Accurate estimation of surface shortwave downward radiation (SWDR) is crucial for understanding global climate change. However, satellite-derived SWDR products often exhibit substantial biases over highly reflective surfaces, largely due to poorly quantified adjacency effects originating from the surrounding environment under enhanced multiple scattering conditions. To address this issue, this study proposes an adjacency effect correction scheme for highly reflective surfaces grounded on the mechanisms of atmospheric radiative transfer. Its core novelty lies in the development of separate quantitative models for direct (Dir) and diffuse radiation (Diff) adjacency effects, which further incorporate an albedo-threshold-based adaptive correction strategy that enables targeted corrections for both SWDR and Dir. Unlike purely data-driven approaches, the scheme ensures radiation budget consistency by restoring authentic surface radiative states, thereby enhancing physical interpretability. The corrected shortwave radiation has been validated using polar sites and non-polar sites during the ice and snow seasons. Specifically, the mean biases of SWDR and Dir across all sites have been reduced by 95.0% and 84.9%, respectively, compared to those before correction. Antarctica exhibited greater accuracy improvements compared to the Arctic, with maximum bias reductions of 40.6 W/m2 (96%) and 32.8 W/m2 (95.3%) for SWDR and Dir, respectively. Non-polar sites also demonstrated enhanced accuracy, with the mean bias reduction exceeding 10 W/m2. The maximum reductions reached 25 W/m2 (65.1%) and 28.4 W/m2 (58.7%) for SWDR and Dir, respectively. Furthermore, the corrected SWDR and Dir over highly reflective surfaces showed improved spatial consistency with true-color composite images and cloud optical thickness maps. The simple and efficient scheme effectively mitigates adjacency effect-induced underestimations over highly reflective surfaces. It can be integrated into retrieval algorithms or applied as a post-processing module to existing products, achieving an 11.1 W/m2 (56.6%) bias reduction in the CERES SSF dataset and showing promising potential for refining high-accuracy shortwave radiation products, particularly in polar regions.

This study introduces FYAI, a global, long-term atmospheric ice water path (IWP) and suspended ice water path (SIWP) dataset spanning 2010-2024, derived from passive microwave observations (MWHS-I/II) onboard China's Fengyun-3 series satellites. The dataset is generated using a machine learning framework featuring a lightweight multilayer perceptron architecture enhanced with gated residual units. This design robustly handles the inherent uncertainties in satellite brightness temperatures and the spatial mismatch between passive microwave footprints and active radar/lidar training data. By establishing rigorous spatiotemporal collocation with CloudSat 2C-ICE products, FYAI provides two operational product levels adhering to standard Earth observation data processing definitions: (1) Level-2 (L2) products, offering instantaneous orbital-resolution IWP and SIWP at a nominal 15 km nadir resolution for 2010-2024; and (2) Level-3 (L3) products, comprising monthly global gridded composites at 1 degrees & times;1 degrees resolution (2010-2024). FYAI bridges the gap between instantaneous pixel-level precision and broad spatiotemporal coverage, offering a comprehensive, decadal-scale record of global atmospheric ice content. This dataset, specifically designed to support long-term climate analysis and model validation, is openly available in netCDF4 format for community use (10.11888/Atmos.tpdc.303143, Yang et al., 2025)

