Global Model Research Articles

Investigating the “Too Bright” Issue Pertaining to Non‐PBL Clouds Over the South Pacific Trade‐Wind Region in CMIP6 Global Climate Models

AbstractThis paper examines the “too bright” issue pertaining to non‐planetary boundary layer (PBL) clouds over the South Pacific trade‐wind region and its potential link to the falling ice radiative effects (FIREs). We run sensitivity experiments with CESM2‐CAM6 (CESM2) global climate model with FIREs on (SON) and off (NOS). The model exhibits more in‐cloud liquid water content (CLWC) and droplet above the PBL in NOS, leading to larger shortwave (SW) reflectivity at the top of the atmosphere than in SON over the trade wind regions. CMIP6 models are divided into three subsets: separately calculates the radiative effects of cloud ice and falling ice (SON2), combined (SON1) and without falling ice (NOS). SON2 models exhibit improved CLWC and SW reflectivity similar to CESM2‐SON, while NOS and SON1 models are akin to CESM2‐NOS owing to weaker surface wind stress and warmer ocean surface, caused by the lack of FIREs over the convective zones.

Image forgery localization integrating multi-scale and boundary features

Abstract Image forgery localization identifies tampered regions within an image by extracting distinctive forgery features. Current methods mainly use convolutional neural networks (CNNs) to extract features. However, CNNs’ limited receptive field emphasizes local features, impeding the global modeling of crucial lower-level features like edges and textures, leading to decreased precision. Moreover, prior methods use pyramid networks for multi-scale feature extraction but show deficiencies in multi-scale and interlayer modeling, leading to inadequate multi-scale information representation and limiting flexibility to tampered regions of varying sizes. To address these issues, this paper proposes a Transformer-based model integrating multi-scale and boundary features. The model employs a Pyramid Vision Transformer as the encoder, using self-attention over convolution to enhance global context modeling. Building on this, the model incorporates a multi-scale feature enhancement module that enriches forgery features by paralleling various convolutional layers. Features at various encoder stages are integrated through a cross-stage interaction module, enabling multi-level feature correlation for a strong feature representation. Furthermore, the model includes a forgery boundary information-guided branch, which focuses precisely on tampered region structures without introducing irrelevant noise. Experiments demonstrate that our model surpasses previous methods in localization accuracy, with F1 and AUC improving by 8.5% and 2.2% in pre-training, respectively.

Modelling marine heatwaves impact on shallow and upper mesophotic tropical coral reefs

Abstract Coral reefs ecosystems, often compared to rain forests for their high biodiversity, are threatened by ocean warming causing coral bleaching when the symbiotic relationship between dinoflagellates and corals breaks under high ocean temperatures. Thermal stress from marine heatwaves (MHWs) occur both at the surface and subsurface with subsurface MHWs lasting longer with potentially higher cumulative intensities. However, global coral bleaching models generally ignore the differences in thermal stress between surface and sea-bed levels. Here, we define MHWs at sea-bed level to model coral bleaching with daily resolution from 6 May 1993 to 31 October 2023, for 9944 tropical coral reefs between 0 and 60 m depths. We show that deeper reefs experience on average higher thermal stress and bleaching compared to surface reefs. Using surface temperature data to model bleaching for deeper corals underestimates bleaching intensities by an average of 6% ± 9% compared to the subsurface calibrated model. Our study is a starting point for more accurate coral bleaching modelling, providing additional evidence to reshape our perception of deeper coral reefs as potential refugees from climate change.

