Climatology Research Articles

How is climate science used to inform national-level adaptation planning in southern Africa?

ABSTRACT Climate model projections are increasingly being included within adaptation planning across sectors but there is limited understanding of how they are being used, and to what extent they improve adaptation planning. This article investigates how climate projections inform adaptation planning processes in the National Communications (NCs) to the United Nations Framework Convention on Climate Change (UNFCCC) in 16 southern African countries through a document analysis together with 18 key informant interviews. The study found that all the NCs include future climate model projections for the mid and/or late twenty-first century and focus on average changes in temperature and precipitation; meanwhile, the models, scenarios and time periods used vary between countries. The climate analysis is often detached from the adaptation planning section of the NC. The impacts and adaptation sections focus on key risks, such as flooding and drought and have limited recognition of uncertainties, suggesting plans are made without considering the full range of plausible futures. The role of climate science in the adaptation planning process varies, with some evidence of highly collaborative processes, resulting in evidence-based adaptation options across sectors and scales. In many cases, boundary agents play a key role in interpreting and communicating climate projections. We suggest that providing additional climate projections is unlikely to improve national adaptation planning, despite their scientific benefits. Instead, the focus should be on developing approaches and collaborative processes to distil and interpret climate information in different contexts, to enable decision-makers to understand the range of plausible futures, including changes in climate alongside growing populations, urbanization and changing economies.

Global-Scale Multidecadal Variability in Climate Models and Observations. Part I: Forced Response

Abstract Twentieth-century climate variations exhibit, on top of a secular global warming trend, multidecadal variability with globally coherent patterns. Identifying and attributing such patterns requires one to combine modern reanalyses of atmospheric and oceanic observations with estimates of climate response to variable forcings based on ensemble simulations of the twentieth-century climate by the state-of-the-art global coupled models. This contribution is the first installment of the series of papers in which an identical sequence of pattern-recognition methods is applied to a 38-model ensemble of historical simulations within phases 5 and 6 of the Coupled Model Intercomparison Project (CMIP5/6) and to two twentieth-century reanalysis datasets to succinctly describe their global-scale multidecadal variability. The focus here is on characterizing the globally distributed forced response in near-surface air temperature (SAT) and sea level pressure (SLP) using a combination of the signal-to-noise-maximizing pattern (S/NP) filtering and linear-regression-based rescaling. A particular novel aspect of this work lies in isolating the forced-response structures common across the entire model ensemble versus the residual responses of individual models, thus setting the stage for exploring whether the latter responses can serve as a viable explanation of the observed multidecadal climate teleconnections. CMIP5/6 models’ common forced response is dominated by two S/NP modes, whose amplitudes differ from model to model. The S/NP-1 describes polar and land-intensified global warming with a slower warming rate prior to mid-60s and a faster warming afterward. The S/NP-2 component is characterized by a pronounced multidecadal variability without much of a centennial-scale trend and exhibits antisymmetric temperature response between the hemispheres. The two mode’s SLP signature is associated with the downward trend over the Southern Ocean and Antarctica after mid-60s and is much weaker than the SLP background internal variability elsewhere. The CMIP models that have large contributions from S/NP-1 and small contribution from S/NP-2 to their forced response tend to exhibit large twentieth-century global-mean SAT trend and vice versa. The residual forced responses of individual models unexplained by S/NPs 1 and 2 do exhibit multidecadal variations with the appearance of long-range teleconnections, albeit with a much smaller amplitude than the observed climate’s globally connected deviations from the model estimated forced trends. Significance Statement Understanding the discrepancies between the observed and simulated multidecadal climate variability constitutes one of the pressing problems in climate science. These discrepancies manifest a coherent global signal which may reflect biases in the simulated response to external forcings or point to the observed internal variability unmatched by the climate models. This study provides a compact characterization of the global-scale forced responses in terms of just a few dominant patterns common across an ensemble of state-of-the-art climate models and the residual forced variability of individual models. The relative amplitudes of these patterns gauge climate sensitivity of individual models, while the residual signals have by far insufficient magnitude to explain the observed multidecadal climate undulations on top of the secular warming trend.

