Digital Twins (DTs) are increasingly used to virtualise physical systems at different scales, enabling monitoring, simulation, and predictions to support decision-making. However, while individual DTs are effective in stand-alone settings, ecosystem-scale deployments require multiple autonomous and distributed DTs to cooperate across system boundaries despite differences in modelling approaches or software technologies, making interoperability and runtime coordination critical challenges. Although Federated Digital Twin Ecosystems have emerged as a promising direction, existing research remains at the conceptual stage, offering high-level architectures while leaving the practical integration of heterogeneous DTs underexplored. This paper proposes the Federation Node Manager, a modular integration mechanism that connects local DTs to a federated environment through controlled capability exposure, protocol and schema adaptation, and timely state and event exchange for coordinated operations. We present a conceptual design and a prototype implementation, and demonstrate their feasibility in the smart mobility domain for emergency response scenarios. The proposed mechanism serves as an enabling component within a broader service-oriented federated DT ecosystem.
Spatio-temporal forecasting on sensor graphs is commonly tackled with a single backbone architecture applied uniformly across all nodes, although graph regions can exhibit different dynamics. Road segments differ in functional class, structure, and traffic behavior, suggesting that node-wise expert specialization can be useful. We propose GC-MoE, a graph-conditioned mixture of experts framework that assigns each node a personalized combination of frozen forecasting experts based on graph topology and the recent traffic input window. GC-MoE combines frozen pretrained spatio-temporal GNN experts with an input-aware, spatially contextualized router while training only a lightweight routing module. We also study a bounded graph-conditioned output refinement layer as an optional extension and include node-adaptive ST-LoRA adapters only as an ablation diagnostic. Across four standard benchmarks (PEMS04, PEMS07, METR-LA, and PEMS-BAY), GC-MoE improves MAE over a zero-parameter ensemble baseline, with competitive RMSE and MAPE, while training only 17K parameters on top of 1.5M frozen expert weights. The implementation is available at https://github.com/Ahghaffari/gc_moe.
Rainfall-induced landslides pose a growing risk worldwide as climate change intensifies extreme rainfall events. To provide sufficient evacuation time, landslide early warning systems (LEWS) for real-time disaster monitoring must estimate near-future landslide risk by integrating observed rainfall with short-term rainfall forecasts from spatio-temporal environmental data streams. Although recent landslide prediction methods have improved predictive performance using statistical and deep learning approaches, most assume accurate rainfall inputs. In operational settings, however, landslide prediction relies on rainfall forecasts, which often contain spatial displacement of rainfall fields due to forecasting uncertainties. Such displacement can alter local accumulated rainfall and degrade prediction accuracy. To address this challenge, we propose a novel LEWS robust to rainfall field displacement. The key idea is to learn latent representations from rainfall and terrain data that remain stable under displacement in rainfall field motion, enabling reliable geospatial data integration for landslide risk estimation. The landslide prediction model is trained using Rainfall-Motion-Aware Contrastive Learning (RMCL), which introduces temporally correlated rainfall field perturbations to emulate forecast-induced displacement in rainfall-driven spatio-temporal environmental data streams. Experiments were conducted using two years of rainfall and terrain data across Japan, covering 19 regions with landslide events. The proposed system achieved up to 37
Despite having billions of data sensing devices that virtually acquire data about anything, anywhere, not all smart applications are intelligent. Context-awareness being at the core of intelligence, it is a contradiction founded on the impotence of these applications to effectively leverage this plethora of big data and infer context in real-time. Hence, in this paper, we propose a conceptual architecture and instantiate a working prototype of the Distributed Contextual Intelligence System (DCIS) - a ubiquitous, mobile edge-based context management mechanism. Unlike any traditional Context Management System, DCIS is an ultra lightweight, event-driven, dynamically scaling agent-based system that overcomes the barriers to ubiquitous contextfacilitating context anywhere, anytime, and everywhere so that any application can be intelligent. Our prototype - Zoolocity, is based on edge devices that execute light-weight context inference on data that it can acquire and dynamically collaborate with nearby edges in real-time. Thus, unlike any previous works that do not provide any evidence or motivation towards distributed context management, we demonstrate how applications can use DCIS agents to become fully context-aware about any scene it is involved in, unlocking any intelligent feature. This is enabled by two key novel contributions: a method to spontaneously share interoperable context and infer geospatially sparse complex activities in near real-time using generalised entity interaction modelling.
Vehicular traffic is one of the major sources of air pollution in urban settings, making it essential to clearly understand how much and where vehicle emissions impact residents. Recent approaches manage to yield pollution maps at the microscopic level by processing GPS trajectories of vehicles. That is achieved by applying mathematical models to estimate instantaneous emissions from GPS data, extending estimates to areas without data through missing data imputation, and further considering air dispersion factors. In this work, we leverage such inferred knowledge to implement an emission-aware pedestrian routing strategy and to study its impact on the reduction of exposure to vehicular pollutants and walking time. The study is realized through simulations of large masses of pedestrians over a medium-sized city in Italy, analyzing the interplay between the two factors - exposure versus walking time - in terms of time efficiency of paths and changes over existing habits both at a global and at an individual level. Experiments suggest that exposure-aware routing can yield a significant margin of improvement in health over most paths with minor effects on mobility, making it feasible and effective.