Tropical Cyclone Wind Shear-Relative Asymmetry in Reanalyses

Abstract While tropical cyclones (TCs) are axisymmetric vortices to the first order, they often exhibit noteworthy structural asymmetries. These often result from environmental vertical wind shear, which tilts the vortex and induces a wavenumber 1 pattern in the circulation and precipitation fields. Reanalyses and climate models have improved in representing the TC structure and climatology, but their relatively coarse resolution and dependence on parameterized physics cast doubt on their ability to capture the asymmetric TC structure. We perform the most comprehensive process-oriented assessment of TC asymmetry to date in reanalyses. Specifically, we analyze the composite shear-relative TC structure in ERA5 and Climate Forecast System Reanalysis (CFSR), which vary in their resolutions, physical parameterization suites, and data assimilation techniques. These structures are compared with aircraft reconnaissance radar observations. In agreement with the observations, the strongest tangential winds are usually found left-of-shear, while inner core rainfall, ascent, vortex tilt, and low-level inflow are favored directly downshear or in the downshear-left quadrant. Outer rainband convection generally peaks in the downshear-right quadrant. Thermodynamic asymmetries are also apparent, with anomalous low-level moisture right-of-shear, midlevel warmth in the upshear-right quadrant (uptilt), and cloud properties suggestive of a realistic precipitation life cycle from growth to fallout. We also decompose rainfall contributions from the convective parameterization and large-scale cloud schemes and highlight the roles of vorticity advection, buoyancy advection, and diabatic processes in driving asymmetric vertical motions in the inner core and outer rainband regions. Our results suggest that process-level studies of TC asymmetry and TC–wind shear interaction under future warming are viable using climate models. Significance Statement Asymmetries are common in tropical cyclones (TCs), influencing their intensity, track, and hazards. Vertical wind shear often plays a leading-order role in causing these asymmetries. It is uncertain how well asymmetric structures and processes are captured in reanalyses and global climate models (GCMs) with grid spacings of 0.25° and coarser. In this study, we first evaluate TC asymmetry in reanalyses, which have the benefit of being forced by observations. This helps to assess whether the resolutions associated with GCMs sufficiently capture asymmetric structures and processes and motivates upcoming work with free-running GCMs to study how TC asymmetry may change in a warming climate.

The key role of forest disturbance in reconciling estimates of the northern carbon sink

AbstractNorthern forests are an important carbon sink, but our understanding of the driving factors is limited due to discrepancies between dynamic global vegetation models (DGVMs) and atmospheric inversions. We show that DGVMs simulate a 50% lower sink (1.1 ± 0.5 PgC yr−1 over 2001–2021) across North America, Europe, Russia, and China compared to atmospheric inversions (2.2 ± 0.6 PgC yr−1). We explain why DGVMs underestimate the carbon sink by considering how they represent disturbance processes, specifically the overestimation of fire emissions, and the lack of robust forest demography resulting in lower forest regrowth rates than observed. We reconcile net sink estimates by using alternative disturbance-related fluxes. We estimate carbon uptake through forest regrowth by combining satellite-derived forest age and biomass maps. We calculate a regrowth flux of 1.1 ± 0.1 PgC yr−1, and combine this with satellite-derived estimates of fire emissions (0.4 ± 0.1 PgC yr−1), land-use change emissions from bookkeeping models (0.9 ± 0.2 PgC yr−1), and the DGVM-estimated sink from CO2 fertilisation, nitrogen deposition, and climate change (2.2 ± 0.9 PgC yr−1). The resulting ‘bottom-up’ net flux of 2.1 ± 0.9 PgC yr−1 agrees with atmospheric inversions. The reconciliation holds at regional scales, increasing confidence in our results.

RFL-LSU: A Robust Federated Learning Approach with Localized Stepwise Updates

Distributed intelligence enables the widespread deployment of AI technology, greatly promoting the development of AI. Federated learning is a widely used distributed intelligence technology that allows iterative optimization of global model while protecting user data privacy. Currently, federated learning faces some security threats, as its open architecture provides attackers with opportunities to disrupt the learning process by submitting malicious updates or inserting backdoors. In this article, we propose a robust federated learning method to defend against potential malicious attacks. Specifically, we enhance the algorithm’s performance and stability by implementing localized stepwise updates on the client side and element-wise anomaly detection on the server side. We conducted experiments in a more realistic non-i.i.d. scenario and compared the results with other typical federated learning methods. Our results demonstrate that our approach exhibits strong robustness under non-i.i.d. data distributions, outperforming other methods in terms of test accuracy and resilience to attacks.

Optimizing Federated Learning Efficiency: A Comparative Analysis of Model Compression Techniques for Communication Reduction

Abstract. The rapidly growing number of connected devices continuously produces massive amounts of heterogeneous data, posing challenges to data privacy and security when leveraging them to train the high-quality big model. Federated learning, which enables numerous users to train a global model without cooperatively exchanging data, has emerged as a viable alternative. Though achieving significant progress, existing federated learning methods still struggle with large communication volumes, especially when the local devices only have limited computing and communication capabilities. To alleviate the efficiency issue, this paper compares the effect of three compression methods in promoting the training of federated learning models, including pruning, quantization, and knowledge distillation. The findings reveal that these methods reduce resource consumption while maintaining high model performance. Combining the pruning, quantization, and knowledge distillation technology through sequential application and parameter aggregation helps balance the model size and performance. Our cascading lightweight strategy, which preserves each method's unique edge while promoting deeper collaboration between them, has been shown beneficial through extensive testing.