Exploring Research Trends on Climate Change: Insights into Port Resilience and Sustainability

Ports play a critical role in global trade, yet they are both contributors to and recipients of climate change. This study conducts a bibliometric analysis to examine the relationship between climate change and port operations, identifying key themes, methodological approaches, and research gaps within the literature, including the need for standardized vulnerability assessments, policy-driven adaptation strategies, and a more integrated, cross-regional approach to port resilience. A bibliometric analysis was performed using peer-reviewed journal articles indexed in the Web of Science. The study employed keyword-based searches, document screening, and bibliometric techniques such as publication trends, keyword co-occurrence analysis, citation impact evaluation, and research clustering. Studies were classified based on research focus, methods, data sources, and geographic scope. The analysis reveals three major research phases: theoretical contributions, empirical expansion, and an intensified focus on adaptation and resilience. Key trends include sea-level rise, port vulnerability, climate adaptation strategies, and emission reduction. Quantitative methods dominate, though qualitative studies contribute to governance and policy discussions. Research on port resilience is becoming increasingly interdisciplinary, integrating engineering, environmental science, policy, and maritime economics. The study identifies the United States, China, Europe, and the Mediterranean as the most focused regions in port-climate research. Understanding climate risks and adaptation strategies is essential for policymakers and stakeholders. The findings highlight the current push for resilience planning, emission mitigation, and policy interventions, with efforts by governmental and international organizations well documented in climate research. This study provides insights into emerging trends and research gaps, enhancing discussions on sustainable port management and climate resilience.

Corrigendum: World Climate Research Programme lighthouse activity: an assessment of major research gaps in solar radiation modification research

Corrigendum: World Climate Research Programme lighthouse activity: an assessment of major research gaps in solar radiation modification research

How Epistemic Beliefs about Climate Change Predict Climate Change Conspiracy Beliefs

Belief in climate change conspiracy theories (CCCT) can undermine support for measures against climate change. In two studies, we therefore aim to gain a clearer understanding of the factors that contribute to CCCT. A significant factor associated with CCCT is distrust in science, which is also correlated with epistemic beliefs (EBs) (e.g., beliefs are about the nature of knowledge and the process of knowing). EBs influence how individuals respond to knowledge claims, address contradictory evidence, and integrate new information. We hypothesize that EBs are linked to belief in CCCT via distrust in science. To test this hypothesis, we conducted one correlational study and one experimental study (n = 404 and n = 104, both pre-registered). Study 1 found that participants were more likely to endorse climate-related conspiracy beliefs if they viewed climate knowledge as tentative, relied on intuition to understand climate change, and had weaker beliefs in the interconnectedness of climate knowledge and its reliance on experts and scientific sources. As anticipated, distrust in climate science significantly mediated the relationships between the EBs subscales and belief in CCCT. Additionally, political ideology moderated the effect of believing knowledge originates from experts and science on distrust in climate science — this effect was pronounced among participants identifying with the political center while being weaker among left-wing participants. In Study 2, we were unable to establish a causal relationship between certainty beliefs and belief in CCCT. In conclusion, we suggest that educational initiatives aimed at fostering EBs may reduce science distrust and conspiracy beliefs.

STEM-LTS: Integrating Semantic-Temporal Dynamics in LLM-driven Time Series Analysis

Time series forecasting plays a crucial role in domains such as finance, healthcare, and climate science. However, as modern time series data become increasingly complex, featuring high dimensionality, intricate spatiotemporal dependencies, and multi-scale evolutionary patterns, traditional analytical methods and existing predictive models face significant challenges. Although Large Language Models (LLMs) excel in capturing long-range dependencies, they still struggle with multi-scale dynamics and seasonal patterns. Moreover, while LLMs' semantic representation capabilities are rich, they often lack explicit alignment with the numerical patterns and temporal structures of time series data, leading to limitations in predictive accuracy and interpretability. To address these challenges, this paper proposes a novel framework, STEM-LTS (Semantic-TEmporal Modeling for Large-scale Time Series). STEM-LTS enhances the ability to capture complex spatiotemporal dependencies by integrating time series decomposition techniques with LLM-based modeling. The semantic-temporal alignment mechanism within the framework significantly improves LLMs' ability to interpret and forecast time series data. Additionally, we develop an adaptive multi-task learning strategy to optimize the model's performance across multiple dimensions. Through extensive experiments on various real-world datasets, we demonstrate that STEM-LTS achieves significant improvements in prediction accuracy, robustness to noise, and interpretability. Our work not only advances LLM-based time series analysis but also offers new perspectives on handling complex temporal data.

