Temporal Integration Research Articles

Advancing terrestrial ecology by improving cross-temporal research and collaboration

Abstract Ecology spans spatial and temporal scales and is inclusive of the history of life on Earth. However, research that occurs at millennial timescales or longer has historically been defined as paleoecology and has not always been well integrated with modern (neo-) ecology. This bifurcation has been previously highlighted, with calls for improved engagement among the subdisciplines, but their priority research areas have not been directly compared. To characterize the research agendas for terrestrial ecological research across different temporal scales, we compared two previous studies, Sutherland and colleagues (2013; neoecology) and Seddon and colleagues (2014; paleoecology), that outlined priority research questions. We identified several themes with potential for temporal integration and explored case studies that highlight cross-temporal collaboration. Finally, a path forward is outlined, focusing on education and training, research infrastructure, and collaboration. Our aim is to improve our understanding of biodiversity patterns and processes by promoting an inclusive and integrative approach that treats time as a foundational concept in ecology.

The effect of distance on audiovisual temporal integration in an outdoor virtual environment

In this study, we explore the influence of stimulus distance on human tolerance for (physical) asynchronies in virtual reality (VR). A repeated audiovisual (AV) stimulus with sound and light bursts was presented to the participants in an outdoor virtual environment (VE) using a head-mounted display (HMD). The investigation focused on quantifying the point of subjective simultaneity (PSS) for both visual and auditory stimuli. A synchrony judgment method (SJ-3) was used for 11 stimulus onset asynchronies (SOA) and five egocentric distances from 10 m up to 50 m with 10 m increments. The data analysis showed negative PSS values that decreased with distance, resulting in a negative slope (-3 ms/m) of the regression line between PSS values and simulated distances. In contrast to the recent study conducted in the indoor VE, we conclude that the presented study in the outdoor VE does not incorporate a distance compensation mechanism and refutes the hypothesis of an ‘implicit estimation’ of sound-arrival time. The reasons behind the observed contrast are discussed in this paper. Moreover, the negative slope of the regression line (−3 ms/m) is similar to previous research, concluding that there is a temporal integration of auditory-visual information within human neural processes without distance compensation.

Discretely Nonlinearly Stable Weight-Adjusted Flux Reconstruction High-Order Method for Compressible Flows on Curvilinear Grids

To achieve genuine predictive capability, an algorithm must consistently deliver accurate results over prolonged temporal integration periods, avoiding the unwarranted growth of aliasing errors that compromise the discrete solution. Provable nonlinear stability bounds the discrete approximation and ensures that the discretization does not diverge. Nonlinear stability is accomplished by satisfying a secondary conservation law, namely for compressible flows; the second law of thermodynamics. For high-order methods, discrete nonlinear stability and entropy stability, have been successfully implemented for discontinuous Galerkin (DG) and residual distribution schemes, where the stability proofs depend on properties of L2-norms. In this paper, nonlinearly stable flux reconstruction (NSFR) schemes are developed for three-dimensional compressible flow in curvilinear coordinates. NSFR is derived by merging the energy stable flux reconstruction (ESFR) framework with entropy stable DG schemes. NSFR is demonstrated to use larger time-steps than DG due to the ESFR correction functions, at the cost of larger error levels at design order convergence for equivalent degrees of freedom, while preserving discrete nonlinear stability. NSFR differs from ESFR schemes in the literature since it incorporates the FR correction functions on the volume terms through the use of a modified mass matrix. We also prove that discrete kinetic energy stability cannot be preserved to machine precision for quadrature rules where the surface quadrature is not a subset of the volume quadrature. This result stems from the inverse mapping from the kinetic energy variables to the conservative variables not existing for the kinetic energy projected variables. This paper also presents the NSFR modified mass matrix in a weight-adjusted form. This form reduces the computational cost in curvilinear coordinates because the dense matrix inversion is approximated by a pre-computed projection operator and the inverse of a diagonal matrix on-the-fly and exploits the tensor product basis functions to utilize sum-factorization. The nonlinear stability properties of the scheme are verified on a nonsymmetric curvilinear grid for the inviscid Taylor-Green vortex problem and the correct orders of convergence were obtained on a curvilinear mesh for a manufactured solution. Lastly, we perform a computational cost comparison between conservative DG, overintegrated DG, and our proposed entropy conserving NSFR scheme, and find that our proposed entropy conserving NSFR scheme is computationally competitive with the conservative DG scheme.