Aerosol layer height (ALH) and aerosol optical thickness (AOT) are crucial parameters for characterizing aerosol vertical structure and radiative effects. Although significant advances have been made in passive satellite-based AOT retrieval, the simultaneous and accurate retrieval of ALH and AOT remains a challenge. Hyperspectral measurements are typically used for ALH retrieval; however, lookup table-based retrieval algorithms normally extract two or three spectral bands and are difficult to leverage a larger set of spectral bands from hyperspectral data due to the 'curse of dimensionality'. Optimal estimation methods are capable of utilizing hyperspectral data, whereas they are computationally intensive and constrained by a priori assumptions. To address these challenges and enable high-accuracy, rapid, and simultaneous retrieval of ALH and AOT, this study develops a novel algorithm, SIMART (Simultaneous Inversion of ALH and AOT using Machine learning and Radiative Transfer). The algorithm leverages hyperspectral observations from the O(2)A and O2B absorption bands, as well as blue and shortwave infrared bands, acquired by the TROPOspheric Monitoring Instrument (TROPOMI) sensor. SIMART integrates three key innovations within a radiative transfer (RT) framework: (1) selection of representative aerosol profiles informed by sensitivity analysis and active lidar observations, (2) identification and incorporation of optimal spectral features with high sensitivity, and (3) simulation of realistic scenarios across a wide range of aerosol and satellite observation conditions. Employing a physics-based approach, SIMART generates an RT simulation dataset to train an artificial neural network model, where the complex relationship between inputs (e.g., satellite-observed spectra or the ratios between spectra) and outputs (i.e., ALH and AOT) can be established to guarantee the retrieval flexibility and efficiency. In addition, to maximize spectral information and suppress noise, the TROPOMI data are spectrally convolved using narrowband filters prior to model input. Validation against CALIOP data for global dust and smoke events demonstrates SIMART's superior performance in ALH retrieval, achieving a mean bias error (MBE) below 0.1 km and a correlation coefficient (R) exceeding 0.75. Furthermore, the SIMART-derived AOT exhibits strong correlations with the ground-based AERONET measurements (MBE = 0.004, R = 0.96) and the satellite-based VIIRS product (MBE < 0.05, R = 0.92). Compared to the official TROPOMI ALH product (v02.04.00), SIMART improves the R value by nearly 40% and reduces the root mean square error by approximately 49%. Although the updated TROPOMI version (v02.08.00) has also mitigated this bias, SIMART demonstrates superior performance. This algorithm demonstrates strong potential for operational, high-accuracy, and simultaneous ALH and AOT retrieval from hyperspectral passive satellite data. Furthermore, its framework is highly adaptable and can be flexibly extended to other hyperspectral sensors.

Radiative transfer (RT) modeling in coupled atmosphere–ocean systems provides a fundamental theoretical basis for retrieving atmospheric and oceanic optical parameters from remote sensing observations. The accurate retrieval of these parameters relies on forward RT calculations performed by flexible and high-precision radiative transfer models (RTMs). Despite substantial progress, RTMs that simultaneously achieve benchmark-level hyperspectral gas absorption with enhanced computational efficiency, incorporate irregular particle scattering, and account for polarized RT in coupled atmosphere–ocean systems, remain limited. In this study, we develop CARE-RTM, a comprehensive RTM designed to support polarimetric and hyperspectral simulations across ultraviolet to infrared wavelengths in coupled atmosphere–ocean systems, through sophisticated modeling of absorption and scattering processes in both atmospheric and oceanic substances. Particularly, CARE-RTM incorporates a GPU-accelerated line-by-line scheme with line-mixing refinement, a Voronoi ice crystal scattering model, and comprehensive bio-optical ocean modules, enabling the accurate treatment of hyperspectral gas absorption, irregularly shaped ice cloud scattering, and underwater light simulation, respectively. Validation against benchmark solutions confirms the high level of accuracy of CARE-RTM, with errors typically below 0.02

The Tibetan Plateau (TP), known as the "Asian Water Tower", plays a critical role in the regulation of the water cycle in the region. Obtaining high spatiotemporal resolution, and all-weather total precipitable water (TPW) data is essential for understanding water vapor transport mechanisms, improving precipitation forecasting, and managing regional water resources over the TP. However, existing single-sensor remote sensing techniques cannot provide high spatiotemporal resolution TPW data under cloudy conditions. Multi-source fusion approaches often produce anomalous distributions in the fused TPW data due to inter-sensor biases, particularly over the complex terrain of the TP. This study proposed a multi-source remote sensing TPW fusion framework that integrates TPW products from eight microwave satellites and the Himawari-8/9 (H8/9) geostationary satellite to produce an all-weather TPW data with the highest spatiotemporal resolution at present. Methodologically, two correction strategies were developed. First, a bias correction approach was proposed using H8/9 TPW data as a reference to calibrate multi-source microwave remote sensing TPW and reduce inter-sensor discrepancies. Second, an adaptive correction method was created to improve the accuracy and spatial continuity of the fused TPW data under cloudy conditions. Based on the newly developed fusion framework, an all-weather TPW dataset with hourly temporal and 0.02 degrees spatial resolution covering the TP from 2016 to 2022 was produced for the first time. The new dataset has been published by the National Tibetan Plateau Data Center and is available at: 10.11888/Atmos.tpdc.301518 (Ji et al., 2025b). Taking the 2017 product as an example, it was verified against GNSS TPW. The RMSE of the fused TPW product at the hourly scale was 3.79 mm, which was 10.82 % and 6.19 % lower than MIMIC-TPW2 and ERA5, respectively. Compared to ERA5 with a spatial resolution of 0.25 degrees, the fused product achieves a 12.5-fold improvement in spatial resolution, which make it possible to significantly grasp the transportation of water vapor in the valley of Yarlung Zsangbo River. It also demonstrates higher reliability in station-sparse regions, providing high-quality, high-resolution vapor data to support vapor flux estimation and forecasting of extreme weather events over the TP.

