Targeted observations for Typhoon Maria (2018) using the Geostationary Interferometric Infrared Sounder (GIIRS) at a temporal resolution of 15 min provide unprecedented information for impact studies of high‐temporal‐resolution from geostationary hyperspectral IR sounders. This study investigates the impacts of different temporal resolutions for GIIRS assimilation on Maria analyses and forecasts. Results reveal that assimilating higher resolution data captures more detailed temporal and spatial structures and stronger warm anomalies. Additionally, the track forecast for Maria from assimilating higher temporal resolution GIIRS radiances is better than those from assimilating radiances with lower temporal resolution, and both are better than the control experiment. The accuracy of the different temporal resolutions GIIRS experiments (from 3 h to 15 min) is improved (from 18% to 43%), and the intensity forecast errors are also reduced (from 12% to 18%). Meanwhile, high‐temporal GIIRS assimilation also improves the coastal precipitation forecasts during typhoon landfall.

The theory of classical variational assimilation assumes that biases are unbiased Gaussian. This article investigates the bias characteristic estimate and the bias correction of Geostationary Interferometric Infrared Sounder (GIIRS) on board the FY‐4A satellite. Quality‐control procedures for GIIRS long‐wave temperature channels brightness temperature include cloud detection based on the Advanced Geosynchronous Radiation Imager (AGRI) and outlier removal. The mean biases for most channels are within ±2 K after quality control except for the contaminated channels. Statistical evaluation of the first‐guess departures of GIIRS observations in GRAPES (Global/Regional Assimilation and Prediction Enhanced System) global 4D‐Var reveal that biases for the long‐wave temperature channels depend on fields of view (FOVs) and latitudinal distribution. Additionally, the diurnal variation of biases is obvious only for the upper tropospheric channels, and the biases for high tropospheric channels are smaller than the biases for low tropospheric channels. Finally, off‐line bias correction that was used in this study accounts for the field‐of‐regard (FOR) dependence and the diurnal variation bias characteristics of GIIRS. After bias correction, the results show that biases of long‐wave temperature sounding channels are reduced to ±0.02 K, and the standard deviations are less than 1 K except for the contaminated channels. The probability density function of the differences between observations and simulations for some common assimilation channels is closer to the unbiased Gaussian distribution.

Last spring, super dust storms reappeared in East Asia after being absent for one and a half decades. The event caused enormous losses in both Mongolia and China. Accurate simulation of such super sandstorms is valuable for the quantification of health damage, aviation risks, and profound impacts on the Earth system, but also to reveal the climatic driving force and the process of desertification. However, accurate simulation of dust life cycles is challenging, mainly due to imperfect knowledge of emissions. In this study, the emissions that lead to the 2021 spring dust storms are estimated through assimilation of MODIS AOD and ground-based PM₁₀ concentration data simultaneously. With this, the dust concentrations during these super storms could be reproduced and validated with concentration observations. The multi-observation assimilation is also compared against emission inversion that assimilates AOD or PM₁₀ concentration measurements alone, and the added values are analyzed. The emission inversion results reveal that wind-blown dust emissions originated from both China and Mongolia during spring 2021. Specifically, 19.9×10⁶ and 37.5×10⁶ t of particles were released in the Chinese and Mongolian Gobi, respectively, during these severe dust events. By source apportionment it was revealed that the Mongolian Gobi poses more severe threats to the densely populated regions of the Fenwei Plain (FWP) and the North China Plain (NCP) located in northern China than does the Chinese Gobi. It was estimated that 63 % of the dust deposited in FWP was due to transnational transport from Mongolia. For NCP, the long-distance transport dust from Mongolia contributes about 69 % to the dust deposition.

Recently Transformer and Convolution neural network (CNN) based models have shown promising results in Automatic Speech Recognition (ASR), outperforming Recurrent neural networks (RNNs). Transformer models are good at capturing content-based global interactions, while CNNs exploit local features effectively. In this work, we achieve the best of both worlds by studying how to combine convolution neural networks and transformers to model both local and global dependencies of an audio sequence in a parameter-efficient way. To this regard, we propose the convolution-augmented transformer for speech recognition, named Conformer. Conformer significantly outperforms the previous Transformer and CNN based models achieving state-of-the-art accuracies. On the widely used LibriSpeech benchmark, our model achieves WER of 2.1%/4.3% without using a language model and 1.9%/3.9% with an external language model on test/testother. We also observe competitive performance of 2.7%/6.3% with a small model of only 10M parameters.

Four‐dimensional (4D) wind fields were derived from radiance measurements of the Geosynchronous Interferometric Infrared Sounder (GIIRS) onboard the FengYun‐4A satellite with 15‐min temporal resolution during Typhoon Maria (2018). Results are evaluated with independent ERA5 reanalysis, Global Data Assimilation System (GDAS) analysis and dropsonde wind profiles, and show a statistical root mean squared error less than 2 m/s for U and V components in troposphere against ERA5 and GDAS. The temporal variation of the wind fields from GIIRS at 15‐min intervals is consistent with that of the hourly ERA5. The added value of wind profiles over the numerical weather predictions (NWP) background field is also revealed. Further experiments confirm that higher temporal resolution from geostationary infrared (IR) sounder measurements could provide better dynamic information. 4D dynamic information can be extracted from high temporal resolution geostationary hyperspectral IR radiances in a consistent and continuous manner that can be used together with the thermodynamic information for various quantitative applications such as NWP data assimilation, near real‐time weather monitoring, situational awareness and nowcasting.

Aerosol optical depths (AODs) from the new Himawari‐8 satellite instrument have been assimilated in a dust simulation model over East Asia. This advanced geostationary instrument is capable of monitoring the East Asian dust storms which usually have great spatial and temporal variability. The quality of the data has been verified through a comparison with AErosol RObotic NETwork AODs. This study focuses on extreme dust events only when dust aerosols are dominant; promising results are obtained in AOD assimilation experiments during a case in May 2017. The dust emission fields that drive the simulation model are strongly improved by the inverse modeling, and consequently, the simulated dust concentrations are in better agreements with the observed AOD as well as ground‐based observations of PM₁₀. However, some satellite AODs show significant inconsistence with the simulations and the PM₁₀ and AErosol RObotic NETwork observations, which might arise from retrieval errors over a partially clouded scene. The data assimilation procedure therefore includes a screening method to exclude these observations in order to avoid unrealistic results. A dust mask screening method is designed, which is based on selecting only those observations where the deterministic model produces a substantial amount of dust. This screen algorithm is tested to give more accurate result compared to the traditional method based on background covariance in the case study. Note that our screen method would exclude valuable information in case the model is not able to simulate the dust plume shape correctly; hence, applications in related studies require inspections of simulations and observations by user.