Advanced piezoelectric fluid energy harvesters by monolithic fluid–structure–piezoelectric coupling: A full-scale finite element model

A full-scale finite element model is presented for monolithic fluid–structure interaction (FSI) simulations of thin-walled piezoelectric fluid energy harvesters (PFEHs). Unlike widely used beam/plate-based models, our model employs a solid finite element discretization to precisely represent the complex PFEH designs involving microstructured transducers and non-uniform cantilevers. These features, plus the local FSI effects, are often ignored by simplified models. We applied the Galerkin method to formulate the weak form of the mixed equation system, integrating the flow dynamics, the geometrically nonlinear cantilever, the piezoelectric components, the electrode, and the output circuit within a closed-circuit electro-mechanical coupled system. The coupling of the multiple domains is achieved through boundary-fitted discretization within a monolithic scheme, using shifted-Crank–Nicolson temporal integration. This work explored implementing piezoelectric FSI systems within the FEniCS-based TurtleFSI library, and experimented techniques such as employing penalty functions for achieving electrode components with uniform electric potentials. We investigated various advanced PFEH features, including the baseplate design, the arrangement and microstructure of the piezoelectric components, and their influence on the system's dynamic and energy output behavior. The results confirmed the model's key advantages: full-scale modeling allows the integration of complex base structures and transducer microstructures in PFEH design. Combined with monolithic FSI coupling, it offers greater versatility, supporting a wider range of fluid environments and configurations in both wind and hydropower harvesting. Additionally, the modeling strategy can be intended not only to enhance power output, but also to minimize material usage, reduce mechanical fatigue, and extend the operational lifespan of PFEH systems.

A new function of offset response in the primate auditory cortex: marker of temporal integration

Offset responses are traditionally viewed as indicators of sound cessation. Here, we investigate offset responses to auditory click trains, examining how they are modulated by inter-click intervals (ICIs) and train duration. Using extracellular recordings and electrocorticography (ECoG) in non-human primates, alongside electroencephalography (EEG) in humans, we show that offset responses are significantly influenced by both ICI and train length, thereby establishing them as markers of temporal integration. We introduce the concept of the ‘Neuronal Integrative Window’ (NIW), defined as the temporal span during which neurons integrate stimuli to produce or modulate the temporal integration signal. Our data reveal that on the neuronal level, the auditory cortex (AC) exhibits a more expansive NIW than the medial geniculate body (MGB), integrating stimuli over longer durations and showing a preference for larger ICIs. Furthermore, our results indicate that offset responses could serve as potential biomarkers for neurological and psychiatric conditions, highlighted by their sensitivity to pharmacological modulation with ketamine. This study advances our understanding of auditory temporal processing and proposes a novel approach for assessing and monitoring brain health.

Exploring the Phenotype and Possible Mechanisms of Palinopsia in Visual Snow Syndrome.

Palinopsia (persistent afterimages and/or trailing) is a common but poorly understood symptom of the neurological condition visual snow syndrome. This study aimed to collect a phenotypical description of palinopsia in visual snow syndrome and probe for abnormalities in temporal visual processing, hypothesizing that palinopsia could arise from increased visibility of normal afterimage signals or prolonged visible persistence. Thirty controls and 31 participants with visual snow syndrome (18 with migraine) took part. Participants completed a palinopsia symptom questionnaire. Contrast detection thresholds were measured before and after brief exposure to a spatial grating because deficient contrast adaptation could increase afterimage visibility. Temporal integration and segregation were assessed using missing-element and odd-element tasks, respectively, because prolonged persistence would promote integration at wide temporal offsets. To distinguish the effects of visual snow syndrome from comorbid migraine, 25 people with migraine alone participated in an additional experiment. Palinopsia was common in visual snow syndrome, typically presenting as unformed images that were frequently noticed. Contrary to our hypotheses, we found neither reduced contrast adaptation (F(3.22, 190.21) = 0.71, P = 0.56) nor significantly prolonged temporal integration thresholds (F(1, 59) = 2.35, P = 0.13) in visual snow syndrome. Instead, participants with visual snow syndrome could segregate stimuli in closer succession than controls (F(1, 59) = 4.62, P = 0.04, ηp2 = 0.073) regardless of co-occurring migraine (F(2, 53) = 1.22, P = 0.30). In contrast, individuals with migraine alone exhibited impaired integration (F(2, 53) = 4.44, P = 0.017, ηp2 = 0.14). Although neither deficient contrast adaptation nor prolonged visible persistence explains palinopsia, temporal resolution of spatial cues is enhanced and potentially more flexible in visual snow syndrome.