Quantum-Enhanced AI and Machine Learning: Transforming Predictive Analytics

Background: Artificial Intelligence (AI) and Machine Learning (ML) have significantly transformed predictive analytics across domains such as healthcare, finance, and environmental sciences. However, traditional ML models face limitations when dealing with large-scale, high-dimensional datasets, particularly due to computational inefficiencies and the "curse of dimensionality." Quantum computing offers promising solutions by leveraging superposition and entanglement to perform complex computations at exponential speeds. Objective: This study aims to explore how quantum computing can enhance AI and ML models, particularly in the domain of predictive analytics. It proposes a comprehensive Quantum-AI framework that integrates quantum technologies into existing predictive systems to overcome the challenges posed by classical approaches. Methods: The study employed hybrid quantum-classical models using frameworks like Qiskit, TensorFlow Quantum, and Pennylane. Datasets were sourced from real-world applications in finance, healthcare, and climate science. Experiments were designed to compare classical and quantum- enhanced models on metrics such as accuracy, scalability, and computational time. Specific algorithms used include Quantum Support Vector Machines (QSVM) and Quantum Neural Networks (QNN). Result: The proposed quantum-enhanced models showed significant improvements: • Accuracy increased from 85% (classical) to 92% (quantum) in finance-related predictions. • Diagnostic models in healthcare achieved a 94% accuracy rate and reduced training time by 50%. • Climate modeling tasks achieved 90% improved accuracy and 20% faster simulation times. The results validate the effectiveness of quantum models in processing large, complex datasets more efficiently. Conclusion: Quantum-enhanced AI presents a transformative approach to predictive analytics by addressing the core limitations of classical ML systems. This study demonstrates the practical feasibility and industrial relevance of integrating quantum computing into AI workflows. It opens avenues for further research in developing scalable, domain-specific quantum algorithms and hybrid AI models for real-world applications.

Quantum computing, the genetic history of the Maghreb, deep learning for climate science, and polio vaccinations

Quantum computing, the genetic history of the Maghreb, deep learning for climate science, and polio vaccinations

A Framework for Trauma-Informed Climate Research: Interrupting the Relationship Between Climate Trauma and Social Inertia

A Framework for Trauma-Informed Climate Research: Interrupting the Relationship Between Climate Trauma and Social Inertia

Determining historic (1950 - 2000) average precipitation and temperature for Pakistan by using climate downscaling technique

There is an immense hydropower potential in Pakistan. To utilize those resources effectively, it’s compulsory to know about climate history that helps in hydropower project feasibility. For that, there is an assessment required of a specific site from a hydrology perspective i.e. water level or evaporation rate, where spatial data regarding different parameters such as precipitation and temperature is compulsory. Based on this, a climate downscaling technique is used in this paper that requires the input time series data sets of temperature and precipitation, based on 30 arc seconds and 0.5˚ resolution. A downscaled dataset has been generated by using time series input data sets of the Global Precipitation Climatology Centre (GPCC) and Climate Research Unit (CRU) that are of 0.5˚ resolution. Along with such low-resolution datasets, a 30˚ high-resolution time series data sets of WorldClim are also used for temperature and precipitation. A downscaled historical (1950 - 2000) gridded data of precipitation and temperature for Pakistan is generated through the statistical climate downscaling technique to interpolate the coarse resolution data into fine spatial resolution data with the help of Piecewise Cubic Hermite Interpolation Polynomial (PCHIP) interpolation scheme. These downscaled historical precipitation and temperature data in ASCII format are programmed in MATLAB, defining the boundary region of Pakistan to calculate their average annual, seasonal, and monthly temperature and precipitation values. The average annual, seasonal, and monthly precipitation and temperature historical data are plotted for Pakistan using a spreadsheet generated in MATLAB. Such plotted data can be used in the prediction of water stream flow rate and evaporation rate by using hydrology models in the future that help in assessing the hydropower of the Pakistan before commencement of the hydro projects.

The SAIL dataset of marine atmospheric electric field observations over the Atlantic Ocean

Abstract. A unique dataset of marine atmospheric electric field observations over the Atlantic Ocean is described. The data are relevant not only for atmospheric electricity studies, but more generally for studies of the Earth's atmosphere and climate variability, as well as space–Earth interaction studies. In addition to the atmospheric electric field data, the dataset includes simultaneous measurements of other atmospheric variables, including gamma radiation, visibility, and solar radiation. These ancillary observations not only support interpretation and understanding of the atmospheric electric field data, but also are of interest in themselves. The entire framework from data collection to final derived datasets has been duly documented to ensure traceability and reproducibility of the whole data curation chain. All the data, from raw measurements to final datasets, are preserved in data repositories with a corresponding assigned DOI. Final datasets are available from the Figshare repository (https://figshare.com/projects/SAIL_Data/178500, SAIL Data, 2025), and computational notebooks containing the code used at every step of the data curation chain are available from the Zenodo repository (https://zenodo.org/communities/sail, Project SAIL community, 2025).

