The challenge that motivated the work is structural. Airspace is divided into sectors, and each sector’s complexity depends on two kinds of relationships that are difficult to model together. Spatially, sectors are connected both by simple geographical adjacency, meaning they share a border, and by dynamic traffic flows, since aircraft entering one sector typically exited another, creating dependencies that shift with the daily rhythm of departures and arrivals. Temporally, complexity evolves across multiple time scales at once: fast fluctuations driven by individual climbing and descending aircraft, medium-term waves tied to airport scheduling, and long-range dependencies that link morning traffic management decisions to afternoon congestion. Existing spatio-temporal prediction frameworks, including the graph neural networks that have dominated traffic forecasting in recent years, tend to rely on static graphs that freeze these relationships in place and on shallow temporal alignment that struggles to connect patterns separated by long gaps in time. The result is a systematic weakness precisely where forecasters need the most help: high-complexity situations and long prediction horizons.

MAST-LLM, which stands for a Multimodal Adaptive Spatio-Temporal framework enhanced by Large Language Models, addresses these weaknesses in three coordinated stages. The first stage is a temporal alignment phase built on a Mixture-of-Experts architecture. A Mixture-of-Experts model is a neural network design in which multiple specialized subnetworks, the experts, each process the input, and a gating mechanism learns to weight their contributions depending on the character of the data at hand. In MAST-LLM, this design lets the framework capture heterogeneous temporal patterns, so that the distinct rhythms of different air traffic variables can each be handled by appropriately specialized experts. Critically, this phase also aligns domain-specific air traffic sequences with the representations that a pre-trained language model has already learned. Rather than forcing raw aviation data into a model built for text, the framework translates traffic dynamics into a form that the language model’s internal representations can meaningfully encode, bridging the gap between two very different data worlds.

The second stage is a full spatio-temporal fine-tuning phase, in which the framework integrates adaptive multimodal spatial representations with multi-scale temporal features. The word multimodal here refers to the combination of different kinds of information about the airspace, such as structural spatial relationships and dynamic traffic measurements, into a shared representation that can adapt as conditions change rather than remaining fixed. The centerpiece of this stage is a mechanism the authors call Bidirectional Spatio-Temporal Attention, or BSTA. Attention mechanisms, first popularized by the transformer architecture underlying modern language models, allow a network to weigh the relevance of every element of its input against every other element. BSTA extends this idea so that spatial and temporal information interact in both directions: spatial structure informs how temporal patterns are interpreted, and temporal evolution informs how spatial relationships are weighted. This joint interaction modeling is what allows the framework to reason about the airspace as a single coupled system rather than as separate spatial and temporal problems stitched together.

The third stage exploits the property that makes large language models attractive for this task in the first place: their capacity for global reasoning and long-range dependency modeling. Language models are trained on sequences in which meaning can depend on context established thousands of tokens earlier, and their architectures are built to preserve and use that distant context. By establishing a unified representation that captures both local dynamics, the minute-to-minute behavior of individual sectors, and global contextual dependencies, the patterns that propagate across an entire region’s airspace over hours, MAST-LLM inherits this long-context strength. The authors argue that this is precisely what earlier frameworks lacked: a way for a prediction about one sector at one moment to draw on evidence from distant sectors and distant times within a single coherent computation.

The empirical results reported in the paper are striking. In extensive experiments, MAST-LLM achieved superior performance at short- and medium-term forecasting horizons and remained competitive at longer horizons, with improvements of up to 9.5 percentage points in accuracy and 12.9 percentage points in macro F1-score over the strongest baselines. The macro F1-score is a particularly meaningful metric here because it averages the F1-score, which balances precision and recall, across all complexity classes, ensuring that improvements on rare but dangerous high-complexity conditions count as much as improvements on routine ones. The authors supplemented the headline numbers with comprehensive ablation studies, which remove individual components to verify that each one contributes, along with factor importance analyses, sensitivity analyses, and visualization analyses that together confirm the effectiveness and robustness of every core element of the design. Statistical significance was assessed with paired two-sided t-tests against the strongest baseline, with exact p-values reported in the paper’s appendix.