Impacts of climate change and deforestation on fire risk in the Amazon basin

Fires in the Amazon rainforest are a result of anthropogenic activity in the region. As the Amazon rainforest becomes vulnerable to fire through deforestation and climate change, it is imperative to understand how the fire risk will change in future years—not only for the potential of the forest to become a source of carbon to the atmosphere if burned but also for the potential loss of unique biodiversity and its critical role in regional and global climate. Therefore, we used the Eta regional climate model driven by 3 global climate models to simulate 2 distinct scenarios: one with a high CO2 emission (RCP8.5-DEF15) path and another with a high CO2 emission path and the total deforestation of the Amazon rainforest (RCP8.5-DEFTOT). Combining the multi-model ensemble of temperature and precipitation, we calculated the Keetch-Byram Drought Index (KBDI) and analyzed the results for the dry and wet seasons. In both scenarios, precipitation is reduced and maximum temperature increased. The greatest changes occur during the dry season and with the RCP8.5-DEFTOT scenario. In RCP8.5-DEF15, the KBDI is greatly affected, showing positive anomalies in both seasons. With the addition of deforestation, the fire risk is further exacerbated, thus highlighting the role of the forest in suppressing optimal conditions for fire. Additionally, more than half of the basin area is predicted to experience a high risk of fire in extreme years. Our results emphasize the importance of reducing global emissions and deforestation of the Amazon, given that fire activity could threaten the future of the Amazon rainforest.

Projections of future bioclimatic indicators using bias-corrected CMIP6 models: a case study in a tropical monsoon region.

This study evaluates the potential impacts of climate change on Bangladesh by analyzing 19 bioclimatic indicators based on temperature and precipitation. Data from 18 bias-corrected CMIP6 global climate models (GCMs) were used, covering four Shared Socioeconomic Pathways (SSPs)-SSP126, SSP245, SSP370, and SSP585-across three future timeframes: near-term (2015-2044), mid-term (2045-2074), and long-term (2075-2100). Under the high-emission SSP585 scenario, average temperatures are projected to rise by up to 3.76°C, and annual precipitation could increase by 52.6%, reaching up to 3446.38mm by the end of the century. The maximum temperature (Bio5) could reach 32.91°C, while the minimum temperature (Bio6) might rise by 4.43°C, particularly during winter. Precipitation seasonality (Bio15) is projected to increase by as much as 7.9% in the northwest, indicating heightened variability between wet and dry seasons. The diurnal temperature range (Bio2) is expected to decrease by up to - 1.3°C, signifying reduced nighttime cooling, which could exacerbate heat stress. Significant reductions in temperature seasonality (Bio4) are forecast for the northeast, with notable declines in isothermality (Bio3) under SSP585, pointing to increased climatic extremes. These climatic shifts pose severe risks to agricultural productivity, water resource availability, and biodiversity, particularly in flood-prone regions. The findings highlight the need for urgent adaptation measures, including improved flood management systems, efficient water resource use, and climate-resilient agricultural practices. By providing robust region-specific projections, this study offers critical insights for policymakers and stakeholders to mitigate the adverse effects of climate change and safeguard environmental and economic sustainability in Bangladesh.

Effect of climate change on the habitat suitability of the relict species Zelkova carpinifolia Spach using ensembled species distribution modelling.

