementation of the DFI in limited-area high resolution numerical forecasts are given.