The work builds on a substantial lineage of research into both airspace complexity and spatio-temporal machine learning. Measures of air traffic complexity stretch back decades, from early workload prediction studies by Chatterji and Sridhar to probabilistic complexity measures in three-dimensional airspace developed by Prandini and colleagues, and to spatiotemporal graph indicators proposed by Isufaj and collaborators. On the machine learning side, the framework draws on the spatio-temporal graph convolutional networks introduced by Yu, Yin, and Zhu in 2018 and the diffusion convolutional recurrent networks of Li and colleagues from the same year, as well as more recent attention-based architectures such as GMAN and PDFormer. Notably, the same research group had previously developed MAST-GNN, a multimodal adaptive spatio-temporal graph neural network for the same prediction task, and MAST-LLM can be seen as an evolution of that line of work, replacing static graph reasoning with the adaptive, language-model-driven approach.

The study also sits within a rapidly growing movement to apply large language models to time-series and traffic problems. Recent research has shown that pre-trained language models can be reprogrammed for general time-series analysis, as in the One Fits All work of Zhou and colleagues, and for dedicated forecasting frameworks such as Time-LLM by Jin and colleagues and LLM4TS by Chang and colleagues. Parallel efforts have applied these ideas to wind power and wind speed forecasting with BERT4ST, STELLM, and STCA-LLM, to spatio-temporal imputation with GATGPT, and to urban traffic prediction with ST-LLM plus. A 2026 survey by Long and colleagues in IEEE Transactions on Big Data catalogues the accelerating adoption of language models across traffic forecasting applications. MAST-LLM distinguishes itself within this crowded field by combining the Mixture-of-Experts alignment strategy with bidirectional spatio-temporal attention, a pairing the authors present as tailored to the specific structure of airspace dynamics rather than borrowed wholesale from other domains.

The practical implications could be considerable. Accurate complexity forecasts feed directly into demand-capacity balancing, the process by which air navigation service providers decide how much traffic each sector can safely absorb and where flow restrictions should be imposed. Better predictions, especially at medium and long horizons, give managers more lead time to reroute flights, adjust sector configurations, and staff control positions appropriately, potentially reducing both delays and controller overload. The authors have made the airspace complexity dataset used in the study publicly available through a project repository, lowering the barrier for other groups to build on the approach. The research was supported by funders including the National Natural Science Foundation of China and the Ministry of Education of China, and the authors note that generative AI tools were used to assist with language refinement of the manuscript, with all content reviewed and verified by the team.

For the field of air traffic management, the study offers a proof of concept that the reasoning machinery of large language models, originally built for human language, can be redirected toward the physical dynamics of the sky. For the broader machine learning community, it adds to mounting evidence that pre-trained language models serve as powerful general-purpose sequence models whose learned representations transfer far beyond text. Whether such frameworks can be deployed in the safety-critical, certification-heavy environment of real air traffic control remains an open question, and the authors’ results, while strong, come from benchmark evaluation rather than live operations. Still, as global air traffic continues to grow toward and beyond pre-pandemic levels, tools that can anticipate when a sector is about to become unmanageable, minutes or hours in advance, address one of the most consequential prediction problems in transportation, and MAST-LLM demonstrates that the newest generation of AI models may be up to the task.

Subject of Research: Large language model-driven multimodal spatio-temporal prediction of air traffic complexity

Article Title: MAST-LLM: A large language model-driven multimodal adaptive spatio-temporal framework for air traffic complexity prediction

Article References: Cheng, R., Fan, J., Zhang, C., Wu, M., Chen, R., Chai, M., & Wang, A. (2026). MAST-LLM: A large language model-driven multimodal adaptive spatio-temporal framework for air traffic complexity prediction. Applied Intelligence, 56(15), Article 442. https://doi.org/10.1007/s10489-026-07469-7

Image Credits: AI Generated

DOI: 10.1007/s10489-026-07469-7

Keywords: air traffic complexity, large language models, spatio-temporal prediction, mixture of experts, attention mechanism, graph neural networks, air traffic management, time series forecasting, Applied Intelligence, machine learning, airspace sectors, macro F1-score

The core insight is deceptively simple. Most traffic forecasting models learn correlations: when sensor A reads a slowdown, sensor B often does too. But correlations say nothing about direction. Granger causality, a technique dating back to the 1960s, asks a more useful question: does the past of one time series improve predictions of another? If knowing the history of sensor A lets you forecast sensor B better than B’s own history alone, then A carries directional predictive information about B. The Pisa team treats this statistical signal as a prior, a structural hint about how congestion actually propagates along a road network.