Zelkova carpinifolia is a Tertiary relict tree distributed in Hyrcanian and Colchic forests. Most of its habitat has been destroyed in the last century. This study aimed to model potentially suitable habitat areas for Zelkova carpinifolia from the past to the future. The Last Glacial Maximum (LGM) and Future (2061-2080) models include 19 bioclimatic variables from the CCSM4 global circulation model Pearson correlation coefficient was used to assess collinearity between variables and ten variables were selected for distribution modelling. Habitat suitability was estimated using the Biodiversity Modelling (BIOMOD) ensemble modelling method by combining the results of ten algorithm models using the R package "biomod2". The area under the curve (AUC) of the receiver operating characteristic (ROC) curve and true skills statistics (TSS) were calculated to evaluate the performance of the models. The contributions of the environmental variables were calculated separately for each algorithm model. According to the results obtained, the most effective bioclimatic variable in the distribution of the species is temperature seasonality (Bio4). The modelling results revealed that Zelkova carpinifolia survived in suitable refuge areas in western Asia during the LGM. These distribution areas have remained largely unchanged and even expanded. The future model results predict that the suitable habitats of the species will narrow in the Hyrcanian forests south of Caspian Sea and that more suitable conditions will be found around the Caucasus. Given the increasing destruction of these valuable plant species due to human activities and the expected negative impacts of climate change in the future, it is important to develop policies and strategies for the protection of Zelkova carpinifolia's habitat, the creation of nature reserves, and sustainability.

Tutorial on Molecular Latent Space Simulators (LSSs): Spatially and Temporally Continuous Data-Driven Surrogate Dynamical Models of Molecular Systems.

The inherently serial nature and requirement for short integration time steps in the numerical integration of molecular dynamics (MD) calculations place strong limitations on the accessible simulation time scales and statistical uncertainties in sampling slowly relaxing dynamical modes and rare events. Molecular latent space simulators (LSSs) are a data-driven approach to learning a surrogate dynamical model of the molecular system from modest MD training trajectories that can generate synthetic trajectories at a fraction of the computational cost. The training data may comprise single long trajectories or multiple short, discontinuous trajectories collected over, for example, distributed computing resources. Provided the training data provide sufficient sampling of the relevant thermodynamic states and dynamical transitions to robustly learn the underlying microscopic propagator, an LSS furnishes a global model of the dynamics capable of producing temporally and spatially continuous molecular trajectories. Trained LSS models have produced simulation trajectories at up to 6 orders of magnitude lower cost than standard MD to enable dense sampling of molecular phase space and large reduction of the statistical errors in structural, thermodynamic, and kinetic observables. The LSS employs three deep learning architectures to solve three independent learning problems over the training data: (i) an encoding of the high-dimensional MD into a low-dimensional slow latent space using state-free reversible VAMPnets (SRVs), (ii) a propagator of the microscopic dynamics within the low-dimensional latent space using mixture density networks (MDNs), and (iii) a generative decoding of the low-dimensional latent coordinates back to the original high-dimensional molecular configuration space using conditional Wasserstein generative adversarial networks (cWGANs) or denoising diffusion probability models (DDPMs). In this software tutorial, we introduce the mathematical and numerical background and theory of LSS and present example applications of a user-friendly Python package software implementation to alanine dipeptide and a 28-residue beta-beta-alpha (BBA) protein within simple Google Colab notebooks.

ENSO predictive analytics based on layered rendering

AbstractEl Niño-Southern Oscillation (ENSO) is a periodic climate phenomenon in the equatorial Pacific that significantly influences global climate patterns. Accurate prediction and monitoring of ENSO events are essential for meteorological agencies and governmental institutions. This study introduces a content-guided attention module within a convolutional neural network to improve prediction accuracy. This module models inter-channel relationships and enhances information interaction by integrating channel and spatial attention weights. These advancements substantially improve prediction accuracy and help overcome the spring prediction barrier in ENSO forecasting. The research emphasizes global feature modeling and proposes a novel content-guided ENSO prediction model. It also includes an ocean data generation model utilizing global attention. Furthermore, a layered rendering technique is employed to invert ocean data, facilitating detailed analysis and contributing to the development of an ocean synthetic dataset.

Balloon Baseline Stratospheric Aerosol Profiles (B2SAP)—Perturbations in the Southern Hemisphere, 2019–2022