Mapping stem volume in fast-growing eucalypt plantations: integrating spectral, textural, and temporal remote sensing information with forest inventories and spatial models

Key messageAccurate volume estimation in Eucalyptus plantation stands was achieved by a linear model using SPOT and Landsat multispectral imagery, specifically texture indices and pixel-scale NDVI time integrals, which reflect the local plantation growth history. Spatial modelling techniques such as Kriging with External Drift and Generalized Additive Model slightly improved predictions by accounting for spatial correlation of volume between sample points.ContextForest inventories are widely used to estimate stand production. To capture the inherent spatial variability within stands, spatial modelling techniques such as Kriging with External Drift (KED) and the generalized additive model (GAM) have emerged. These models incorporate information on spatial correlation and auxiliary variables that can be obtained from satellite imagery.AimsOur study explored the use of reflectance data from SPOT and Landsat multispectral imagery. We focused on texture indices and temporal integration of vegetation indices as auxiliary variables in KED and GAM to predict stem volume of fast-growing Eucalyptus sp. plantations in Brazil.MethodsThe components extracted from the high-resolution SPOT-6 image included spectral band values, band ratio metrics, key vegetation indices (NDVI, SAVI, and ARVI), texture measurements, and indices derived from texture analysis. Additionally, we included the accumulated NDVI time series acquired from Landsat 5, 7, and 8 satellites between the planting date and the forest inventory measurement date.ResultsThe best linear model of stem volume using remotely sensed predictors gave an R-squared value of 0.95 and a Root Mean Square Error (RMSE) of 12.44 m3/ha. The R-squared increased to 0.96 and the RMSE decreased to 10.6 m3/ha when the same predictors were included as auxiliary variables in the KED and GAM spatial models.ConclusionThe linear model using remotely sensed predictors contributed most to volume prediction, but the addition of spatial coordinates in the KED and GAM spatial models improved local volume predictions.

Better early than late: the temporal dynamics of pointing cues during cross-situational word learning

Abstract Learning the meaning of a word is a difficult task due to the variety of possible referents present in the environment. Visual cues such as gestures frequently accompany speech and have the potential to reduce referential uncertainty and promote learning, but the dynamics of pointing cues and speech integration are not yet known. If word learning is influenced by when, as well as whether, a learner is directed correctly to a target, then this would suggest temporal integration of visual and speech information can affect the strength of association of word–referent mappings. Across two pre-registered studies, we tested the conditions under which pointing cues promote learning. In a cross-situational word learning paradigm, we showed that the benefit of a pointing cue was greatest when the cue preceded the speech label, rather than following the label (Study 1). In an eye-tracking study (Study 2), the early cue advantage was due to participants’ attention being directed to the referent during label utterance, and this advantage was apparent even at initial exposures of word–referent pairs. Pointing cues promote time-coupled integration of visual and auditory information that aids encoding of word–referent pairs, demonstrating the cognitive benefits of pointing cues occurring prior to speech.

Synaptic weight dynamics underlying memory consolidation: Implications for learning rules, circuit organization, and circuit function

Systems consolidation is a common feature of learning and memory systems, in which a long-term memory initially stored in one brain region becomes persistently stored in another region. We studied the dynamics of systems consolidation in simple circuit architectures with two sites of plasticity, one in an early-learning and one in a late-learning brain area. We show that the synaptic dynamics of the circuit during consolidation of an analog memory can be understood as a temporal integration process, by which transient changes in activity driven by plasticity in the early-learning area are accumulated into persistent synaptic changes at the late-learning site. This simple principle naturally leads to a speed-accuracy tradeoff in systems consolidation and provides insight into how the circuit mitigates the stability-plasticity dilemma of storing new memories while preserving core features of older ones. Furthermore, it imposes two constraints on the circuit. First, the plasticity rule at the late-learning site must stably support a continuum of possible outputs for a given input. We show that this is readily achieved by heterosynaptic but not standard Hebbian rules. Second, to turn off the consolidation process and prevent erroneous changes at the late-learning site, neural activity in the early-learning area must be reset to its baseline activity. We provide two biologically plausible implementations for this reset that propose functional roles in stabilizing consolidation for core elements of the cerebellar circuit.

