Assimilating satellite retrieved temperature and humidity profiles in cloudy regions remains attractive for numerical weather prediction (NWP). Current operational systems primarily rely on a limited number of radiance channels from hyperspectral sounders. However, compactly retrieved profiles from multiple instruments and all available channels, especially those generated efficiently using machine learning algorithms, present a promising alternative. These retrievals can be produced under both clear and cloudy conditions, making it crucial to understand how cloudiness affects the impact of assimilation. In this study, atmospheric temperature and humidity profiles jointly retrieved from the Cross-track Infrared Sounder (CrIS) and Advanced Technology Microwave Sounder (ATMS) onboard NOAA-20 satellite were assimilated into a regional NWP model using a three-dimensional variational (3D-Var) method. The impacts of different cloud amounts and cloud-top heights on tropical cyclone (TC) forecasts were examined for Typhoon Muifa. Assimilating NOAA Unique Combined Atmospheric Processing System (NUCAPS) profiles improved the thermodynamic structure of the initial field and enhanced typhoon forecasts. The all-sky profile assimilation reduced track forecast errors by up to 200 km after 48 hours, and increased the equitable threat score (ETS) for 72-hour accumulated precipitation by about 74

The Geostationary Environment Monitoring Spectrometer (GEMS), onboard GEO-KOMPSAT-2B (GK-2B), is the world's first ultraviolet-visible hyperspectral sensor from geostationary orbit, it is designed to enhance air pollutants monitoring capability across the Asia-Pacific region with unique capability on monitoring the diurnal variations of aerosol and air pollution gases. This work aims to validate the accuracy of the official aerosol optical properties products including Aerosol Optical Depth (AOD) and Single Scattering Albedo (SSA) retrieved from GEMS and understand the possible causes of retrieval errors. Validation with AERONET at 443 nm shows that about 63 % (Highly Absorbing Fine, HAF), 65 % (Dust), and 58 % (Non-absorbing, NA) of GEMS AOD matchups agree within 30 % expected error (EE). GEMS AOD exhibits underestimation at high aerosol loading (AOD > 0.4) and overestimation at low aerosol loading (AOD < 0.2). For the HAF type, GEMS AOD retrieval accuracy improves in the noon/afternoon compared to morning. However, the temporal variation in retrieval accuracy of Dust and NA types are comparatively less pronounced. Regional performance of GEMS AOD for HAF type performs better in Korea and Beijing, while worse in India and equatorial regions. GEMS SSA presents good performance in Korea and Beijing, with 81.1 %(HAF)/81.3 %(NA) and 72.5 %(HAF)/65.9 %(NA) falling within +/- 0.05 absolute difference, respectively. The limitations of GEMS aerosol classification algorithms based on UV-AI and Visible-AI are a significant cause of errors in AOD and SSA retrieval. Additionally, a priori ALH based on CALIOP ALH climatology also contribute to the AOD retrieval error, particularly for HAF type. These factors should be considered in future algorithm.

Air pollution and climate change are two significant concerns threatening sustainable human development. Tracking and mapping atmospheric pollutants and greenhouse gases are essential to managing these issues. Due to its extensive spatiotemporal coverage, satellite remote sensing is indispensable in Earth observation systems to provide measurements of atmospheric chemical species. Since 2008, China has been gathering global atmospheric chemical composition data from space on a daily basis, and the Fengyun satellite series provides the nation's longest record of ozone and aerosol monitoring. The Medium Resolution Spectral Imager (MERSI) onboard the Fengyun-3 (FY-3) series and the Advanced Geostationary Radiation Imager (AGRI) aboard the Fengyun-4 (FY-4) series satellites monitor aerosols in sun-synchronous and geosynchronous orbits, respectively. The Total Ozone Unit (TOU) and the Solar Backscatter Ultraviolet Sounder (SBUS) are China's first instruments to measure global total ozone columns and vertical profiles. The Ozone Monitoring Suite-Nadir (OMS-N) is the successor to the TOU, monitoring ozone and trace gases like nitrogen dioxide (NO2) and sulfur dioxide (SO2) in the ultraviolet (UV)-visible spectrum. The OMS-Limb (OMS-L) provides the capability to obtain stratospheric profiles of ozone and other species. In the infrared wavelength range, the Hyperspectral Infrared Atmospheric Sounder (HIRAS) retrieves vertical information for ozone as well as other trace gases, including carbon monoxide (CO) and ammonia (NH3). For NH3 and CO, HIRAS in low-Earth orbit maps their global distribution, and the Geostationary Interferometric Infrared Sounder (GIIRS) in geostationary orbit tracks their temporal variation over East Asia. The Greenhouse-Gases Absorption Spectrometer (GAS) onboard FY-3D and FY-3H is designed to detect the greenhouse gases carbon dioxide (CO2) and methane (CH4).

