Aerosol forecasting is important for air-quality management, health risk assessment and climate change mitigation1,2. However, it is more complex than weather forecasting, owing to the interactions between aerosol physicochemical processes and atmospheric dynamics, resulting in high uncertainty and computational costs3,4. Here we develop a machine-learning-driven Global Aerosol-Meteorology Forecasting System (AI-GAMFS), which provides reliable 5-day, 3-hourly forecasts of aerosol optical components and surface concentrations. AI-GAMFS combines a vision transformer and U-Net in a backbone network, robustly capturing the complex aerosol-meteorology interactions via global attention and spatiotemporal encoding. Trained on 42 years of aerosol reanalysis data and initialized with Global Earth Observing System Forward Processing (GEOS-FP) analyses, AI-GAMFS delivers operational 5-day forecasts in 1 minute. Evaluation with independent ground-based observations suggests improved performance compared with the Copernicus Atmosphere Monitoring Service5 and regional dust models6-9 in forecasting aerosol optical depth and dust components. Compared with GEOS-FP10, it has a lower root-mean-square error for global aerosol optical depth, with comparable dust forecasting skill and improved surface aerosol component forecasts over the USA and China. Our results provide a step forward in leveraging machine learning to refine aerosol forecasting and may help warn against aerosol pollution events such as dust storms and wildfires.

Orbiting at Sun-Earth Lagrange-1 point, the Earth Polychromatic Imaging Camera (EPIC) provides sunlit Earth images at 10 channels from ultraviolet (UV) to near-infrared (NIR) every 1–2 h. Here, we presented the version 2 of the algorithm to retrieve aerosol optical depth (AOD) and aerosol optical centroid height (AOCH) from visible (443 and 680 nm) and O2 absorption (688 and 764 nm) bands of EPIC (hereafter AOCH-v2 algorithm). The EPIC AOCH-v2 algorithm differs from the version 1 (AOCH-v1) algorithm in several aspects. First, an independent calculation of ultraviolet aerosol index (UVAI) was added by using two UV bands (340 and 388 nm) to identify absorbing aerosols for AOCH retrieval. Second, we updated the real part of aerosol refractive index to represent a new smoke aerosol model for large smoke AOD scenarios over North America, while retaining the original smoke model generated from the Aerosol Robotic Network (AERONET) climatology for background aerosol cases. Third, a new AOD retrieval scheme was developed to constrain the surface reflectance using the climatological surface reflectance ratios. Finally, the cloud mask scheme was improved for cloud screening over bright surfaces. The EPIC AOD from the AOCH-v2 algorithm displays a better agreement with AERONET than AOCH-v1 algorithm. The evaluations against Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) measurements indicate that the AOCH overestimation in AOCH-v1 algorithm is significantly mitigated by AOCH-v2 algorithm, suggesting uncertainties in the EPIC AOCH retrievals are notably reduced through a better characterization of surface reflectance, enhanced AOD retrievals, and an updated smoke aerosol model.

Artificial Light at Night (ALAN) poses risks to public health and ecosystems. While long-term remote sensing has tracked global nighttime light, spectrally resolved light at night (spectral light at night (SLAN)) data from geostationary orbit only became available with NASA's Tropospheric Emissions: Monitoring of Pollution (TEMPO) mission. Its Level 1 twilight radiance product enables detailed spatiotemporal analysis of ALAN's spectral characteristics. This study introduces algorithms for processing TEMPO Level 1 data into Level 2 SLAN data using two methods for solar background removal: scattering-angle correction and spectral correction. The first method uses the relationship between scattering angle and radiance to suppress background, while the second uses the inherent spectral relations in the solar radiation in 290-390 nm range, where artificial light emissions are minimized, to estimate direct twilight contributions. Combined, these methods yield more stable correction results. The results are validated against two independent data sets: the Visible Infrared Imaging Radiometer Suite (VIIRS) Day/Night Band (DNB) Black Marble product and astronaut photographs taken from the International Space Station (ISS) over Houston. Comparisons include (a) city light classifications from Black Marble versus both raw and corrected TEMPO data, (b) pixel-to-pixel radiance measurements between TEMPO and VIIRS, and (c) spectral angle analyses between ISS and TEMPO data. We also compare CONUS nighttime light maps from VIIRS DNB, TEMPO VIS, and TEMPO UV; TEMPO VIS agrees closely with VIIRS DNB, while both diverge from TEMPO UV in gas-flaring regions, highlighting the complementary information these instruments provide.