AbstractVolcanic and pyrocumulonimbus (pyroCB) injections into the stratosphere perturb the aerosol layer and can have important radiative and chemical impacts on timescales spanning from months to several years. Repeated in situ balloon‐borne measurements of aerosol size and number concentration (>140 nm in diameter), ozone, water vapor, and atmospheric state variables made at midlatitudes in the southern hemisphere (SH) since 2019 enable us to better characterize such events. We use this record and coincident lidar extinction profiles to study several moderate to large stratospheric perturbations in the SH between 2019 and 2022 in detail, including the Australian New Year Super Outbreak (ANYSO) pyroCB in 2020. Median vertical profiles of aerosol number concentration, effective radius, and surface area in SH midlatitudes are also compared with those recorded in Northern Hemisphere midlatitudes under baseline conditions using an identical payload. These data depict the variability in stratospheric aerosol properties in the SH midlatitudes during this period and provide a benchmark for global sectional aerosol models. They reveal that sulfate particle size distributions under baseline conditions and in volcanic plumes are relatively well represented in the Community Earth System Model—Community Aerosol Radiation Model for Atmospheres (CESM‐CARMA), but more observations of biomass burning plumes are needed to improve model skill in simulating pyroCB. Comparisons between in situ and lidar observations also highlight a need for more observations of aerosol composition and refractive index in both fresh and aging biomass burning plumes.

Past foraminiferal acclimatization capacity is limited during future warming.

Climate change affects marine organisms, causing migrations, biomass reduction and extinctions1,2. However, the abilities of marine species to adapt to these changes remain poorly constrained on both geological and anthropogenic timescales. Here we combine the fossil record and a global trait-based plankton model to study optimal temperatures of marine calcifying zooplankton (foraminifera, Rhizaria) through time. The results show that spinose foraminifera with algal symbionts acclimatized to deglacial warming at the end of the Last Glacial Maximum (LGM, 19-21 thousand years ago, ka), whereas foraminifera without symbionts (non-spinose or spinose) kept the same thermal preference and migrated polewards. However, when forcing the trait-based plankton model with rapid transient warming over the coming century (1.5 °C, 2 °C, 3 °C and 4 °C relative to pre-industrial baseline), the model suggests that the acclimatization capacities of all ecogroups are limited and insufficient to track warming rates. Therefore, foraminifera are projected to migrate polewards and reduce their global carbon biomass by 5.7-15.1% (depending on the warming) by 2100 relative to 1900-1950. Our study highlights the different challenges posed by anthropogenic and geological warming for marine plankton and their ecosystem functions.

Optimization of global gravitational potential models with detailed gravity anomalies in Southern VietnamOptimization of global gravitational potential models with detailed gravity anomalies in Southern Vietnam

This study evaluates the accuracy of several global gravity field models, specifically EGM 2008, EICGEN-6C4, SGG-UGM-1, SGG-UGM-2, and GECO, using detailed gravity anomaly data from 1968 points in the southern plains of Vietnam. The evaluation process followed a rigorous methodology to ensure precision. Results indicate that these models exhibit similar accuracy levels in the experimental area, with a mean error of approximately 6.2 mGal. Among them, The SGG-UGM-2 model, demonstrated the highest precision, with a fluctuation range of -29.2425 to +74.9010 mGal. These newer generation models, which incorporate additional satellite data, offer enhanced reliability compared to the EGM2008 model. The fluctuation ranges for other models also varied: EGM 2008 from -29.8662 to +82.2140 mGal, EIGEN-6C4 from -28.3626 to +74.2292 mGal, SGG-UGM-1 from -28.9500 to +75.5315 mGal, and GECO from -30.5641 to +73.6775 mGal. This research highlights the critical importance of accurate gravity field models for various applications, including geodesy, geophysics, natural disaster management, resource exploration, and military operations. The methodologies employed for accuracy assessment are adaptable to other regions with similar datasets, underscoring the significance of precise geophysical measurements in Vietnam's diverse geographic landscape. The findings support the essential role of gravity anomaly data in understanding Earth's physical properties and emphasize the need for continued evaluation across different regions in Vietnam to establish comprehensive applicability. This research paves the way for advanced studies and practical applications, contributing to a deeper understanding of Earth's physical properties.

The comparative role of physical system processes in Hudson Strait ice stream cycling: a comprehensive model-based test of Heinrich event hypotheses