GC-MGT is built from four cooperating modules. The first is a hierarchical dynamic dependency module that runs a linear Granger test over sliding windows of one day’s observations, sampled every five minutes, with a lag order of twelve steps. Because the model runs thousands of pairwise tests, the authors apply the Benjamini-Hochberg procedure to control the false discovery rate at five percent, keeping only statistically defensible directional links. That fixed Granger graph is then multiplied by a sigmoid-bounded neural refinement that adapts edge strengths to the current traffic state, capturing nonlinear, time-varying interactions that a purely linear test would miss. Regularization terms encourage sparsity and temporal smoothness so the dependency graph does not thrash from one input window to the next.

The second module fuses three different views of the network. A physical topology graph encodes actual road connections, weighted by shortest-path distance with a five-kilometer connectivity threshold. A semantic graph, learned through multi-head attention over recent observations, captures sensors whose traffic patterns behave alike even if they sit far apart. The causality-inspired graph from the first module supplies directionality. A small neural network generates learnable, state-dependent weights that blend all three into a single fused adjacency matrix at every time step. During free-flowing conditions the semantic graph may dominate; during a merge-zone jam, the physical and directional graphs take over.

The third module is the Transformer itself, but modified in two crucial ways. The fused graph is injected directly into the attention logits as a bias term, so the model’s attention literally leans along the directions where traffic influence flows. In parallel, four convolutional branches with kernel sizes of one, three, five, and seven extract patterns at multiple temporal scales, from abrupt braking events to daily commute rhythms. A learned gating mechanism balances the convolutional and attention streams, and a lightweight non-autoregressive decoder produces all forecast horizons simultaneously, up to sixty minutes ahead.

The fourth module is the most unusual: a reality check performed inside a traffic simulator. Using the Simulation of Urban Mobility platform, or SUMO, the team reconstructed the Florence-Pisa-Livorno highway corridor, calibrating it with a Greenshields fundamental diagram until simulated speeds matched observations with a mean absolute percentage error of just 8.6 percent. They then ran three intervention experiments, reducing a mainline speed limit from 110 to 80 kilometers per hour, cutting ramp inflow by 30 percent, and closing an acceleration lane, each repeated ten times with different random seeds. By comparing the model’s predicted responses to SUMO’s simulated propagation, and measuring the overlap of affected sensors with a Jaccard index, they tested whether the learned dependencies actually reflect how disturbances travel through a real highway.

The results are striking. Across three datasets, the California PeMS08 benchmark with 170 sensors, the San Francisco Bay Area PEMS-BAY dataset with 325 loop detectors, and the FI-PI-LI Tuscan highway with its notoriously tight Ginestra Fiorentina interchange, GC-MGT beat every baseline at every horizon. At the hardest sixty-minute forecast, relative to PDFormer, the strongest competing model, GC-MGT cut mean absolute error by 13.3, 21.3, and 12.0 percent across the three datasets, and root mean squared error by 3.2, 24.5, and 19.2 percent. The advantage was largest on FI-PI-LI, where complex merging zones produce chaotic, disturbance-driven fluctuations that conventional models handle poorly. Paired t-tests on five independent runs confirmed the improvements were statistically significant, not artifacts of random initialization.

Ablation experiments revealed exactly where the performance comes from. Removing the entire directional-dependency mechanism was devastating, inflating MAE by 23.3 percent on PeMS08, 38.1 percent on PEMS-BAY, and 54.6 percent on FI-PI-LI. The Granger prior and the neural refinement each contributed independently, and adaptive multi-graph fusion consistently outperformed fixed weighting. Crucially, in the SUMO intervention tests, GC-MGT achieved the lowest prediction error under intervention and the highest affected-sensor similarity in all three scenarios, reaching a Jaccard index of 0.87 in the speed-limit experiment, compared with 0.80 for Graph WaveNet and 0.75 for PDFormer. The model’s internal dependency structure, in other words, behaves like genuine traffic physics.