Photonic Neuromorphic Processing with On‐Chip Electrically‐Driven Microring Spiking Neuron

AbstractGuided by brain‐like temporal processing and event‐driven manner, neuromorphic computing has emerged as a competitive paradigm to realize artificial intelligence with high energy efficiency. Silicon photonics offers an ideal hardware platform with mutual foundry fabrication process and well‐developed device libraries, however, its huge potential to build integrated neuromorphic systems is significantly hindered due to the lack of scalable on‐chip photonic spiking neurons. Here, the first integrated electrically‐driven spiking neuron based on a silicon microring under the carrier injection working mode is reported, which is capable of emulating fundamental neural dynamics including excitability threshold, temporal integration, refractory period, controllable spike inhibition, and precise time encoding at a speed of 250 MHz. By programming time‐multiplexed spike representations, photonic spiking convolution is experimentally realized for image edge feature detection. Besides, a spiking convolutional neural network is constructed by combining photonic convolutional layers with a software‐implemented fully‐connected layer, which yields a classification accuracy of 94.1% on the benchmark Modified National Institute of Standards and Technology database. Moreover, it is theoretically verified that it's promising to further improve the operation speed to a gigahertz level by developing an electro‐optical co‐simulation model. The proposed microring neuron constitutes the final building block of scalable spike activation, thus representing a great breakthrough to boost the development of on‐chip neuromorphic information processing.

No evidence for unconscious initiation and following of arithmetic rules: A replication study.

The field of consciousness studies has yielded various-sometimes contradicting-accounts regarding the function of consciousness, ranging from denying it has such function to claiming that any high-level cognitive function requires consciousness. Empirical findings supporting both accounts were reported, yet some of them have been recently revisited based on failures to replicate. Here, we aimed at replicating a remarkable finding reported by Ric and Muller (2012); participants were able to follow an unseen instruction, integrate it with a subsequently presented pair of unseen digits, and accordingly either add the digits (resulting in a priming effect), or simply represent them. This finding thus demonstrates unconscious task-switching, temporal integration (involving mental chaining), and arithmetic operation. Finding such high-level processes in the absence of awareness is of pivotal importance to our understanding of consciousness, as it challenges prominent theories in the field (e.g., the global neuronal workspace). Accordingly, in light of the self-correction wave in psychological science in general and in the field of consciousness studies in particular, we report here a preregistered replication aimed at testing the reproducibility of this finding, while also better controlling for subjects' awareness of both the instruction and the digits. Across two highly powered experiments, our results failed to replicate the original effect. We, therefore, conclude that the current evidence does not support the claim that arithmetic operations (specifically, addition) can be flexibly initiated without awareness, in line with the current arguments for a more limited scope of unconscious processing. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Recommendations on bioacoustical metrics relevant for regulating exposure to anthropogenic underwater sounda).

Metrics to be used in noise impact assessment must integrate the physical acoustic characteristics of the sound field with relevant biology of animals. Several metrics have been established to determine and regulate underwater noise exposure to aquatic fauna. However, recent advances in understanding cause-effect relationships indicate that additional metrics are needed to fully describe and quantify the impact of sound fields on aquatic fauna. Existing regulations have primarily focused on marine mammals and are based on the dichotomy of sound types as being either impulsive or non-impulsive. This classification of sound types, however, is overly simplistic and insufficient for adequate impact assessments of sound on animals. It is recommended that the definition of impulsiveness be refined by incorporating kurtosis as an additional parameter and applying an appropriate conversion factor. Auditory frequency weighting functions, which scale the importance of particular sound frequencies to account for an animal's sensitivity to those frequencies, should be applied. Minimum phase filters are recommended for calculating weighted sound pressure. Temporal observation windows should be reported as signal duration influences its detectability by animals. Acknowledging that auditory integration time differs across species and is frequency dependent, standardized temporal integration windows are proposed for various signal types.

