Jiang Zhu

Ocean heat content (OHC) dominates Earth’s energy imbalance, but its uncertainty has not been fully understood. Here, we construct a Large-Ensemble to comprehensively quantify the OHC uncertainty from global to regional scales since 1955, and consider eight groups of error sources based on the community’s efforts. The global upper 2000 m OHC estimated uncertainty (defined as 5–95% quartile error range) has reduced six-fold since 1955 from ~111 ZJ (1955–1965) to ~19 ZJ (2014–2023), mainly due to reduced mapping/sampling uncertainty. This indicates a robust upper 2000 m OHC increase since 1955 of 449 [385, 519] ZJ. Quality-Control (QC) has become a primary error source recently, due to the difficulties detecting outliers in eddy-rich regions. Trend uncertainty sources differ from annual OHC: for 2005–2023, the upper 2000m trend is 10.9 [9.2, 11.4] ZJ yr ⁻¹ (0.68 [0.57, 0.71] Wm ⁻² ), with uncertainties from spatial gaps, mapping, QC, and climatology choice. The ocean warming rate increase from 2005–2023 is small but robust: 0.34 [0.19, 0.46] Wm ⁻² dec ⁻¹ . These comprehensive uncertainty estimates guide future improvements in the global ocean observing system.

Emission inversion based on surface or satellite data assimilation is an important way to constrain NO _x emissions, thereby facilitating the development of effective mitigation strategies and assessment of CO ₂ emissions. However, few studies have been conducted to compare their inversion performance under the same assimilation framework, and the efforts to combine these two data sources remain limited. In this study, a multi‐source data assimilation system based on an ensemble Kalman filter and the Community Multiscale Air Quality Modeling System is developed, which can simultaneously assimilate surface and satellite observations. A nearly two‐fold discrepancy is identified between the inversion results of surface and satellite assimilation, where assimilating one type of observation often degrades performance for the other. Although direct combined assimilation of surface and satellite observation simultaneously optimized surface and column concentration simulations, it yields only moderate performance. The inconsistent simulation errors in surface and column concentrations are found to be the main reasons for these issues and may reflect the contrast emission errors between near‐surface and upper layers. To address this, we proposed a three‐dimensional (3D) combined assimilation method by utilizing the varied vertical sensitivity of surface and satellite data to emissions. The method shows capability in reconciling the conflicts inherent in single‐source data assimilation and highlights potential biases in single‐source inversion results and a larger fraction of surface NO _x emissions in China was identified. These findings highlight the necessity of multi‐source data assimilation and the consideration of vertical emission uncertainty in future inversion studies.

Accurate simulation of regional carbon dioxide (CO ₂ ) concentrations is critical for urban carbon monitoring, inverse modeling and mitigation. However, large uncertainties persist due to differences in anthropogenic emission inventories in China. Focusing on Jiangsu Province in the Yangtze River Delta, where the mean inter‐inventory spread in annual city‐level total emissions exceeds 60% and spatial discrepancies are substantial, we assessed how six widely used inventories affect modeled CO ₂ fields. Using a 3‐km WRF‐Chem‐VPRM framework, we performed simulations for July and December 2022. To isolate the impact of inventory‐related uncertainties, we designed sensitivity experiments that separately perturbed inventory selection, emission magnitude, spatial distribution, and temporal and vertical allocation used in the inventory‐to‐model matching. All experiments were driven by identical meteorological conditions to ensure comparability. Model outputs showed good consistency with meteorological observations, CarbonTracker near‐surface CO ₂ , and OCO‐2 XCO ₂ ( R ≈ 0.83), while also capturing physically reasonable near‐surface diurnal behavior. Inventory selection led to a nighttime urban domain‐averaged standard deviation of 8.2 ppm. Sensitivity results indicated that spatial allocation differences contributed more to modeled CO ₂ variability than total emission magnitude. Under stable boundary‐layer conditions, vertical allocation emerged as a key uncertainty source, producing 25–50 ppm differences in surface CO ₂ . These results demonstrate that inventory‐to‐model matching, especially vertical allocation, can exceed the impact of inventory selection for high‐resolution urban CO ₂ simulations under stable nighttime conditions. This study provides a quantitative basis for diagnosing inventory‐induced variability and supports the development of fine‐resolution, vertically resolved inventories for robust urban CO ₂ modeling and future inversion efforts.

Ocean heat content (OHC) is a crucial climate indicator and provides an estimate of Earth’s energy imbalance, but its accuracy and uncertainty have not been fully understood. Here, we constructed a Large-Ensemble to comprehensively quantify the OHC uncertainty from global to regional scales since 1955, and consider eight groups of error sources based on the community’s efforts. We find a six-fold reduction in the global upper 2000 m OHC estimated uncertainty (5–95% quartile error range) since 1955 from ~ 111 ZJ (1955–1965) to ~ 19 ZJ (2014–2023), attributed to the reduction of errors from all sources especially mapping/sampling. This indicates a robust upper 2000 m OHC increase since 1955 of 449 [385, 519] ZJ. Quality-Control (QC) has emerged as a primary error source in the recent decade, because of the difficulties in detecting outliers in large perturbations in eddy-rich regions. The trend uncertainty has different sources from the annual OHC: from 2005–2023, the upper 2000m OHC trend is 10.9 [9.2, 11.4] ZJ yr ^− 1 or equivalently 0.68 [0.57, 0.71] Wm ^− 2 , with uncertainty mainly stemming from spatial gaps and mapping errors, QC, and climatology choice. An increase in ocean warming rate from 2005–2023 is small but robust: 0.34 [0.19, 0.46] Wm ^− 2 dec ^− 1 . The comprehensive OHC uncertainty estimates provide a basis for model evaluation and a guide for future improvements in the global ocean observing system.