Recent surges in wildfire emissions have exacerbated surface ozone pollution in the United States. Using deep learning, we developed a gapless daily surface ozone dataset at 1-kilometer resolution for 2003-2024. This dataset revealed a reversal in national policy-relevant ozone trends that had gone undetected by the sparse monitoring network: from -0.65 parts per billion (ppb) per year (2003-2015) to +0.13 ppb per year (2015-2024). The reversal was primarily driven by increasing wildfire emissions, offsetting 3.9 years of mitigation progress. Premature deaths from fire-sourced ozone have increased by 318 deaths per year since 2013, with post-2013 mortality 46% higher than pre-2013 mortality. During 2022-2024, wildfire emissions exposed 43 million people to nonattainment conditions, effectively preventing a 4-ppb tightening of the ozone standard. These results underscore the growing challenges of sustaining air quality progress as wildfires intensify under climate change.

Optical remote sensing images are essential for Earth surface analysis; however, their applications are limited by cloud occlusion, as annual global cloud coverage exceeds 60%. The introduction of temporal compensation has addressed the limitation of single-image approaches, which struggled with thick cloud cover. The integration of synthetic aperture radar (SAR) data has further improved their cloud removal performance. Nevertheless, existing methods still face notable challenges. Specifically, they lack fine-grained, long-range compensation and adaptive mechanisms to effectively distinguish the high-frequency information from noise. Moreover, current approaches lack an optimization mechanism to ensure the acquisition of the most representative features. To address these issues, this paper proposes a novel Optical-SAR fusion method for cloud removal that explicitly combines controlled spatio-temporal aggregation with causal regularization (CR). A pixel-wise channel attention (PCA) module with high-frequency refinement (HRR) is introduced to enable full-resolution feature extraction while introducing frequency compensation advantages under acceptable computational complexity. Temporal attention aggregation is applied to facilitate effective temporal compensation. Furthermore, the CR module is employed to suppress statistically correlated but semantically irrelevant features arising from Optical-SAR fusion and temporal aggregation, improving robustness and generalization. Extensive experiments on both simulated and real-world datasets confirm the effectiveness of the proposed spatio-temporal feature extraction architecture, showing measurable yet modest metric improvements and state-of-the-art generalization under realistic cloud contamination. https://github.com/wjven/CR-PCA-HRR.

Large particle size and thick coatings have been detected in pyrocumulonimbus (pyroCb) smoke injected into the lower stratosphere, enhancing both extinction and absorption of smoke. Here, we quantified how particle size and coating, via internal mixing of non-absorbing species on black carbon, influence pyroCb smoke radiative forcing and stratospheric heating using core-shell Mie calculations coupled with chemical transport and radiative transfer models. Airborne measurements indicate a number median radius of 0.25 mu m for days-old pyroCb smoke, larger than typical lower tropospheric smoke. This larger size approximately doubles aerosol extinction and enhances shortwave radiative forcing at the top of the atmosphere by 35%-60%, more consistent with satellite-observed radiative fluxes following 2019-2020 Australian pyroCb event. Coating enhances stratospheric heating by 1-1.5 K, twice the enhancement caused by larger particle size, yielding better agreement with stratospheric temperature anomaly observed from satellite during the first 3 months after injection.

Accurate retrieval of nighttime water cloud microphysical properties, including cloud optical thickness (COT) and cloud effective radius (CER), remains a long-standing challenge. This study develops a physics-informed machine-learning framework, UI-Cloud (Unified Illumination Cloud Microphysics Retrieval Framework), to retrieve nighttime COT and CER from observations of the Visible Infrared Imaging Radiometer Suite (VIIRS). The framework unifies solar and lunar illumination regimes by training on daytime data and applying physically constrained corrections to nighttime radiance inputs. The VIIRS Day/Night Band (DNB) top-of-atmosphere reflectance serves as a bridging variable linking daytime and nighttime conditions. Lookup tables generated using the UNified Linearized Vector Radiative Transfer Model (UNL-VRTM) quantify day–night brightness temperature (BT) offsets and enable dynamic correction of residual solar contributions in the 4.065 μm thermal emissive band (M13).,The daytime model demonstrates strong agreement with VIIRS standard retrievals (R = 0.98). Direct application to nighttime data results in CER overestimation, whereas incorporation of the UNL-VRTM-based M13 BT correction yields nighttime global mean values (COT = 16.0; CER = 16.8 μm) that are consistent with daytime references (14.6 and 15.9 μm). Validation against liquid water path (LWP) retrievals from the Advanced Microwave Scanning Radiometer 2 (AMSR2) demonstrates RMSE reduction from 261.9 to 147.9 g m⁻² and bias decrease from 180% to 62%. Retrieval accuracy is high for Moon Illumination Fraction (MIF) above 70% and remains robust down to approximately 50%, enabling the UI-Cloud framework to extend diurnal cloud microphysical observations beyond daytime-only conditions into lunar-illuminated nights.