The four-dimensional variational (4D-Var) data assimilation framework conventionally assumes static observation error covariance throughout the assimilation window. This assumption neglects the progressive growth of background error during model integration, potentially leading to underutilization of observations in the later portion of the window. Here, we propose a dynamic observation error adjustment strategy that incorporates time-varying scaling factors to assign greater weights to later-window observations, thereby compensating for background error growth. This approach involves only scalar modifications to the observation error covariance matrix and is, in principle, applicable to all observation types within the 4D-Var framework. We demonstrate its effectiveness by assimilating three-dimensional (3D) wind retrievals from the Geostationary Interferometric Infrared Sounder (GIIRS) onboard the Fengyun-4B (FY-4B) satellite into the China Meteorological Administration Global Forecast System (CMA-GFS), using Typhoon Khanun (2023) as a case study. The standard deviation of observation-minus-background departures exhibits 6-hour periodic fluctuations corresponding to the assimilation window length, providing observational evidence of background error growth that underpins our dynamic adjustment strategy. Relative to fixed observation error experiments, the dynamic adjustment reduces the observation component of the cost function by 7–12% during the later-window period. Assimilation of GIIRS-derived winds substantially improves typhoon track forecasts, with mean track errors reduced by 15.3% and 23.3% in the fixed and dynamic observation error experiments, respectively, compared to the control experiment. The dynamic adjustment strategy further enhances medium-range forecasts of wind fields and geopotential height. This study offers a novel perspective on observation error optimization for 4D-Var assimilation.

Abstract. Accurate spectral transformation across satellite sensors with similar but different spectral response functions (SRFs) are essential for applying the same retrieval algorithms. A novel physics-constrained transfer learning (TL) framework is developed for transferring satellite radiance observations across different sensors while preserving physical consistency. It integrates a core Spectral-Fidelity-Preserving (SFP) model based on extensive radiative transfer simulations, allowing broad adaptability for radiance transformation under diverse satellite observational conditions. Sensitivity experiments demonstrate the robustness of the TL framework relating to radiometric calibration uncertainties, particularly in infrared (IR) channels, and further highlight the critical role of SRF similarity between sensors. Specifically, the scaling factor between the SRFs of the target and reference channels should be constrained within the range of 0.5 – 1.5. Meanwhile, the shift in central wavenumber should remain below 200 cm⁻¹ for visible or near IR channels, and more strictly below 20 cm⁻¹ for infrared window channels (e.g., 10.80 µm). Applying to radiance observations from Fengyun-4A/B (FY-4A/B) geostationary (GEO) satellites explicitly indicates that the TL approach improves retrieval accuracy for key geophysical parameters such as cloud amount profile and quantitative precipitation estimation, when compared those without applying TL. Thus, the TL approach enhances cross-satellite data consistency and provides a practical tool for operational satellite data applications (e.g., adopt algorithms of F-4A to FY-4B without operational interruption).