Aerosol–cloud interactions (ACI), which substantially offset anthropogenic greenhouse warming, remain a major source of uncertainty in current climate assessments. Satellite-based estimates of ACI radiative forcing (RFaci) serve as a key benchmark for climate predictions and for evaluating improvements in climate models. However, these estimates remain poorly constrained. A critical limitation is that satellite assessments typically rely on column-integral aerosol proxies (e.g., AOD, AODf, AI, etc.), which may not accurately represent cloud-base cloud condensation nuclei (CCN)—the particles that actually form cloud droplets. This limitation has given rise to a puzzling phenomenon: over the Southern Hemisphere, cloud droplet concentrations have declined despite increases in column-integral aerosol proxies. While some previous studies have noted this droplet–aerosol trends discrepancy, they often relied on limited datasets and single aerosol proxies, without providing systematic validation, causal analysis, or quantification of its implications for RFaci. This gap has been a significant obstacle to reducing uncertainties in ACI forcing.To address this challenge, we first combined multi-source observations to provide robust, quantitative evidence of the droplet–aerosol trends discrepancy across the Southern Hemisphere from 2003 to 2020. We then used a source–sink framework to explore the underlying physical mechanisms, finding that the discrepancy arises from elevated cloud bases systematically reducing CCN availability, while enhanced precipitation accelerates droplet removal. By explicitly accounting for these processes, this study provides a physically grounded estimate of aerosol–cloud radiative forcing, constraining RFaci to −1.37 W m⁻². Previous global assessments relying on column-integral variables are therefore biased by –35% to +26%, with discrepancies reaching +42% over the Southern Hemisphere.This work reconciles a long-standing discrepancy between observed droplet and column-integral aerosol trends, highlighting the critical importance of considering cloud-base CCN in future ACI radiative forcing estimations. It provides a physically grounded constraint on aerosol forcing based on cloud-base CCN, supporting more precise estimates of climate sensitivity and guiding model development.

ABSTRACT Cloud horizontal heterogeneity within instrument footprints constitutes a major error source in ice cloud retrievals. We present a novel algorithm considering cloud horizontal heterogeneity based on lightweight convolutional neural network (LCNN) designed for Ice Cloud Imagers (ICI), which retrieves high‐resolution ice water path (IWP) by leveraging spatial correlations in low‐resolution satellite observations. Our approach employs a prior database generated through three‐dimensional radiative transfer simulations, containing collocated high‐resolution atmospheric profiles and cloud parameters, as well as the corresponding brightness temperature differences (BTDs). Then, the LCNN is trained and tested by the simulated prior database. Compared to Bayesian Monte Carlo integration (BMCI), the LCNN reduces mean absolute error (MAE) by 12.05 g/m2 while tripling spatial resolution (from 16 to 5.3 km). This resolution enhancement mitigates errors induced by ice cloud horizontal heterogeneity, achieving a 23.6% reduction in MAE attributing to plane‐parallel cloud assumptions within footprints. The framework demonstrates the viability of deep learning for submillimeter‐wave cloud sensing, providing a pathway to resolve sub‐footprint cloud variability in future satellite missions.