The high temporal resolution of Himawari-8 (H8) geostationary satellite has distinctive advantages in capturing diurnal variation of aerosol properties. However, the operational H8 aerosol products including Aerosol Optical Depth (AOD) are subject to considerable uncertainties due to assumptions and simplified surface reflectance. To fully exploit the H8 multi-spectral measurements, a data-driven retrieval framework based on a Deep Belief Network (DBN) was developed to simultaneously retrieve aerosol optical and microphysical parameters. By directly modeling H8 spectral reflectance at the top of the atmosphere (TOA) with matched AERONET products in 82 sites across diverse surface types and emission sources, four aerosol parameters including AOD, fine and coarse AOD (FAOD and CAOD), and single scattering albedo (SSA) are retrieved. Ground-based validation shows that the H8 DBN retrievals achieve high accuracies for AOD and FAOD, with correlation coefficients of 0.935 and 0.930, respectively, substantially outperforming the operational JAXA product (R = 0.703 for AOD). The retrieved CAOD and SSA also show good agreement with AERONET observations. Comparisons with MODIS and JAXA H8 aerosol products show very high consistency in both spatial and temporal variations. H8 DBN retrievals can clearly distinguish biomass burning smoke and dust plumes by their particle size and absorption. Additionally, hourly H8 DBN retrievals accurately reflect diurnal variations of aerosol properties. Consequently, the developed DBN-based approach provides a flexible and robust retrieval method for dynamic aerosol properties over East Asia.

This paper explores gaps, opportunities and future technology needs for satellite Earth Observation (EO) to address wildland fire science and applications for the next Decadal Survey time-period focusing on: 1) pre-fire fuels, 2) active fire monitoring, 3) smoke and aerosol observation, and 4) post-fire recovery. This paper stems from a workshop of over 40 experts in wildland fire science and applications. We find that advancing wildland fire science and applications for the 2027 – 2037 Decadal Survey time-period and beyond, will require more dynamic, integrated, and interoperable EO systems. There is a need to move beyond static and abstracted variables, towards validated, precise and meaningful measurements. High accuracy thermal, imaging spectroscopy, LiDAR, SAR, and (shortwave) optical and spectrometer observations—combined with coordinated field campaigns for validation and calibration—will enable the retrieval of detailed, physically meaningful measurements of fire behavior and impacts. Meeting these needs also demands integrated and interoperable EO architectures that provide concurrent measurements of fuels, weather, and active fire conditions of combustion processes and energetic distribution, leverage AI/ML for low-latency onboard processing and adhere to standardized data sharing principles. These advances will allow next-generation fire prediction models to ingest dynamic EO inputs toward process-level simulation of fire behavior and smoke modeling, and support near-real-time operational forecasting and tactical responses. As the EO landscape becomes more diverse and with the increase of fire frequency and intensity, new dedicated collaboration paradigms cross-cutting Earth system spheres (atmosphere, hydrosphere, biosphere, cryosphere, geosphere) and integrating public, private, and nonprofit sectors will be essential to deliver comprehensive wildland fire observations required for future resilience.

The Tropospheric Emissions: Monitoring of Pollution (TEMPO) instrument provides continuous, high-resolution observations of atmospheric pollutants over North America from geostationary orbit. This study introduces an on-orbit spectral calibration algorithm implemented in the TEMPO Version 3 Level 0-1 processor, covering both operational irradiance and radiance wavelength calibrations and offline slit function retrievals. Irradiance wavelength calibration accuracy was evaluated, with the TSIS-1 hybrid solar reference spectrum chosen due to its low fitting residuals. Accordingly, first- and second-order Chebyshev-polynomial fittings are applied to UV and VIS, respectively, to derive the wavelength grid. Earth-view radiance wavelength calibration updates the wavelength grid based on the latest solar irradiance calibration result by fitting a wavelength shift. To optimize efficiency and accuracy, a narrow spectral window of 100 channels (320-340 nm for UV and 630-650 nm for VIS) was selected, with wavelength shift uncertainties of 0.002 nm (UV) and 0.006 nm (VIS). Radiance calibration results shows that the wavelength shifts of inhomogeneous pixels vary relatively significantly. We perform a 22-month trend analysis of the TEMPO solar irradiance spectral performance. Compared to first light, the wavelength shift gradually increases, reaching 0.08-0.09 nm in July 2024, and then remains stable. The offline slit function parameters, retrieved from several narrow spectral windows using a super-Gaussian function, show minor variations during the 22-month on-obit operation and did not deviate significantly from prelaunch. This study supports the long-term L1b data processing for TEMPO and provides an instrument spectral calibration framework applicable for future geostationary orbit spectrometers.