Accurate cloud detection is essential for the quantitative applications of satellite imager observations, but nighttime cloud detection has challenges due to limited spectral bands, for example, physical methods using only infrared (IR) bands without using spatial textures as input for cloud detection often result in high uncertainties, especially in some situations such as cryosphere surface. Although numerous segmentation-style deep learning cloud detection algorithms have proposed in previous studies, they are inadequate for nighttime due to the difficulty in acquiring two-dimensional truth data for training and validation. To overcome these challenges, the Transformer based Nighttime Cloud Detection (TNCD) framework, which integrates spatial features and utilizes an advanced Transformer architecture with relative position encoding, layer scaling, and channel attention mechanisms, is proposed and investigated for nighttime cloud detection. The model was trained on labels derived from CALIOP data, utilizing a dataset comprising nearly one hundred million segments from MODIS. Independent validation indicates that TNCD achieves robust and consistent performance across various scenarios, with an overall accuracy (OA) of 93.26 % and over 90 % in cryosphere regions. The proposed algorithm avoids the pattern noise appeared in the traditional physical methodology due to the utilization of auxiliary data at coarser resolutions, it also mitigates the negative impact of stripes in IR images for cloud detection. Moreover, TNCD shows high transferable practicability across sensors, with over 90 % OA for MERSI. More importantly, our research underscores the importance of water vapor absorption bands for nighttime cloud detection over the cryosphere. TNCD's high accuracy and robustness provide unique methodology that could be used operationally for nighttime cloud detection.

Reliable precipitation monitoring is essential for disaster risk reduction, water resources management, and agricultural decision-making. Multi-source satellite observations, particularly the combination of geostationary infrared and passive microwave measurements, have become a primary means of precipitation detection. Traditional multi-source satellite precipitation estimation methods remain computationally inefficient, and many deep learning methods lack the flexibility to incorporate new sensors without retraining the full model. Here we introduce PRISMA (Precipitation Inference from Satellite Modalities via generAtive modeling), a plug-and-play latent generative framework for multi-sensor precipitation estimation. PRISMA learns an unconditional precipitation prior from IMERG Final fields and constrains it through independently trained, sensor-specific conditional branches, allowing new observation sources to be incorporated without retraining the generative backbone. Applied to FY-4B AGRI infrared and GPM GMI microwave observations, PRISMA improves Critical Success Index by up to 40.3

High-resolution atmospheric thermodynamic information is essential for resolving localized weather processes in complex urbanized regions such as the Yangtze River Delta (YRD). However, existing global reanalysis products are limited by coarse spatial resolution, constraining their ability to represent fine-scale thermodynamic structures. In this study, we develop YRD1km, a 1-km hourly regional thermodynamic reanalysis dataset for the YRD, generated through dynamical downscaling of ERA5 using the WRF model with a three-level nested configuration. The framework integrates optimized physical parameterizations, a combined observation and analysis nudging strategy, and updated high-resolution land-use information. The current dataset covers the summer seasons (June–August) from 2021 to 2023, a period characterized by frequent convection and strong land–atmosphere interactions. Comprehensive validation against dense surface and radiosonde observations demonstrates that YRD1km consistently outperforms ERA5 in near-surface temperature, relative humidity, and surface pressure, with reductions of approximately 35

Using Volcanic ash transport and dispersion models to provide timely and reliable short-term volcanic ash dispersion forecasts is critical for aviation safety. Both the choice of initialization approach and uncertainties in satellite-derived eruption parameters play a key role in determining forecast accuracy. However, an evaluation of different initialization strategies, together with the propagation of satellite observation uncertainties into forecast errors, remains limited. In this study, the May 2010 Eyjafjallajökull eruption was used as a case to comparatively evaluate different initialization strategies and to quantify the impact of satellite-derived parameter uncertainties on short-term forecasts. Satellite-based initialization and a uniform line source approach were assessed against satellite and lidar observations. Uncertainties in satellite-retrieved cloud top height and mass loading were quantified and propagated into dispersion simulations. Results indicate that forecasts based on uniform line source initialization tend to overestimate spatial coverage and produce false alarms, reflecting sensitivity to the specification of eruption source parameters. Sensitivity analysis shows that plume height and particle size distribution are the dominant sources of uncertainty controlling ash spatial extent, leading to Critical Success Index variations of up to 0.2, while the fine ash emission rate primarily affects mass loading magnitude. Satellite constraints can effectively reduce parameter uncertainties and improve forecast consistency. Bias correction of satellite-derived parameters further improves model performance, reducing RMSE and increasing R2 in cloud top height predictions. This study provides a comparative evaluation of initialization strategies and examines uncertainties in eruption source parameters and satellite retrievals, offering reference for volcanic ash forecasting.