High-precision, real-time precipitation estimation is critical for improving the accuracy of meteorological forecasting and disaster warnings. However, mainstream satellite remote sensing inversion techniques are often limited by their reliance on cloud-top information. In addition, estimation accuracy is further constrained by the severe category imbalance in precipitation samples and the widespread use of low-spatial-resolution reanalysis data, making it difficult for existing products to meet the practical application requirements. To address these challenges, this study proposes a high-precision real-time precipitation inversion algorithm named geostationary satellite precipitation estimation-dynamic updating machine learning (GSPE-DML), which is based on the Fengyun-4B (FY-4B) satellite and machine learning. The algorithm is designed with two core innovations. First, it incorporates high-resolution (3 km) CMA-BJ v2.0 reanalysis data to replace conventional low-resolution datasets such as ERA5, thereby capturing more detailed information on the subcloud vertical atmospheric structure. Second, a real-time dynamic sample balance mechanism is introduced, which adaptively adjusts category weights during model training to effectively mitigate the precipitation sample imbalance problem. A comparative analysis conducted across different temporal scales indicates that the performance of GSPE-DML is better than that of real-time precipitation products such as FY-4B quantitative precipitation estimation (FY4B-QPE), APCP, and Global Satellite Mapping of Precipitation (GSMAP)-NRT across several key metrics. On the hourly scale, in particular, GSPE-DML maintained a Heidke skill score (HSS) above 0.52 and achieved a correlation coefficient (CC) of 0.45 with ground observations. Notably, its false alarm rate (FAR) was significantly lower than that of the GSMAP-NRT product across all four quarters. A case study of Typhoon "Doksuri" further validated the robustness of the algorithm. The inversion results showed minimal deviation (-0.01) from ground observations and high consistency in spatial patterns, successfully reconstructing the intense precipitation band near the typhoon's eyewall and the weak precipitation structure within its outer spiral rainbands.

Soil temperature (ST) is a fundamental meteorological variable that governs energy, water, and carbon exchanges across the land-atmosphere interface, with profound implications for hydrological forecasting, ecological modeling, and climate change projections. Despite its critical importance, global subsurface soil temperature profiles remain poorly characterized by existing remote sensing capabilities, which are limited to instantaneous, single-depth retrievals with inadequate temporal sampling. This study proposes a Diurnal Soil Thermal (DST) model founded on the one-dimensional heat conduction equation, representing daytime temperature propagation as a harmonic wave and nighttime cooling as an exponential decay. Building upon this physical model, we propose the Diurnal Soil Thermal-Mapping Algorithm for Profiles (DST-MAP), which integrates land surface temperature observations from Fengyun-3 multi-frequency microwave radiometer and MODIS thermal infrared sensors to retrieve diurnal temperature cycle parameters. The damping depth characterizing vertical attenuation is constrained by soil texture and satellite-derived soil moisture using a physically based parameterization. DST-MAP generates continuous hourly soil temperature estimates at 5, 10, 20, and 30 cm depths. Validation against in situ measurements from 450 + stations demonstrate robust performance, with correlation coefficients exceeding 0.87 and root-mean-square errors (RMSE) of 3.5-4.5 K across all depths. Accuracy is highest in shallow layers, progressively attenuating with depth yet remaining reliable even at 30 cm. Triple Collocation Analysis (TCA) among DST-MAP, MERRA-2, and GLDAS indicates theoretical uncertainties below 4 K for most land areas, with DST-MAP exhibiting superior consistency in arid regions compared to MERRA-2 and comparable performance to GLDAS. This work establishes an observation-driven, physically grounded framework for hourly-resolved subsurface temperature profiling that captures the complete diurnal thermal evolution, providing essential support for the development of next-generation land surface hydrothermal coupling models and satellite payloads.

Clouds strongly modulate the radiation balance of the Earth and the atmosphere system by reflection of shortwave radiation and absorption of longwave radiation. In this study, we develop a shortwave cloud radiative effect (SWCRE) algorithm for the cloud, atmospheric radiation, and renewal energy application (CARE) datasets (Version 1.2) from geostationary satellites of Himawari-8/9 and FY-4A. First, the lookup tables (LUTs) for retrieval of cloud optical thickness (COT) and cloud effective radius (CER) (LUTs-A), and shortwave radiation at the surface and TOA (LUTs-B) were generated by the radiative transfer model (RSTAR). Second, COT and CER for water and ice clouds were retrieved based on the LUT method with LUTs-A, respectively. Third, shortwave radiation at the surface and TOA were estimated under assumptions with clouds (all-sky) and without clouds (clear-sky). Finally, an SWCRE at the surface and TOA was estimated based on the difference between all- and clear-sky shortwave radiation. Validations of SWCRE from Himawari-8 against the five years (2016-2020) Clouds and the Earth's Radiant Energy System (CERES) product show high accuracy with root-mean-squared error (RMSE) of 5.5 and 4.6 Wm-2 at the surface and TOA, respectively; R values of 0.96 and 0.97. High accuracy, spatial-temporal resolutions of CARE SWCRE products can be further used for Earth's radiation balance study.