Modelling of wall-bounded cavitating flow and spray mixing in multi-component environments using the PC-SAFT equation of state

This work introduces a numerical multiphase model for multi-component mixtures, utilizing tabulated data for physical and transport properties across a spectrum of conditions from near-vacuum pressures to supercritical states. The property data are derived using Perturbed Chain Statistical Associating Fluid Theory (PC-SAFT), vapor-liquid equilibrium (VLE) calculations, entropy scaling methodologies, and Group Contribution (GC) methods. These techniques accurately reflect the thermodynamic behaviors of real fluids, avoiding the empirical estimation of Equation of State (EoS) input parameters. Implemented in OpenFOAM, the fluid dynamics solver is designed to address the three-dimensional Navier-Stokes equations for multi-component mixtures. The methodology integrates operator splitting to manage hyperbolic and parabolic steps distinctively. Hyperbolic terms are solved using the HLLC (Harten-Lax-van Leer-Contact) solver with temporal integration performed via a third-order Strong-Stability-Preserving Runge–Kutta (SSP-RK3) method. Viscous stress tensor contributions in the momentum equation are handled through an implicit velocity correction equation, while parabolic terms in the energy equation are explicitly solved. The simulation efficiency is further enhanced by adaptive Local Time Stepping and the Immersed Boundary (IB) method, which addresses interactions between the fluid and solid boundaries. Turbulence is resolved using the Wall Adaptive Large Eddy (WALE) model. Applied to high-pressure diesel fuel spray injections into non-reacting (nitrogen) gas environments, the model has been validated against Engine Combustion Network (ECN) data for the Spray-C configuration, featuring a fully cavitating multi-hole orifice. Results demonstrate that the model achieves accurate predictions across a broad range of tested conditions without the need for tuning or calibration parameters.

A Study on Designing Cyber Training and Cyber Range to Effectively Respond to Cyber Threats

As cyberattacks become increasingly sophisticated with advancements in information and communication technology, the impact of cyberspace threats is growing in both civilian and defense sectors. The utilization of cyber capabilities in operations is on the rise, prompting major nations to continuously enhance their cyber capabilities. This study aims to establish a systematic approach to cyber operations training and propose a framework for the development of cyber training. A hybrid cyber training system is designed as a plan for temporal and spatial integration to simultaneously combine simulation-based training with real-world target training. To develop this concept, a literature review was conducted, expert consultations were held, and data were collected and analyzed through visits to relevant organizations and units. Additionally, the fundamental components of cyber training were examined from environmental, scenario-based, and operational perspectives, leading to the presentation of a development direction for effective cyber training. This study is anticipated to enhance response capabilities to evolving cyber threats and attacks, improve cyber operational proficiency, and secure cyber power to achieve dominance in cyberspace.

Spatially and temporally high-order dynamic nonlinear variational multiscale methods for generalized Newtonian flows

Non-Newtonian fluids are of interest in industrial sectors, biological problems and other natural phenomena. This work proposes rheologically-dependent, spatially and temporally high-order non-residual stabilized finite element formulations. The accuracy of the methods is assessed by tackling highly-convective time-dependent power-law flows. The spatial approximation uses Lagrangian finite elements up to fourth order. The temporal integration is done via backward differentiation formulas of order one, two and three. A key aspect of our work is using a non-residual orthogonal variational multiscale (VMS) formulation to stabilize dominant convection and to allow equal-order interpolation of velocity and pressure. Our VMS method uses dynamic nonlinear subscales, which have not been tested so far for generalized Newtonian fluids. In this work, the use of high-order temporal discretizations for the subscale components is systematically evaluated. Numerical experiments consider the flow over a confined cylinder for Reynolds numbers between 40 and 400 and power-law indices between 0.4 and 1.8. Numerical testing demonstrates the method to be stable in all combinations of spatial and temporal orders. Our results show that using high-order spatial discretizations more accurately approximates boundary layers and viscosity fields. Moreover, higher temporal orders allow using larger time steps while still capturing highly dynamic behaviors with better resolution in frequency spectra.

Temporal integration in human auditory cortex is predominantly yoked to absolute time, not structure duration.

Sound structures such as phonemes and words have highly variable durations. Thus, there is a fundamental difference between integrating across absolute time (e.g., 100 ms) vs. sound structure (e.g., phonemes). Auditory and cognitive models have traditionally cast neural integration in terms of time and structure, respectively, but the extent to which cortical computations reflect time or structure remains unknown. To answer this question, we rescaled the duration of all speech structures using time stretching/compression and measured integration windows in the human auditory cortex using a new experimental/computational method applied to spatiotemporally precise intracranial recordings. We observed significantly longer integration windows for stretched speech, but this lengthening was very small (~5%) relative to the change in structure durations, even in non-primary regions strongly implicated in speech-specific processing. These findings demonstrate that time-yoked computations dominate throughout the human auditory cortex, placing important constraints on neurocomputational models of structure processing.

The effect of object perception on event integration and segregation.