The parameterization of raindrop size distribution (DSD) is critical for the satellite precipitation retrieval algorithms. Utilizing the multiple ground-based two-dimensional video (2DVD) and Particle Size and Velocity (Parsivel) disdrometers data from the Precipitation Validation Network (Guangdong) of the Fengyun satellites during April–September 2024, this study evaluates the uncertainty of DSD parameterization on a dual-frequency (DF) precipitation retrieval algorithm over South China. It is shown that the composite raindrop spectra generally conform to the gamma distribution, with the shape parameter μ on average of 4.5–4.8, which is higher than the fixed μ = 3 used in the Global Precipitation Measurement mission (GPM) Dual-Frequency Precipitation Radar (DPR) algorithms. By varying the μ value in the DSD gamma model, the effects on the retrieved mass-weighted mean diameter (Dm), normalized intercept parameter (Nw), and rain rate are examined. As μ increases from 1 to 6, the underestimation of Dm shifts to overestimation, while for lgNw and rain rate, it is the opposite. The overestimation of rainfall, especially at the range of 8’32 mm h−1, mainly comes from underestimated Dm and overestimated lgNw. On the contrary, overestimation of Dm and underestimation of lgNw mainly lead to underestimated rainfall, especially when rain rate is above 64 mm h−1. Comprehensive analysis shows that the DSD gamma distribution with μ in the range of 4’5 may be more suitable for South China. These results provide valuable reference for optimizing the DSD module of the precipitation retrieval algorithm for the Fengyun-3G (FY-3G) satellite.

To perform a full matrix capture (FMC) with a laser-induced ultrasonic phased array, both the generation and detection lasers must be independently scanned to acquire signals from all combinations of source and receiver positions. In contrast, a laser diffuse ultrasonic phased array (LDUPA) enables rapid acquisition of an equivalent FMC dataset by scanning only a detection laser while ultrasound is generated from a fixed transducer. This FMC dataset is subsequently reconstructed by applying a diffuse correlation technique to the raw receive dataset. While the reconstructed FMC dataset is used for total focusing method (TFM) images, the early, coherent raw receive dataset can also be used directly to form images via a receive-only focusing method (RFM). This study compares ultrasound images generated from both the raw receive datasets and the reconstructed FMC datasets across various inspection configurations, including different transmitter positions and incident wave angles. Results from simulations and experiments show that RFM images exhibit excellent detection capability within the incident wave’s illumination area. TFM images, in contrast, demonstrate good detection across the entire receiver aperture. The results also highlight the flexibility in transmitter positioning and incident wave angle, facilitating inspections that can adapt to accessibility constraints.

Atmospheric ammonia (NH3) is challenging to monitor accurately due to its short lifetime, high variability, and weak signal. Here, an advancing real-time NH3 retrieval model was developed that leverages high-frequency Fengyun-4B (FY-4B) GIIRS observations and meteorological data to generate 2-hourly NH3 maps over East Asia in tens of seconds, without costly radiative transfer simulations. Validations against the IASI product and FY-4B optimal estimation retrievals demonstrate high robustness (R > 0.75). A key advantage of this model is its ability to extract multiple source features, thereby producing physically consistent NH3 retrievals under weak-signal conditions, effectively eliminating 31.7% negative NH3 estimates produced by IASI at nighttime. This capability enables the timely tracking of the diurnal evolution of regional NH3 pollution, providing new insights into its emission patterns and transport processes.

Satellite-derived atmospheric motion vectors (AMVs) provide essential wind field data crucial for numerical weather prediction (NWP) and nowcasting applications. However, current operational AMV products typically offer relatively low spatial resolution, limiting their effectiveness in meeting precise meteorological forecasting requirements. Retrieving AMVs at higher spatial resolutions significantly increases computational demand, hindering their application in real-time operational scenarios. To address this challenge, this article introduces a novel GPU-accelerated (graphics processing unit) algorithm, based on OpenACC (ACCelerators), designed specifically to improve computational efficiency in the target-tracking part of the AMV retrieval algorithm. The proposed method decomposes complex calculations into parallel tasks suitable for efficient processing by GPUs. Additionally, to address the limited memory capacity of individual GPUs, a block-based computational strategy is developed, allowing for efficient use of multiple GPUs to process larger datasets. A comparative analysis at 48, 12, and 6 km resolutions showed that, at 6 km, OpenMP on a 48-core CPU achieved a 40 & times; speedup, while single- and dual-GPU configurations reached approximately 50 & times; and 110 & times;, respectively. The hybrid OpenACC+OpenMP strategy, combining one GPU with 48 CPU cores, delivered the highest acceleration at around 140 & times;, with all GPU-based methods maintaining near-lossless accuracy relative to the CPU baseline. This article provides a practical and effective solution for real-time high-resolution AMV retrieval, significantly improving the timeliness of critical meteorological services such as typhoon monitoring and data assimilation in NWP systems.