High spatiotemporal resolution remote sensing products are essential for advancing Earth system science. These products, which include key atmospheric and surface radiation parameters, are crucial not only for studying cloud-radiation-climate interactions and global radiative energy balance, but also for understanding multi-sphere interactions within the Earth system. The Cloud Remote Sensing, Atmospheric Radiation and Renewable Energy Application (CARE) algorithm system and products provide a comprehensive suite of around 30 parameters, including cloud and aerosol properties, atmospheric water vapor, and surface radiation budget. CARE products are primarily generated using observations from new-generation geostationary satellites like FY-4 and Himawari-8, combined with data from polar-orbiting satellites including FY-3 and MODIS, enabling multi-scale data coverage across East Asia and the globe. A key advantage of CARE products is their high spatiotemporal resolution: global products achieve a 5 km and 30-min spatiotemporal resolution, enabling detailed characterization of diurnal variations in parameters such as cloud cover, cloud water content and surface radiation flux. Notably, parameters like downward shortwave radiation show higher accuracy compared to other existing datasets. The CARE system integrates a full-spectrum ice crystal scattering model, the high-performance CARE radiative transfer model (CARE-RTM), advanced remote sensing retrieval algorithms incorporating artificial intelligence (AI) technology, and a near-real-time monitoring platform to facilitate product development. This study summarizes the recent development of CARE models, algorithms, and products, highlighting the unique features of the full-spectrum ice crystal scattering model, the enhanced RTM, high-performance remote sensing retrieval algorithms with accuracy evaluation, and the broad use of CARE-derived datasets in atmospheric and climate research.

The dynamical and microphysical processes that govern the lifecycle of hail-producing deep convective clouds (DCCs) remain poorly understood, limiting severe weather prediction. Here, we dissect a severe hailstorm that occurred over Inner Mongolia using multi-source observations, including Himawari-8 satellite data, Doppler radar, and a lightning mapping network. Our analysis reveals a tightly coupled co-evolution of cloud-top microphysical properties, cloud-top kinematics, and electrical activity. A key finding is the synchronization during rapid updraft intensification of a collapsing cloud-top effective radius (from similar to 40 mu m to similar to 20 mu m) with a surge in total lightning flash rate. Rapid updrafts likely shorten particle residence time, limiting particle growth while accelerating mixed-phase collisions and non-inductive charging, thereby promoting lightning jump activity. Critically, these abrupt changes in updraft velocity and lightning activity preceded surface hailfall and peak rainfall by approximately 30-40 min and 2 h, respectively. This study provides quantitative evidence that the integration of satellite, radar, and lightning observations can elucidate the microphysical pathways leading to severe convective weather and offers valuable lead time for improved nowcasting.

The base heights and geometric thickness of clouds are key macrophysical parameters critical for climate, weather, and aviation safety. Existing methods exhibit various limitations in obtaining cloud base height (CBH) measurements. In order to complement and extend existing methods, this paper presents a retrieval method for single-layer cloud base height based on the oxygen A band (OA). On this basis, multi-angle polarization bands can be used to obtain the top heights of clouds with high precision. Then, by combining the top height of a cloud with its base height, the cloud geometric thickness (CGT) can be acquired. Simulation experiments involving radiative transfer models indicate that the OA exhibits regular sensitivity to the CBH. Within the OA, the ratio of narrow-channel (763 nm) to wide-channel (765 nm) radiation intensities increases as the CBH increases. Owing to the uniform distribution of oxygen in the atmosphere, the absorption in the OA is stable. Additionally, the wealth of information derived from multi-angle remote sensing can further increase the accuracy of retrievals. Therefore, the multi-angle OA is incorporated into the model training process. CBH obtained from CloudSat are used as the true values. The longitude, latitude, and multi-angle OA information obtained from PARASOL Level 1 is utilized to retrieve the CBH. After several machine learning algorithms are compared, the deep neural network (DNN) model with the best accuracy is selected as the retrieval model. The method of CBH retrieval based on multi-angle OA remote sensing and the DNN has a mean absolute error (MAE) of 0.75 km, a bias (absolute) of 0.23 km, correlation coefficient (R) of 0.82, and a mean relative percentage difference (MRPD) of 34%. In the CGT retrieval, the MAE was 0.72 km, the bias was-0.15 km, R was 0.93, and the MRPD was 42%. In comparison with the results from PARASOL Level 2 and CloudSat, the present study achieves a lower MAE by 0.97 km, a lower RMSE by 1.23 km, a higher R by 0.17, and a reduced MRPD by 27%. This study demonstrates the efficacy of a hybrid approach that integrates multi-angle OA remote sensing with deep learning for global retrieval of single-layer cloud CBH and CGT.