The perceptual system must integrate information from various points in time and space to interpret continuous sensory input into meaningful units, such as visual objects or events. To explore the relationship between the perception of spatial objects and temporal events, we modified the missing element task, a typical temporal integration task, by inserting a simple spatial object. The aim was to determine whether the perceptual processing of the object would have an impact on the frequency of temporal integration and segregation. Temporal integration was most successful when the missing element was located within the object, less successful when there was no object, and least successful when the missing element appeared outside the object. The advantage of the location of the missing element within the object was observed at display durations from 30 ms to 150 ms. Interestingly, the object provided the same benefit for integration and segregation despite their opposing perceptual demands. This study demonstrates the relationship that exists between the processing of temporal events and spatial objects, and shows how such spatial information can facilitate temporal integration.

Maternal immune activation alters temporal Precision of spike generation of CA1 pyramidal neurons by Unbalancing GABAergic inhibition in the Offspring

Infection during pregnancy represents a risk factor for neuropsychiatric disorders associated with neurodevelopmental alterations. A growing body of evidence from rodents and non-human primates shows that maternal inflammation induced by viral or bacterial infections results in several neurobiological alterations in the offspring. These changes may play an important role in the pathophysiology of psychiatric disorders like schizophrenia and autism spectrum disorders, whose clinical features include impairments in cognitive processing and social performance. Such alterations are causally associated with the maternal inflammatory response to infection rather than with the infection itself. Previously, we reported that CA1 pyramidal neurons of mice exposed to MIA exhibit increased excitability accompanied by a reduction in dendritic complexity. However, potential alterations in cellular and synaptic rules that shape the neuronal computational properties of the offspring remain to be determined. In this study, using mice as subjects, we identified a series of cellular and synaptic alterations endured by CA1 pyramidal neurons of the dorsal hippocampus in a lipopolysaccharide-induced maternal immune activation (MIA) model. Our data indicate that MIA reshapes the excitation-inhibition balance by decreasing the perisomatic GABAergic inhibition predominantly mediated by cholecystokinin-expressing Interneurons but not parvalbumin-expressing interneurons impinging on CA1 pyramidal neurons. These alterations yield a dysregulated amplification of the temporal and spatial synaptic integration. In addition, MIA-exposed offspring displayed social and anxiety-like abnormalities. These findings collectively contribute to understanding the cellular and synaptic alterations underlying the behavioral symptoms present in neurodevelopmental disorders associated with MIA.

A novel task to measure temporal integration in the human visual system

A novel task to measure temporal integration in the human visual system

Domain knowledge-enhanced multi-spatial multi-temporal [formula omitted] forecasting with integrated monitoring and reanalysis data

Accurate air quality forecasting is crucial for public health, environmental monitoring and protection, and urban planning. However, existing methods fail to effectively utilize multi-scale information, both spatially and temporally. There is a lack of integration between individual monitoring stations and city-wide scales. Temporally, the periodic nature of air quality variations is often overlooked or inadequately considered. To overcome these limitations, we conduct a thorough analysis of the data and tasks, integrating spatio-temporal multi-scale domain knowledge. We present a novel Multi-spatial Multi-temporal air quality forecasting method based on Graph Convolutional Networks and Gated Recurrent Units (M2G2), bridging the gap in air quality forecasting across spatial and temporal scales. The proposed framework consists of two modules: Multi-scale Spatial GCN (MS-GCN) for spatial information fusion and Multi-scale Temporal GRU (MT-GRU) for temporal information integration. In the spatial dimension, the MS-GCN module employs a bidirectional learnable structure and a residual structure, enabling comprehensive information exchange between individual monitoring stations and the city-scale graph. Regarding the temporal dimension, the MT-GRU module adaptively combines information from different temporal scales through parallel hidden states. Leveraging meteorological indicators and four air quality indicators, we present comprehensive comparative analyses and ablation experiments, showcasing the higher accuracy of M2G2 in comparison to nine currently available advanced approaches across all aspects. The improvements of M2G2 over the second-best method on RMSE of 72-h future predictions are as follows: PM2.5: 6%∼10%; PM10: 5%∼7%; NO2: 5%∼16%; O3: 6%∼9%. Furthermore, we demonstrate the effectiveness of each module of M2G2 by ablation study. We conduct a sensitivity analysis of air quality and meteorological data, finding that the introduction of O3 adversely impacts the prediction accuracy of PM2.5.