EEAR-Clim: a high-density observational dataset of daily precipitation and air temperature for the Extended European Alpine Region

Abstract. The Extended European Alpine Region (EEAR) exhibits a well-established and very high density network of in situ weather stations, hardly attained in other mountainous regions of the world. However, the strong fragmentation of the area into national and regional administrations and the diversity of data sources have hampered the full exploitation of the available data for climate research. Here, we present EEAR-Clim, a new observational dataset gathering in situ daily measurements of air temperature and precipitation from a variety of meteorological and hydrological services covering the whole EEAR. The data collected include time series from recordings of diverse lengths up to 2020, with the longest records spanning up to 200 years. The overall observational network encompasses about 9000 in situ weather stations, significantly enhancing data coverage at high elevations compared to existing datasets and achieving an average spatial density of one station per 6.8 km2 over the 1991–2020 period, the most covered by measurements. Data collected from many sources were tested for quality to ensure the internal, temporal, and spatial consistency of time series, including outlier removal. Data homogeneity was assessed through a cross-comparison of the break points detected by three methods that are well established in the literature, namely Climatol, ACMANT, and the RH test. Quantile matching was applied to adjust inhomogeneous periods in time series. Overall, about 4 % of data were flagged as unreliable, and about 20 % of air temperature time series were corrected for one or more inhomogeneous periods. In the case of precipitation time series, fewer break points were detected, confirming the well-known challenge of properly identifying inhomogeneities in noisy data. The high quality, homogeneity, unprecedented spatial density, and completeness of data and the inclusion of the most recent records are important add-on improvements compared to other observational products available for the EEAR. The dataset (https://doi.org/10.5281/zenodo.10951609, Bongiovanni et al., 2024) aims to serve as a powerful tool to better understand climate change and climatic variability over the European Alps.

Монгол орны хамгийн нам цэгийг шинэчлэн тогтоосон судалгаа

This study aimed to reassess and redefine the lowest point in Mongolia using modern geodetic and topographic research methods. Previously, the lowest officially recognized point in Mongolia was Khukh Lake in Dornod Province, recorded at an elevation of 560 meters above sea level. However, this study identified a newly discovered low point in Altai Soum, Gobi-Altai Aimag, specifically within the "A" section of the Great Gobi Strictly Protected Area. The precise elevation of this newly identified point was measured at 526.745 meters. To ensure the accuracy of this finding, the study employed advanced surveying techniques, including high-precision Global Navigation Satellite System (GNSS) measurements, innovative geodetic methodologies, and detailed topographic mapping. These methods provided reliable and verifiable data to support the revision of Mongolia's official geographic records. The identification of this low point holds significant implications for environmental and hydrological studies. The area functions as a natural depression where seasonal water accumulation may occur, making it a valuable site for climate and ecosystem research. Understanding such depressions is crucial for monitoring water balance, sediment deposition, and ecological dynamics in arid and semi-arid regions like Mongolia. Additionally, this discovery has practical applications for updating Mongolia’s geographical databases and can contribute to environmental education and sustainable tourism initiatives. Given the importance of this newly identified lowest point, it is recommended that the site be formally recognized in Mongolia's official geographic records. Furthermore, long-term monitoring and future research should be conducted to assess its environmental dynamics, potential hydrological significance, and conservation needs. Preserving and utilizing this location for scientific research will provide further insights into the geomorphological and climatic processes shaping Mongolia's landscape.