Geostationary hyperspectral infrared sounder observations provide two distinctive advantages: high temporal resolution and fixed observation geometry. However, conventional application frameworks have tended to focus on the quality of individual observations at a single time step and a single footprint without considering the temporal and spatial connections, thereby not able to fully exploit these unique strengths. Based on FY-4A/GIIRS observations, this study establishes a dynamic error characterization framework that employs multidimensional grouping across temporal, spatial, and detector arrays to statistically model the varying observation errors. Built upon this framework, key parameters for quantitative applications, radiance bias correction coefficients, observation error covariance matrices, and channel selection indexes, are derived dynamically. These parameters, which remain static in conventional approaches, are made dynamically adaptive to temporal and spatial variations within this framework. This geostationary hyperspectral sounder's methodology effectively utilizes 98% of the spectral channels, achieving a 0.14 K reduction in boundary layer temperature error and a 4.26% cut in overall humidity error.

This study demonstrates the potential values of geostationary satellite-derived mesoscale atmospheric motion vectors (MAMVs) to improve the analysis and forecast of typhoons. By utilizing high-frequency, high-resolution satellite imagery, MAMVs can provide novel important information on evolving wind fields that can influence typhoon structure, intensity, and track. This enhanced data, when assimilated, can lead to improve the skill of numerical weather prediction (NWP) models for typhoon prediction. To evaluate the effectiveness of MAMVs toward improving typhoon forecasts, this study assimilates MAMVs into NWP model and compares them with assimilating FY-4B/AGRI operational atmospheric motion vectors (AMVs) and control experiments without MAMVs/AMVs assimilation. It is shown that for two typhoon cases, the assimilation of MAMVs can improve typhoon dynamic field and reduce typhoon track forecast errors. These case studies demonstrate a promising approach to advancing typhoon prediction that can improve disaster preparedness.

Accurate information on cloud amount vertical structure is crucial for weather monitoring and understanding climate systems. Active sensors from satellites can provide three-dimensional (3D) cloud structure but with limited geographical coverage, passive sensors from satellites have expanded observation coverage but with limited capability on profiling the clouds. Combing active and passive observations from satellites, together with atmospheric reanalysis data, this study proposes a machine learning approach (CLANN, CLoud Amount Neural Network) to construct three-dimensional (3D) cloud amounts at passive observational coverage. Independent validation is conducted for cloud amount estimates derived from combined data of the Advanced Geostationary Radiation Imager (AGRI) onboard Fengyun-4 A and ERA5 using CALIPSO/CALIOP product as reference. The results indicate notable correlations (Pearson's r = 0.73). The cloud-amount-weighted height showed a high consistency in terms of height positioning between CLANN estimations and CALIOP data, with an RMSE of 1.88 km and a Pearson's r of 0.92. Key features such as water vapor band brightness temperature and upper-layer temperature significantly enhanced model accuracy, as revealed by permutation importance analysis. Sensitivity tests highlighted the critical role of the 1.375 mu m band in cirrus altitude detection, justifying the model's reliance on daytime observations. Additionally, the 3D statistical results from CLANN in 2019 reveal the seasonal variation details of cloud distribution, further demonstrating its application value in climate analysis.