Accurately simulating orography‐induced mountain waves over steep terrain, such as the Tibetan Plateau (TP), remains a major challenge for numerical weather prediction (NWP) models due to grid distortions inherent in traditional terrain‐following coordinates. To address this issue, we developed an AI‐driven adaptive mesh refinement (AMR) framework within the Fluidity‐Atmosphere model, which employs a 3D unstructured mesh to mitigate geometric distortions. A Long Short‐Term Memory (LSTM) neural network is integrated to enhance the AMR process, replacing traditional adaptation criteria with data‐driven predictions. A series of idealized 2D and 3D experiments demonstrate that both the traditional AMR and LSTM‐driven approaches reproduce mountain wave dynamics with higher efficiency than fixed mesh. Furthermore, the LSTM model suppresses numerical noise near terrain, preventing spurious over‐refinement. In realistic simulations over the TP, the LSTM‐enhanced model successfully captured the life cycle of mountain waves, reproducing key physical features such as vertical velocity structures, wave amplitude decay, and upstream phase tilt. Comparative tests further revealed efficiency gains of up to 71.4% over fixed meshes and 23.8% over traditional AMR at high resolution, alongside accuracy improvements in vertical velocity, potential temperature, and wave propagation. These findings validate the LSTM‐based AMR framework as a robust and efficient approach for atmospheric simulations over complex terrain. By intelligently allocating computational resources while preserving physical accuracy, this method offers a scalable pathway toward next‐generation atmospheric modeling, with future applications targeted at realistic meteorological conditions over the TP.

The accuracy of global forecasting of atmospheric composition is essential for protecting public health and advancing climate research. Carbon monoxide (CO), a key pollutant with indirect greenhouse effects, requires timely and accurate prediction. We propose Duo‐AttnOPNets, an operational framework that combines a deep learning‐based forecasting module Duo‐AttnForeNet with a training‐free data assimilation module Duo‐AttnVarNet. Duo‐AttnForeNet employs CSLSTM blocks—built from ConvLSTM cells and dual attention mechanisms—to effectively capture spatio‐temporal dynamics. Duo‐AttnVarNet leverages automatic differentiation and GPU acceleration to enable efficient four‐dimensional variational (4D‐Var) assimilation without extensive manual adjoint coding. We evaluate Duo‐AttnOPNets against the Integrated Forecasting System for Atmospheric Composition (C‐IFS). Results show that Duo‐AttnOPNets achieves comparable or superior accuracy in both forecasting and assimilation, while generating 5‐day forecasts within seconds on a single GPU. These findings demonstrate its potential for real‐time, scalable, and accurate CO forecasting, marking a promising advance in integrating deep learning with traditional variational methods for operational atmospheric modeling.

High-resolution (HR) sea surface temperature (SST) is crucial for understanding ocean dynamics, climate variability, and nearshore ecosystems. While demand for fine-scale SST data in coastal regions grows, HR observations remain limited, and numerical modeling is constrained by prohibitive computational costs. Efficient and accurate downscaling approaches are therefore essential, and recent advances in artificial intelligence (AI) offer promising alternatives. However, current AI-based methods often use artificially degraded low-resolution (LR) data, overlooking the systematic mismatches where LR simulations misrepresent complex, fine-scale ocean features. This study employed a Convolutional Block Attention Module-enhanced UNet model (CBAM-UNet) trained on realistic LR-HR data pairs from the three-layer nested regional ocean model system (ROMS) simulations to enhance practical reliability. The LR multi-mode inputs include SST and sea surface currents, which help encode essential physical ocean processes. Compared with bilinear interpolation and a traditional UNet model using SST-only input, the proposed model reduced Root Mean Square Error (RMSE) by 21.93% and achieved a spatial Pearson Correlation Coefficient (R) of 0.86. In addition, interpretability analysis revealed the contribution of each input channel, aligning well with temperature transport and seasonal variability, and confirming the underlying natural physical constraints learned from real-world data. Beyond pixel-value precision, the physically interpretable behavior of the multi-mode downscaling method demonstrates its capability to reconstruct accurate and dynamically consistent HR SST fields, which is vital for operational applications.

Multiple climate-related stressors affect the ocean, including warming, acidification, deoxygenation and variations in salinity, with profound effects on Earth system cycles, marine ecosystems and human well-being. Nevertheless, a global perspective on the combined impacts of these changes on both surface and subsurface ocean conditions remains unclear. Here, applying a time-of-emergence methodology to observed physical and biogeochemical variables, collectively referred to as compound climatic impact-drivers, we show individual and compound ocean state changes have become increasingly prominent globally over the past 60 years. In particular, observations show the simultaneous emergence of compound climatic impact-drivers in regions spanning the subtropical and tropical Atlantic, the subtropical Pacific, the Arabian Sea and the Mediterranean Sea. We highlight extensive exposure of different ocean layers to compound emergence, characterized by significant intensity, duration and magnitude. These results provide a comprehensive framework and perspective to illustrate the ocean’s vulnerability to pervasive and interconnected changes in a warming climate.