Abstract. Despite their recognized significance on global climate and extensive research efforts, the mechanism(s) driving Heinrich events remain(s) a subject of debate. Here, we use the 3D thermomechanically coupled glacial systems model (GSM) to examine the Hudson Strait ice stream surge cycling and the role of three factors previously hypothesized to play a critical role in Heinrich events: ice shelves, glacial isostatic adjustment, and sub-surface ocean temperature forcings. In contrast to all previous modeling studies examining HEs, the GSM uses a transient last glacial cycle climate forcing, global viscoelastic glacial isostatic adjustment model, and sub-glacial hydrology model. The results presented here are based on a high-variance sub-ensemble retrieved from North American history matching for the last glacial cycle. Over our comparatively wide sampling of the potential parameter space (52 ensemble parameters for climate forcing and process uncertainties), we find two modes of Hudson Strait ice streaming: classic binge–purge versus near-continuous ice streaming with occasional shutdowns and subsequent surge onset overshoot. Our model results indicate that large ice shelves covering the Labrador Sea during the last glacial cycle only occur when extreme calving restrictions are applied. The otherwise minor ice shelves provide insignificant buttressing for the Hudson Strait ice stream. While sub-surface ocean temperature forcing leads to minor differences regarding surge characteristics, glacial isostatic adjustment does have a significant impact. Given input uncertainties, the strongest controls on ice stream surge cycling are the poorly constrained deep geothermal heat flux under Hudson Bay and Hudson Strait and the basal drag law. Decreasing the geothermal heat flux within available constraints and/or using a Coulomb sliding law instead of a Weertman-type power law leads to a shift from the near-continuous streaming mode to the binge–purge mode.

Enhancing Fine Aerosol Simulations in the Remote Atmosphere with Machine Learning

Global aerosol models often underestimate the mass concentration of aerosols in the remote troposphere, as evidenced by aircraft measurements. This study leveraged data from the NASA Atmospheric Tomography Mission (ATom), which provides remote aerosol concentrations, to refine algorithms for simulating these concentrations. Using the GEOS-Chem model, we simulate five fine aerosol types and enhance the simulation results using five machine-learning algorithms: Random Forest, XGBoost, SVM, KNN, and LightGBM, and compare the performance of these algorithms. Additionally, we evaluate the refinement effect of algorithms based on decision trees on a validation dataset. The results demonstrate that GEOS-Chem generally underestimated aerosol mass concentration. Among the tested algorithms, algorithms based on decision trees, particularly the Random Forest algorithm and the LightGBM algorithm, exhibited a superior performance, significantly improving prediction accuracy and computational efficiency in both the training and testing phases, as well as on the validation dataset.

Impact of biomass burning aerosols (BBA) on the tropical African climate in an ocean–atmosphere–aerosol coupled climate model

Abstract. The impact of biomass burning aerosols (BBA) emitted in central Africa on the tropical African climate is studied using the ocean–atmosphere global climate model CNRM-CM, including prognostic aerosols. The direct BBA forcing, cloud feedbacks (semi-direct effects), effects on surface solar radiation, atmospheric dynamics and precipitation are analysed for the 1990–2014 period. During the June–July–August (JJA) season, the CNRM-CM simulations reveal a BBA semi-direct effect exerted on low-level clouds with an increase in the cloud fraction of ∼5 %–10 % over a large part of the tropical ocean. The positive effect of BBA radiative effects on low-level clouds is found to be mainly due to the sea surface temperature response (decrease of ∼0.5 K) associated with solar heating at 700 hPa, which increases the lower-tropospheric stability. Over land, results also indicate a positive effect of BBA on the low-cloud fraction, especially for the coastal regions of Gabon and Angola, with a potentially enhanced impact in these coupled simulations that integrates the response (cooling) of the sea surface temperature (SST). In addition to the BBA radiative effect on SST, the ocean–atmosphere coupled simulations highlight that the oceanic temperature response is noticeable (about −0.2 to −0.4 K) down to ∼80 m depth in JJA between the African coast and 10° W. In parallel to low-level clouds, reductions of ∼5 %–10 % are obtained for mid-level clouds over central Africa, mainly due to BBA-induced surface cooling and lower-tropospheric heating inhibiting convection. In terms of cloud optical properties, the BBA radiative effects induced an increase in the optical depth of about ∼2–3 over the ocean south of the Equator. The result of the BBA direct effect and feedback on tropical clouds modulates the surface solar radiation over the whole of tropical Africa. The strongest surface dimming is over central Africa (∼-30 W m−2), leading to a large reduction in the continental surface temperature (by ∼1 to 2 K), but the solar radiation at the oceanic surface is also affected up to the Brazilian coast. With respect to the hydrological cycle, the CNRM-CM simulations show a negative effect on precipitation over the western African coast, with a decrease of ∼1 to 2 mm d−1. This study also highlights a persistent impact of BBA radiative effects on low-level clouds (increase in cloud fraction, liquid water content and optical depth) during the September–October–November (SON) period, mainly explained by a residual cooling of sea surface temperature over most of the tropical ocean. In SON, the effect on precipitation is mainly simulated over the Gulf of Guinea, with a reduction of ∼1 mm d−1. As for JJA, the analysis clearly highlights the important role of the slow response of the ocean in SON and confirms the need to use coupled modelling platforms to study the impact of BBA on the tropical African climate.