Understanding of natural sciences teachers on addressing climate change concepts in the curriculum: Evidence from Fundisa for Change Keep It Cool Climate Change Project

Climate change is one of the most pressing international concerns of the twenty-first century. Globally, it affects economies, societies, and the environment. It is imperative that all sectors, including the education sector, embrace climate change mitigation measures. A qualitative case study was conducted to investigate teachers' experiences with integrating climate change concepts into the teaching of natural sciences at the Senior Phase in the Amathole West Education District. Permission to conduct the study was sought from the Department of Education and the institution for ethical considerations. Empirical data were collected through one-on-one semi-structured interviews with five natural sciences teachers from five selected secondary schools in the Amathole West Education District. One participant withdrew after data collection, and her data had to be discarded. The second generation of Cultural Historical Activity Theory (CHAT) was used as the theoretical framework. The thematic data analysis technique was employed to analyse the collected data. Key issues highlighted included teachers' content knowledge regarding climate change in natural sciences, a lack of training, and insufficient clear guidelines on integrating climate change topics. The study recommends regular teacher capacitation workshops on integrating climate change into their natural sciences teaching. Furthermore, it suggests the organisation of webinars to discuss issues related to climate change integration into the teaching of natural sciences, as well as encouraging teachers to collaborate with climate scientists to improve their pedagogical content knowledge.

A Creativity Navigator to Guide Teaching for Creativity: Implementation and Teacher Impacts in a Creativity Collaborative of Schools

ABSTRACTWhile creativity has traditionally been linked to arts education, the importance of developing the ability to think creatively across the school curriculum has grown in prominence (James et al. 2019), reflected by its inclusion in OECD PISA 2022 for the first time (OECD, 2024). Creativity enables learners to thrive in a rapidly evolving workplace, with increasing AI integration, and where the ability to solve novel problems with effective solutions is emphasized by employers (WEF, 2023). We present a new Creativity Navigator framework of teaching for creativity (TfC), which integrates creative metacognition, cognitive processes, dispositional models, and creative climate research. Importantly, the framework was co‐developed by teachers and researchers, enabling its implementation through a three‐year Creativity Collaborative involving 16 schools, where intervention group teachers used it to plan TfC across subjects. Implementation included a multilayered approach addressing context, leadership, knowledge, agency, and pedagogies needed to embed TfC. Pre‐post comparisons between intervention and non‐intervention teachers revealed significantly higher scores for intervention teachers at the project endpoint in eight out of nine outcome measures, relating to TfC confidence and efficacy, creativity growth mindset, self‐perceived everyday creativity, and metacognitive knowledge. Findings indicate that our approach effectively supports teachers' practice of TfC in schools.

Optimizing the multi-model ensemble of CMIP6 GCMs for climate simulation over Bangladesh

This study aims to enhance the precision of climate simulations by optimizing a multi-model ensemble of General Circulation Models (GCMs) for simulating precipitation, maximum temperature (Tmax), and minimum temperature (Tmin). Bangladesh, with its susceptibility to rapid seasonal shifts and various forms of flooding, is the focal point of this research. Historical simulations of 19 CMIP6 GCMs are meticulously compared with ERA5 data for 1986–2014. The bilinear interpolation technique is used to harmonize the resolution of GCM data with the observed grid points. Seven distinct error metrics, including Kling-Gupta Efficiency and normalized root mean squared error, quantify the grid-to-grid agreement between GCMs and ERA5 data. The metrics are integrated into the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) for seasonal and annual rankings of GCMs. Finally, the ensemble means of top-performing models are estimated using Bayesian Model Averaging (BMA) and Arithmetic Mean (AM) for relative comparison. The outcomes of this study underscore the variability in GCM performance across different seasons, necessitating the development of an overarching ranking system. Results reveal ACCESS.CM2 is the preeminent GCM for precipitation, with an overall rating matric of 0.99, while INM.CM4.8 and UKESM1.0.LL excel in replicating Tmax and Tmin, with rating matrices of 1.0 and 0.88. In contrast, FGOALS.g3, KACE.1.0.G, and CanESM5 are the most underperformed models in estimating precipitation, Tmx, and Tmn, respectively. Overall, there are five models, ACCESS.ESM1.5, ACCESS.CM2, UKESM1.0.LL, MRI.ESM2.0, EC.Earth3 performed best in simulating both precipitation and temperature. The relative comparison of the ensemble means of the top five models revealed that the accuracy of BMA with Kling Gupta Efficiency (KGE) of 0.82, 0.65, and 0.82 surpasses AM with KGE of 0.59, 0.28, and 0.45 in capturing the spatial pattern of precipitation, Tmax and Tmin, respectively. This study offers invaluable insights into the selection of GCMs and ensemble methodologies for climate simulations in Bangladesh. Improving the accuracy of climate projections in this region can contribute significantly to climate science.