Accurate cloud detection is critical for quantitative applications of satellite-based advanced imager observations, yet nighttime cloud detection presents challenges due to the lack of visible and near-infrared spectral information. Nighttime cloud detection using infrared (IR)-only information needs to be improved. Based on a collocated dataset from Fengyun-3D Medium Resolution Spectral Imager (FY-3D MERSI) Level 1 data and CALIPSO CALIOP lidar Level 2 product, this study proposes a novel framework leveraging Light Gradient-Boosting Machine (LGBM), integrated with grey level co-occurrence matrix (GLCM) features extracted from IR bands, to enhance nighttime cloud detection capabilities. The LGBM model with GLCM features demonstrates significant improvements, achieving an overall accuracy (OA) exceeding 85% and an F1-Score (F1) of nearly 0.9 when validated with an independent CALIOP lidar Level 2 product. Compared to the threshold-based algorithm that has been used operationally, the proposed algorithm exhibits superior and more stable performance across varying solar zenith angles, surface types, and cloud altitudes. Notably, the method produced over 82% OA over the cryosphere surface. Furthermore, compared to LGBM models without GLCM inputs, the enhanced model effectively mitigates the thermal stripe effect of MERSI L1 data, yielding more accurate cloud masks. Further evaluation with collocated MODIS-Aqua cloud mask product indicates that the proposed algorithm delivers more precise cloud detection (OA: 90.30%, F1: 0.9397) compared to that of the MODIS product (OA: 84.66%, F1: 0.9006). This IR-alone algorithm advancement offers a reliable tool for nighttime cloud detection, significantly enhancing the quantitative applications of satellite imager observations.

Measurements from a hyperspectral infrared(HIR)sounder onboard a satellite in geostationary orbit not only provide atmospheric thermodynamic information,but also can be used to infer dynamic information with high temporal resolution.Radiance measurements from the Geostationary Interferometric Infrared Sounder(GIIRS),obtained with 15-min temporal resolution during Typhoon Maria(2018)and 30-min temporal resolution during Typhoon Lekima(2019),were used to derive three-dimensional(3D)horizontal winds by tracking the motion of atmospheric moisture.This work focused on the impact of assimilation of 3D winds on typhoon analyses and forecasts using the operational NWP model of the China Meteorological Administration(CMA-MESO),and improved understanding of the potential benefits of assimilating dynamic information from geostationary sounder data with higher temporal resolution.The standard deviation of the observations minus simulations revealed that the accuracy of the derived 3D winds with 15-min resolution was higher than that of derived winds with 30-min resolution.Experiments showed that the assimilation system can effectively absorb the information of the derived 3D winds,and that dynamic information from clear-sky areas can be transferred to typhoon areas.In typhoon prediction,assimilation of the derived 3D winds had greatest influence on the typhoon track,and less influence on the maximum wind speed.Assimilation of the derived 3D winds reduced the average track error by 17.4％for Typhoon Maria(2018)and by 3.5％for Typhoon Lekima(2019)during their entire 36-h forecasts initiated at different times.Assimilation of GIIRS dynamic information can substantially improve forecasts of heavy precipitation by CMA-MESO.Results indicate that the assimilation of dynamic information from high-temporal-resolution geostationary HIR sounder data adds value for improved numerical weather prediction.

For numerical weather prediction (NWP),data assimilation (DA) combines short-term forecasts and various atmospheric observations to achieve optimal initial conditions,based on which subsequent forecasts are launched.With the rapid advancements in numerical models and observing systems,DA has been significantly evolved.Modern methods now can account for uncertainties of state variables across various spatiotemporal scales,incorporate multiscale observation error statistics,and enforce dynamical constrains and model balances.Meanwhile,observations from various platforms,such as ground-based,aircraft,and satellite,have been assimilated.These include data from polar-orbiting and geostationary satellites,radar-derived radial winds and reflectivity,Global Navigation Satellite System (GNSS) radio occultations,etc.To further utilize the advanced observing systems and DA techniques for high-impact weather predictions,target observation strategies have been developed to identify areas where additional observations can yield the greatest predict improvements.Based on the advancements of DA theories and methods,China’s operational systems have made significant progress,establishing advanced operational DA systems.Over the past decade,the forecast skill of 5-day global weather prediction has improved by approximately 15%.The article reviews a century of development in DA,and discusses future directions,including the advanced DA methods,operational frameworks,integration of novel observations,and the synergy between DA and artificial intelligence.