The authors are careful about what this means. Granger analysis, they stress, demonstrates predictive precedence, not structural causation; the learned graph is a map of directional predictive dependencies, not a proof of mechanism. Still, the practical implications are considerable. Because the framework is both accurate and interpretable, the researchers envision integrating it with digital twins and reinforcement learning controllers to drive proactive traffic management: variable speed limits, adaptive ramp metering, and scenario-based optimization that responds to predicted congestion before it materializes. Congestion costs cities billions in fuel, time, and greenhouse gas emissions each year, and the TomTom traffic index shows most major cities wrestling with worsening gridlock.

What makes this work resonate beyond transportation is its methodological message. Transformers, the architecture behind modern language models, are superb at capturing long-range patterns but notoriously blind to why things happen. By anchoring attention in statistically tested directional dependencies, and then validating those dependencies against simulated interventions, the Pisa team has sketched a template for making black-box forecasting models legible without sacrificing accuracy. It is a rare case where an economic statistic from 1969 and a cutting-edge neural architecture find common cause on a Tuscan highway, and both come out ahead.

Subject of Research: Granger causality-guided deep learning framework for spatio-temporal traffic flow forecasting

Article Title: A Granger causality-guided multi-graph transformer framework for traffic flow forecasting

Article References: Wang, C., Riccardi, C., Fiorentini, N., & Losa, M. (2026). A Granger causality-guided multi-graph transformer framework for traffic flow forecasting. Discover Artificial Intelligence, 6(1), Article 1219. https://doi.org/10.1007/s44163-026-02230-y

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02230-y

Keywords: traffic flow forecasting, Granger causality, transformer, multi-graph neural networks, spatio-temporal prediction, directional dependency, SUMO simulation, deep learning, traffic management, digital twins, intelligent transportation systems, interpretability

The research tackles a problem that has long plagued computational seismology: extreme class imbalance. When earthquake forecasting is cast as a grid-based classification task, the region under study is divided into spatial cells, and the model must predict whether a seismic event will occur in each cell during each time window. For Central Asia, the team discretized the territory into one-degree by one-degree grid cells and attempted to forecast earthquakes of magnitude 3.0 or greater on a weekly basis. Under this formulation, roughly 96 percent of all cell-week combinations contain no event at all, a phenomenon known as zero inflation. The true event prevalence sits at only about 4.5 percent, which means that a model doing nothing more than predicting ‘no earthquake’ everywhere would still appear superficially accurate while being scientifically useless.

This imbalance has profound consequences for how forecasting models must be evaluated. Standard accuracy metrics become meaningless when negative cases dominate by more than twenty to one. The researchers therefore anchored their evaluation in the Precision-Recall Area Under the Curve, a metric that is far more sensitive to performance on the rare positive class. With a prevalence of 4.5 percent, the constant baseline for PR-AUC is 0.045, meaning any model must substantially exceed that value to demonstrate genuine predictive skill. The achieved score of 0.451 represents a tenfold improvement over this baseline, a substantial margin in a domain where even modest gains above chance are considered meaningful by the seismological community.

Central to the study is a carefully engineered feature space grounded in earthquake physics rather than raw statistical patterns. The framework integrates tectonic regime-conditioned normalization, which allows the model to account for the fact that different tectonic settings produce fundamentally different seismic behavior, so that features extracted from a thrust-fault environment are not treated as directly comparable to those from a strike-slip regime. It also incorporates Omori energy decay proxies, mathematical representations of the well-documented tendency of earthquake sequences to produce aftershocks whose frequency decays over time following a mainshock. These proxies give the models a physically interpretable signal about the temporal clustering of seismicity, encoding decades of seismological understanding directly into the input data.