Synoptic quantification of phytoplankton depth-integrated primary production (IPP) has advanced significantly over recent decades by leveraging satellite observations and sophisticated IPP models. However, monthly mean upstream products from polar-orbiting satellites, e.g., the Moderate Resolution Imaging Spectroradiometer (MODIS), are commonly used to generate IPP products, raising a concern about whether neglecting diurnal or daily IPP variabilities may compromise the accuracy of monthly-and annual-scale quantifications. Here, we aim to investigate this concern by comparing IPP quantified using high-frequency data at multiple timescales. A theoretical time-resolved model (TPM) was utilized for IPP modeling, driven by either diurnal photosynthetically available radiation (PAR) from the Advanced Himawari Imager (AHI) onboard Himawari-8 (H8) or daily PAR from MODIS. Preliminary evaluation against in situ measurements corroborated the superiority of AHI-based daily IPP estimation over MODIS, attributed to the robustness of AHI PAR data across varying sky conditions. Satellite IPP products were generated in the full-disk area of H8 between 2016 and 2019 under "daily-to-monthly-to-annual" (DtA) and "monthly-to-annual" (MtA) scenarios for comparison, using other requisite daily and gap-free biogeochemical products. Our analysis unveiled moderate spatiotemporal discrepancies between DtA-based IPP products from AHI and MODIS, confirming an overestimation in MODIS-derived monthly (< 8%) and annual total IPP (similar to 5%). In contrast, under the MtA scenario, MODIS substantially overestimated monthly (similar to 14-30%) and annual total IPP (similar to 20%) and gave biased temporal trends (similar to 1.3-1.6 times higher) compared to DtA-based IPP estimates of AHI. The discrepancies between IPP products were largely subject to the cloud-induced variabilities in daily PAR products and ocean color data coverage. By upscaling our results to the global ocean, it is anticipated that the annual total IPP previously estimated from MODIS with TPM-like models is overestimated by at least 19%. This study emphasizes the necessity of modeling IPP at finer timescales using high-frequency observations and provides insights for improving IPP quantification with the aid of geostationary satellites.

Abstract Cloud detection is a critical procedure in satellite remote sensing. Most meteorological satellite products provide binary cloud masks, which identify whether a pixel in the satellite image is entirely cloudy or clear‐sky, and the subpixel cloud fraction (CF) of partly cloudy pixels is usually unavailable. In this work, we develop a deep‐learning neural network to estimate the subpixel cloud fraction of moderate‐resolution satellite images based on binary cloud masks, and train the neural network with high‐resolution satellite data. We aggregate 10 m Sentinel‐2 cloud masks into physically consistent 1 km binary cloud masks, smooth them with edge‐aware filtering to obtain collocated 1 km CF reference fields, and use these pairs to train DualCloudNet, a dual‐branch U‐Net that fuses complementary feature representations. The network takes the aggregated 1 km cloud mask as input and predicts the corresponding 1 km continuous CF, achieving a Pearson correlation of 0.92 and an RMSE of 0.103 on an independent test data set. The model outperforms a physically motivated statistical estimator, and its accuracy is slightly higher than two other deep‐learning networks that were constructed in the validation stage of this study. Applied without retraining to FY‐3D MERSI‐II and Landsat‐8 imagery, DualCloudNet reproduces pixel‐level cloud structures, demonstrating cross‐sensor generalization. Compared with existing cloud fraction retrieval algorithms, the new algorithm is able to provide a subpixel cloud fraction with low computational cost. The fine‐scale CF information is potentially useful for improving cloud property retrievals and surface radiation budget estimation, where the representation of partly cloudy pixels remains a major source of uncertainty.