The Wind‐Driven Time‐Variable Circulation in the Arctic Mediterranean

AbstractThe Arctic Ocean is a key component of Earth's climate system, and an understanding of ocean dynamics in this region is central for predicting how the Arctic is responding to a changing climate. In this study, we examine the ocean circulation in a high‐resolution numerical model of the Arctic Ocean and Nordic Seas. Based on what is observed in this simulation, we reexamine an existing idealized linear model estimating the time‐variable large‐scale circulation in ocean basins, and test it against the highly nonlinear numerical model. The idealized model is an integral relation derived from the linear momentum equations and assumes that the circulation around a closed depth contour is driven by surface stresses and regulated by bottom friction. We show that the idealized model estimates agree very well with the numerical simulations. This indicates that much of the variability of the large‐scale circulation can be explained by linear processes. In particular, a correct description of the net surface stress over partially ice‐covered areas improves the correlation between linear model and numerical simulations significantly in the Arctic Ocean compared to a previous study. However, undetected in that previous study, we now find that the linear model might be lacking a cyclonic tendency.

Marine Stratocumulus Clouds With More Coarse Sea Spray Aerosols Are Brighter

AbstractThe idea of cooling the Earth by marine cloud brightening is well established. All prior studies considered enhancing cloud albedo only with fine aerosols (FA). Adding coarse sea spray aerosols (CSA, radius>1 μm) has been thought to have the opposite effect. Using nearly a decade of satellite observations and global aerosol reanalysis, we found that the maximum radiative cooling effect from marine stratocumulus occurs when FA is around 3 μg m−3 and CSA is around 30 μg m−3. Under low winds and high stability conditions, optimal FA and CSA can enhance cooling by −95 W m−2, nearly 60% more than adding FA alone. This CRE response to FA and CSA was consistently observed across various cloud‐controlling factors, thus minimizing the probability of being caused by meteorological co‐variability. These findings improve our understanding of how different aerosols affect Earth's climate, improve the evaluation of cooling achieved through marine cloud brightening, and support its feasibility.

Wavelet correlation noise analysis for qubit operation variable time series

In quantum computing, characterizing the full noise profile of qubits can aid in increasing coherence times and fidelities by developing error-mitigating techniques specific to the noise present. This characterization also supports efforts in advancing device fabrication to remove sources of noise. Qubit properties can be subject to non-trivial correlations in space and time, for example, spin qubits in MOS quantum dots are exposed to noise originating from the complex glassy behavior of two-level fluctuator ensembles. Engineering progress in spin qubit experiments generates large amounts of data, necessitating analysis techniques from fields experienced in managing large data sets. Fields such as astrophysics, finance, and climate science use wavelet-based methods to enhance their data analysis. Here, we propose and demonstrate wavelet-based analysis techniques to decompose signals into frequency and time components, enhancing our understanding of noise sources in qubit systems by identifying features at specific times. We apply the analysis to a state-of-the-art two-qubit experiment in a pair of SiMOS quantum dots with feedback applied to relevant operation variables. The observed correlations serve to identify common microscopic causes of noise, such as two-level fluctuators and hyperfine coupled nuclei, as well as to elucidate pathways for multi-qubit operation with more scalable feedback systems.

U‐Net Kalman Filter (UNetKF): An Example of Machine Learning‐Assisted Data Assimilation

AbstractMachine learning techniques have seen a tremendous rise in popularity in weather and climate sciences. Data assimilation (DA), which combines observations and numerical models, has great potential to incorporate machine learning and artificial intelligence (ML/AI) techniques. In this paper, we use U‐Net, a type of convolutional neutral network (CNN), to improve the localized error covariances for the Ensemble Kalman Filter (EnKF) algorithm. Using a 2‐layer quasi‐geostrophic model, U‐Nets are trained using data from EnKF DA experiments. The trained U‐Nets are then successfully implemented in U‐Net Kalman Filter (UNetKF) experiments to predict localized error covariances that possess adaptive localization and some state‐dependent features of the model error covariances. UNetKF is compared to traditional 3‐dimensional variational (3DVar), ensemble 3DVar (En3DVar) and EnKF methods. The performance of UNetKF can match or exceed that of 3DVar, or En3DVar and EnKF for small to moderate ensemble sizes. We also demonstrate that trained U‐Nets can be transferred to a higher‐resolution model for UNetKF implementation, which again performs competitively to 3DVar and EnKF, particularly for small ensemble sizes.