Exploring the impacts of climate change and identifying potential adaptation strategies for sustainable rice production in Thailand's Lower Chao Phraya Basin through crop simulation modeling.

The lower Chao Phraya River Basin (CPRB) in Thailand, a major rice-producing area, is grappling with increased water scarcity alongside more frequent floods and droughts, necessitating effective adaptation strategies to sustain agricultural productivity. This study assesses the impacts of climate change on rice yield and irrigation water use, using the DSSAT-CERES-Rice model. Based on these findings, potential genotype- and management-based adaptation strategies were recommended. The model was calibrated and evaluated using the data from field experiments conducted at the Asian Institute of Technology, Thailand during 2017-2018 and 2021-2022. The grain yield and irrigation water use between baseline (2010-2022) and future climate periods (early-century: 2023-2040, mid-century: 2041-2070, and late-century: 2071-2100) were compared. Future climate projections were based on five Global Climate Models (GCMs) from the NEX-GDDP-CMIP6 project under three scenarios (SSP126, SSP245, and SSP585). The model calibration and evaluation demonstrated very good performance statistics, with a d-index of 0.85 during both calibration and evaluation. The model simulations indicated that the maximum and minimum temperatures in the lower CPRB are projected to increase by ~ 2 °C and ~ 4 °C in the late century under SSP245 and SSP585, respectively. Consequently, rice yields are projected to decline by up to 33%, and irrigation water usage to increase by 53% under SSP585 by the late century. Based on the findings, the following major genotype- and management-based adaptation strategies are recommended: (1) Developing heat-tolerant rice cultivars to mitigate yield losses under future climate scenarios, (2) Developing rice cultivars with extended grain-filling duration to enhance both irrigation water use and yield, (3) Shifting the planting date 1-2 weeks earlier (from baseline planting date of 20 July), and shifting fertilizer application date 1-2 weeks earlier (from baseline fertilizer application date of 20 September) for the panicle initiation stage to improve yield, and (4) Optimizing irrigation thresholds (remaining soil water at which to irrigate) to reduce irrigation water use without compromising yield. Overall, the findings highlight the importance of genotype improvement and adaptive management practices in mitigating the adverse effects of climate change on rice production in the lower CPRB.

Spatio-temporal attention based collaborative local–global learning for traffic flow prediction

Traffic flow prediction is crucial for intelligent transportation systems (ITS), providing valuable insights for traffic control, route planning, and operation management. Existing work often separately models the spatial and temporal dependencies and primarily relies on predefined graphs to represent spatio-temporal dependencies, neglecting the traffic dynamics caused by unexpected events and the global relationships among road segments. Unlike previous models that primarily focus on local feature extraction, we propose a novel collaborative local–global learning model (LOGO) that employs spatio-temporal attention (STA) and graph convolutional networks (GCN). Specifically, LOGO simultaneously extracts hidden traffic features from both local and global perspectives. In local feature extraction, a novel STA is devised to directly attend to spatio-temporal coupling interdependencies instead of separately modeling temporal and spatial dependencies, and to capture in-depth spatio-temporal traffic context with an adaptive graph focusing on the dynamics in traffic flow. In global feature extraction, a global correlation matrix is constructed and GCNs are utilized to propagate messages on the obtained matrix to achieve interactions between both adjacent and similar road segments. Finally, the obtained local and global features are concatenated and fed into a gated aggregation to forecast future traffic flow. Extensive experiments on four real-world traffic datasets sourced from the Caltrans Performance Measurement System (PEMS03, PEMS04, PEMS07, and PEMS08) demonstrate the effectiveness of our proposed model. LOGO achieves the best performance over 18 state-of-the-art baselines and the best prediction performance with the highest improvement of 6.06% on the PEMS07 dataset. Additionally, two real-world case studies further substantiate the robustness and interpretability of LOGO.