Ocean color (OC) data, particularly remote-sensing reflectance (Rrs), are crucial for estimating marine optical property parameters. Although some polar-orbiting satellites provide reliable and operational Rrs through atmospheric correction (AC), achieving comparable accuracy and wide-area synchronous observations using Himawari-8 (H8) over the East Asia–Pacific region with a high-temporal resolution remains a marked challenge. Since H8 is not a dedicated OC satellite, its ability to capture rapid minute-level OC changes is restricted. The primary challenge involves further enhancing the Rrs retrieval accuracy beyond that of hourly products to compensate for the sensor’s signal-to-noise ratio (SNR) limitations. To address these issues, we develop a novel algorithm for retrieving Rrs from H8. This algorithm innovatively integrates the Moderate Resolution Imaging Spectroradiometer (MODIS) with H8 at a 10-min resolution using machine learning. This ability to retrieve high-frequency Rrs from H8 represents an unprecedented advance. After applying a classic AC framework for gas absorption and Rayleigh scattering corrections, an advanced transformer-based algorithm is employed for Rrs retrievals. The results demonstrate high spatiotemporal consistency with MODIS OC products. A validation with AERONET-OC confirms that the new algorithm substantially improves upon the accuracy of the H8 Rrs products, with RMSE reductions of approximately 34%, 26%, and 12% at 470, 510, and 640 nm, respectively. The algorithm effectively mitigates the underestimation of Rrs values from H8 products at 470/510 nm in turbid waters, and the overall overestimation at 640 nm in clear water. This work pioneers OC variation analyses at 10-min intervals using H8.

Estimation of surface longwave cloud radiative forcing (LWCRF) is crucial for understanding cloud-climate interactions but remains challenging due to the complexity of radiative transfer simulations and the difficulty in accuracy evaluation. Using only five readily available parameters: digital elevation model, column water vapor, cloud top temperature, cloud optical thickness, and cloud fraction ratio, this study develops a lightweight algorithm to estimate LWCRF, defined as the contribution of clouds to surface downward longwave radiation. The algorithm achieves a theoretical RMSE (root-mean-square error) of 5.47 W/m2 under rigorous radiative transferbased testing. When applied to the CERES CRS (Clouds and the Earth's Radiant Energy System, Clouds and Radiative Swath) data which is based on pixel-level Langley Fu-Liou radiative transfer simulations, it exhibits an RMSE of 12.03 W/m2 across 812 million global samples, demonstrating strong consistency in spatial patterns. This work provides a practical and efficient alternative when numerous inputs required by radiative transfer model are unavailable. This advancement allows for easy assessment of the impact of clouds on the global radiation balance, thereby advancing the understanding of how much clouds warm or cool the Earth.

Accurate estimation of surface solar radiation and its global, direct, and diffuse components at high spatiotemporal resolution remains challenging, despite the availability of existing products, due to the combined effects of rapidly varying clouds, aerosols, and the strong dependence of many retrieval methods on auxiliary atmospheric datasets. In this study, we develop a transfer learning framework that leverages the Cloud, Atmospheric Radiation and Renewal Energy Application radiation product derived from Himawari-8 and extend it to China’s FY-4A geostationary satellite. Specifically, a deep neural network pretrained with Himawari-8 data was fine-tuned with FY-4A observations, while Bayesian optimization was employed to automatically determine key hyperparameters and enhance model generalization. The framework requires only satellite top-of-atmosphere reflectance and solar-satellite geometry as dynamic inputs, thereby avoiding reliance on auxiliary meteorological datasets that often restrict real-time and large-scale applicability. Validation against 33 sites from 3 networks during 2018 to 2020 demonstrates that the FY-4A retrievals achieve high accuracy at both instantaneous and daily scales. For representative Baseline Surface Radiation Network sites, the root mean square errors of global, direct, and diffuse radiation are 102.2, 117.5, and 83.1 W m−2 at the instantaneous scale, and 28.5, 30.1, and 22.6 W m−2 at the daily mean scale, respectively. Comparative analysis further indicates that FY-4A outperforms the Clouds and the Earth’s Radiant Energy System in estimating direct and diffuse radiation and consistently surpasses ECMWF Reanalysis v5 across all radiation components. This study establishes the high-precision solar radiation estimation algorithm specifically designed for Chinese geostationary satellites, providing a robust data foundation for solar power generation, energy forecasting, climate research, and broader applications in sustainable energy transition.