Structural fault descriptors form a third pillar of the feature design. The geometry, orientation, and proximity of mapped fault systems are among the strongest known controls on where earthquakes occur, and by encoding these structural characteristics as model inputs, the framework ensures that the learning algorithms operate on geologically meaningful quantities rather than arbitrary grid statistics. The final and perhaps most consequential innovation is log-odds baseline initialization, a technique that encodes the historical cell-specific event rate directly into the learning objective. Instead of forcing each model to rediscover from scratch the simple fact that some grid cells are historically far more seismically active than others, the initialization embeds this prior knowledge into the model’s starting point, allowing learning effort to focus on deviations from the historical pattern.

To determine whether performance under such extreme imbalance is governed primarily by model architecture or by structured feature design, the researchers evaluated six heterogeneous architectures under a strict chronological split, ensuring that models were trained only on past data and tested on future periods, exactly as an operational forecasting system would be deployed. The architectures spanned a wide methodological range, including CatBoost and other gradient boosting methods, which excel at tabular data, and Bi-LSTM networks, a bidirectional long short-term memory architecture capable of capturing temporal dependencies in sequential data. Despite their radically different inductive biases and internal mechanics, the models converged on nearly identical PR-AUC and ROC-AUC values, a result the authors interpret as evidence that the information content of the feature space, not the capacity of the learner, is the binding constraint.

Equally notable is what the framework does not do. Many studies confronting severe class imbalance resort to synthetic resampling techniques, such as oversampling the rare event class or undersampling the dominant negative class, to artificially balance the training distribution. These methods can distort the learned probability calibration, producing models whose confidence scores no longer correspond to real-world event likelihoods. The Central Asia framework achieves its tenfold improvement entirely without synthetic resampling, preserving the integrity of the probability estimates. This matters enormously for practical applications, because emergency management authorities require calibrated forecasts whose stated probabilities can be trusted when weighing evacuation decisions, infrastructure inspections, and public warnings.

The authors argue that their findings point to the existence of a macro-scale predictability ceiling in seismic forecasting. If architecturally diverse models, given the same physically structured inputs, all plateau at the same performance level, the implication is that the remaining unpredictability reflects genuine stochasticity in the earthquake process at this spatial and temporal resolution, rather than a deficiency of current algorithms. This interpretation carries a sobering but valuable message for the field: further architectural innovation alone is unlikely to break through the ceiling, while improvements in physical understanding, richer observational data streams, and better-calibrated baselines may still push the boundary outward. It also cautions against the common practice of claiming architectural superiority from small performance differences that may fall within the noise of a shared predictability limit.

The work was funded by the Committee of Science of the Ministry of Science and Higher Education of the Republic of Kazakhstan under a grant for developing a multifunctional system of ground-space monitoring and early warning of natural and technogenic emergencies, underscoring its operational motivation. For a country situated in one of the most seismically active zones of Central Asia, where the collision of the Indian and Eurasian plates drives hazardous tectonics through the Tien Shan and surrounding mountain belts, reliable macro-scale forecasting is not an academic curiosity but a matter of public safety. By demonstrating that disciplined feature engineering, physically informed priors, and rigorous baseline calibration can deliver a tenfold gain in predictive skill without exotic machinery, the Almaty team has provided both a practical forecasting tool and a methodological lesson that resonates far beyond seismology: in data-starved, imbalance-dominated problems, how you frame the question often matters more than how elaborate your model is.

Subject of Research: Machine learning earthquake forecasting under extreme class imbalance in Central Asia

Article Title: Macro-scale earthquake forecasting under class imbalance in Central Asia

Article References: Nurtas, M., Nurakynov, S., Sakabekov, A., Altaibek, A., Kumarkhanova, A., & Merekeyev, A. (2026). Macro-scale earthquake forecasting under class imbalance in Central Asia. Journal of Big Data. https://doi.org/10.1186/s40537-026-01544-z

Image Credits: AI Generated

DOI: 10.1186/s40537-026-01544-z

Keywords: earthquake forecasting, Central Asia, class imbalance, machine learning, CatBoost, gradient boosting, Bi-LSTM, PR-AUC, spatio-temporal prediction, Omori decay, fault descriptors, Kazakhstan