Soil heavy metal concentrations are spatially heterogeneous and strongly influenced by soil physicochemical properties. This study aimed to identify the key physicochemical properties affecting the performance of UAV-based hyperspectral models for estimating Zn, As, Cu, and Pb concentrations in soil. Soil pH, soil organic matter (SOM), and soil moisture content (SMC) were combined with spectral features to construct different input scenarios. The best-performing model for each metal was subsequently interpreted using Shapley additive explanations (SHAP). The predictive accuracy of the models for Zn, As, Cu, and Pb improved by 10

Digital platforms have become primary arenas for international and intercultural communication. Yet practitioners still lack a transparent way to prioritize competing interventions when goals such as reach, authenticity, trust, and safety conflict. This study develops and tests a multi-criteria decision framework that compares the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) and the VlseKriterijumska Optimizacija I Kompromisno Resenje (VIKOR). These two methods were selected because they represent complementary decision logics widely used in multi-criteria decision analysis: TOPSIS ranks alternatives according to their geometric distance from an ideal solution, while VIKOR identifies compromise solutions that minimize the maximum regret among competing criteria. Employing both methods enables the study to capture different trade-off perspectives that arise in international communication planning, thereby improving the transparency and interpretability of intervention prioritization. Using China-anchored outbound communication towards ASEAN, EU, and MENA audiences as illustrative contexts, we construct a criteria system that integrates validated intercultural outcomes (e.g., openness, empathy, interaction management, perceived authenticity) with operational and platform-related factors (e.g., algorithmic visibility, translation and subtitling quality, accessibility, moderation and civility, and cost). A multidisciplinary expert panel (N = 100) provides importance weights and performance assessments for a portfolio of interventions, including localization depth, creator partnerships, cross-platform sequencing, and community moderation design. From these data, we build cleaned, normalized decision matrices and apply TOPSIS (closeness to an ideal solution) and VIKOR (compromise under conflicting criteria), followed by unified robustness and sensitivity analyses. Across contexts, both methods converge on a similar top tier of interventions: strategies that combine deep localization, sustained partnerships with credible creators, and proactive community moderation consistently outperform cost or speed-optimized options. Method divergences are confined to mid-ranked alternatives and align with theoretical differences between ideal-distance and regret-minimizing logics. The findings demonstrate that a dual TOPSIS-VIKOR lens can provide defensible, practice-ready rankings and diagnostics for international communication planning, and the study offers a replicable workflow that other institutions can adapt to their own platforms, audiences, and strategic priorities.

Abstract. We propose a physics-informed Transformer framework to correct biases in the Aerosol Extinction Coefficient (AEC, km-1) profiles simulated by GEOS-Chem. Unlike standard Transformer, our framework features a dual-stream architecture with explicit physical constraints. It employs Gated Feature Fusion to integrate vertical structures (combining GEOS-Chem priors with MERRA-2 profiles) by dynamically identifying height-dependent drivers, and leverages Cross-Attention to incorporate MERRA-2 surface environmental constraints for modulating AEC vertical reconstruction with synoptic contexts. This approach effectively predicts systematic biases relative to Cloud-Aerosol Lidar with Orthogonal Polarization satellite observations and resolves AEC profiles, surpassing methods retrieving only aerosol layer heights. "Leave-One-Year-Out" validation over East Asia during 2017–2019 demonstrates significant AEC fidelity improvements, increasing R from 0.49–0.53 in the GEOS-Chem simulations to 0.66–0.73 and reducing RMSE by approximately 25 %. The model effectively mitigates over-diffusion, significantly reducing AEC simulation biases in the critical near-surface layer while restoring smoothed biomass burning and dust plumes. Additionally, it exhibits robust cross-continental transferability, reproducing bias patterns over North American domain (R=0.70) without retraining, confirming the internalization of universal physicochemical relationships linking atmospheric states to simulation biases. Furthermore, interpretability analysis establishes a feedback loop from data-driven correction to physical model improvement. The model identifies temperature and sensible heat flux as primary drivers to constrain boundary layer mixing, and uses environmental proxies (e.g., vegetation indices) to diagnose deficiencies in dust uplift and secondary aerosol formation. These insights provide a physical basis for refining parameterization schemes in chemical transport models.

Acute exposure to surface nitrogen dioxide (NO 2 ) poses substantial global health risks. Using machine learning, we generated the first global, daily, 1-km-resolution, gap-free surface NO 2 data from satellite observations for 2018–2022. Surface NO 2 shows strong day-to-day variability, with fluctuations reaching 73% of the global mean. We identify pronounced pollution hotspots and large global inequalities in acute NO 2 exposure: although only 28% of inhabited land exceeds the WHO daily guideline (25 μg m -3 ) at least once per year, these exceedances disproportionately affect more than 61% of the global population and nearly all megacities (98%). When exceedances are aggregated over 7-day and 30-day windows, 77% and 56% of megacities remain exposed. The decline in NO 2 -affected areas outpaces reductions in population exposure, highlighting the challenge of mitigating impacts in densely populated urban centers. Acute NO 2 exposure caused approximately 576,000 (95% CI: 473,000–678,000) premature deaths globally in 2019. COVID-19 strictest lockdowns in 2020 temporarily reduced NO 2 levels, but rebounds occurred in 91% of countries by 2022. These results reveal the widespread and under-recognized burden of acute NO 2 exposure and emphasize the need for high-resolution global monitoring to support effective pollution control.

Wildfire smoke is an increasingly important contributor to urban air pollution and public health risk, especially in ozone (O3) non-attainment areas like Chicago. This study assessed the impact of the 2023 wildfire smoke on ground-level O3 concentrations and associated mortality across Chicago's 77 community areas. We integrated NOAA's Hazard Mapping System smoke classifications, high-resolution downscaled O3 data, and GridMET meteorological data to construct a daily community-level dataset for 2014-2023. We estimated counterfactual O3 levels in the absence of wildfire smoke using matching, linear regression, and machine learning models. We separated the effect of smoke on O3 from those caused by meteorological variability. O3 concentrations increased with smoke density, peaking under medium smoke conditions, while the largest smoke-attributable increase (6.7 ppb) occurred under heavy smoke. Estimated daily all-cause mortality rates attributable to smoke-enhanced O3 followed a similar trend, reaching 0.24 deaths per 100,000 population per day under heavy smoke. Spatial

Black carbon (BC) aerosols remain among the most uncertain contributors to anthropogenic climate forcing, as their radiative impact depends sensitively on microphysical evolution and atmospheric loading. This study presents a physics-informed, machine learning (ML) approach to estimate clear-sky BC top-of-atmosphere direct radiative forcing (BC TOA) at high spatial-temporal resolution while retaining physical interpretability. The study derives necessary optical properties for radiative transfer modeling (RTM), by constraining them with multi-platform, multi-waveband observations and their associated uncertainties. The RTM outputs are then used to train the ML surrogates and applied over two contrasting urban agglomerates-Xuzhou, China, and Dhaka, Bangladesh. The ML framework closely reproduces physics-based regional climatological mean (-17.6 +/- 2.2 W m(-2) versus -17.4 +/- 2.6 W m(-2) over Xuzhou; -14.9 +/- 1.1 W m(-2) versus -15.0 +/- 1.2 W m(-2) for Dhaka), while achieving high predictive fidelity R-2 > 0.95; RMSE similar to 1.5-1.8 W m(-2) and strong cross-regional consistency (r > 0.9). SHAP based predictor attribution indicates that BC TOA estimates are strongly associated with BC aerosol optical depth (BCAOD), column number density, and mixing state, with their relative contributions varying non-linearly across cooling-to-warming regimes. Crucially, similar BC loading can yield contrasting absorption-scattering dynamics across region, which are not captured by simplified forcing parameterization. To test transferability, the combined ML model (trained in Xuzhou, China and Dhaka, Bangladesh) was evaluated zero-shot on two additional regions with contrasting aerosol microphysical conditions represented by Delhi, India (urban and agricultural burning sources) and Mongu, Zambia (strong savanna fires). While transference to Delhi is reasonable (Adj. R-2 = 0.91, RMSE = 2.3 W m(-2)), there is a systematic underestimate at Mongu (Adj. R-2 = 0.83; MBE = -4.2 W m(-2)). Feature-space overlap analysis attributes this degradation to a distributional mismatch in key microphysical predictors. Retraining on an expanded dataset including all four regions preserves urban performance while reducing Mongu RMSE by 68 % and bias from -4.2 to -0.8 W m(-2). Together, the physics-informed ML framework and the multi-domain evaluation provide an efficient and transferable tool for constraining BC radiative impacts across real-world heterogeneity. The study also offers new mechanistic insight into how regional properties reshape BC radiative forcing.