Spatial flows between consecutive distribution snapshots describe how one configuration evolves into the next. Such panel flows are inferable, albeit challenging, from observations over space and time. Current flow generation models focus on cross-sectional scenarios and neglect the temporal nature of flows. The panel gravity model, although interpretable, relies on linear assumptions and overlooks the change of geographic contexts. Our work introduces a Deep Spatial Evolution Network (DSEN) to infer panel flows, i.e. the spatial evolution between two snapshots of spatial distributions. DSEN incorporates a cross-event context learner to encode the contextual features, and a gravity-informed spatial evolution decoder to learn latent evolutionary features. Using a device-level mobile positioning dataset in the Twin Cities Metropolitan Area, Minnesota, U.S., DSEN achieves a 14.0% correlation improvement and a 15.3% error reduction compared to baselines in inferring human flows during the 2021 Christmas holiday, while explaining evolution processes and flow directionalities via deep features. Further experiments across sampling ratios and unseen events demonstrated the robustness and generalizability of our model, respectively. Deep flow inference for panel data holds promise for advancing mobility studies at the GeoAI frontier, expanding access to flow data, and informing solutions to pressing human-environment challenges in future cities.

Home location detection is a foundational task in human mobility research and underpins a wide range of data-driven applications. Leveraging mobile phone data, this study develops a behavior-informed and geo-context-aware framework for inferring the home locations of mobile users from their historical visits. Unlike most existing methods that rely on heuristic algorithms or uniform assumptions across devices, our approach captures individual-level behavioral differences in terms of mobility characteristics, such as activity timing, frequency, and spatial range, and accounts for geographic disparities of these behaviors across urban, suburban, and rural areas. The proposed framework first extracts user-specific representative activity locations using a refined density-based clustering algorithm, then classifies users into behaviorally distinct groups, and finally estimates each user's most probable home location for evaluation and downstream applications. Applied to one month of anonymized mobile phone data in Minnesota, USA, our framework achieves higher detection rates and stronger alignment with land use patterns than traditional approaches, particularly for users with sparse or irregular mobility and in less densely populated regions. We further illustrate the practical utility of the proposed framework through two example applications, highlighting its potential to support transportation planning, demographic studies, and broader geospatial applications.

Collective origin-destination (OD) flows refer to the total number of visits or visitors from origins to destinations within a region, providing the basic information for transportation planning, mobility demand, and so on. Recently, the universal visitation law provides a theoretical foundation for understanding the statistical structure of OD flows by modeling their underlying visit density as an inverse-square function weaving distance, visiting frequency, and attractiveness. However, its closed-form assumes that key quantities, such as attractiveness, travel friction, and frequency spectrum, are externally known or can be easily estimated, limiting its applicability in real-world, where such factors are often latent, interdependent, and governed by nonlinearities. In this work, we propose a deep graph learning approach that incorporates the visitation law as a structural prior and applies additional physical constraints on neural networks to guide OD flow imputation. Experiments based on real mobility data show that our method effectively imputes OD flows in a data-driven yet theory-informed manner.

Uncovering the collective behavior of flows among locations is critical to understanding the structure within an ever-changing spatial network. When a network evolves, there may exist subgraphs within which the internal flows generally follow a rule: the change rates of the flow weight are either collectively high or low. Classic network measures such as degree, clustering, and betweenness can be used to quantify the process of network evolution by profiling the overall characteristics over time. However, it remains challenging to elucidate how a spatial network is evolving without looking at structures where collective changes emerge. To bridge this gap, we introduce the concept of the Collective Flow-Evolutionary Pattern (CFEP) as a mesoscopic description for spatial network evolution. Four types of patterns with distinct features are defined to clarify the collective behaviors of the flow-evolutionary characteristics. We provide an analytical framework that utilizes flow change rates between two snapshots of the spatial network to detect CFEPs as optimized flow evolution (evo-groups). Synthetic experiments are presented to validate the method. A case study of large-scale individual mobile positioning data is conducted in the Twin Cities Metropolitan Area, Minnesota, US to demonstrate how CFEP can effectively understand the evolution of human mobility networks.

Extreme snowfall events significantly affect human dynamics, challenging transportation safety. Current research on inferring mobility patterns post-snowfall is hindered by scale-free network connectivity, abrupt contextual changes, intricate mechanisms of spatial interaction, and multivariate feature handling. This paper proposes the Weather-Informed Mobility Network (WIMN), a graph deep learning model designed to infer the post-snowfall human movement flows within mobility networks. WIMN treats each region of interest as a spatially-embedded graph and has three characteristics: (i) a geo-attraction block informed by the gravity law, recalibrating the spatial-wise importance of locations and interactions; (ii) an Encoder-Decoder structure where Encoder learns stable spatial dependencies and Decoder captures the snow-induced dynamic changes; and (iii) a Pareto-inspired hub selection, processing locations hierarchically according to their degree centrality within the network context. Applied to the Twin Cities Metropolitan Area, Minnesota, USA, WIMN outperformed baseline methods by up to 18.5% in inferring the redistribution of human mobility flows. Additionally, WIMN demonstrated spatial explainability across the learned deep weights, snowfall intensity, and flow reductions. These findings highlight WIMN’s effectiveness in understanding short-term human dynamics after extreme weather events, with potential applications in aiding transportation planning and disaster response.

Accurately predicting the next track point of vehicle travel is crucial for various Intelligent Transportation System (ITS) applications, such as travel behavior studies, traffic control, and traffic congestion monitoring. Recent works on trajectory prediction follow a paradigm that first represents the raw trajectory and subsequently makes predictions based on that representation. Currently, trajectory representation methods tend to project trajectory points to road networks by map matching and represent trajectories based on the representation of matched roads. However, precisely matching trajectories to roads is a challenge in ITS, as the matching precision is greatly affected by the quality of the trajectory. Meanwhile, since it is difficult to discern whether trajectory matching results are accurate or confounded, how to effectively utilize this type of uncertain geographic context information is also a challenge, which is defined as the Uncertain Geographic Context Problem (UGCoP) in geographic information science. Therefore, we propose a flexible strategy of subgraph learning, referred to as SLM, for predicting the next track point of vehicles. Specifically, a subgraph generation module is first proposed to extract topology contextual information of the roads around historical trajectory points. Secondly, a subgraph learning module is designed to learn rich spatial and temporal features from generated subgraphs. Finally, the extracted spatiotemporal features will be fed into a prediction module to predict the next track points of vehicles on road networks. Our model enables the effective utilization of uncertain geographic context information of trajectories on road networks while avoiding the error brought by map matching. Extensive experiments based on trajectory datasets in two different cities confirm the effectiveness of our approach.

Spatial heterogeneity refers to uneven distributions of geographical variables. Spatial interpolation methods that utilize spatial heterogeneity are sensitive to the way in which spatial heterogeneity is characterized. This study developed a Generalized Heterogeneity Model (GHM) for characterizing local and stratified heterogeneity within variables and to improve interpolation accuracy. GHM first divides a study area into multiple spatial strata according to the sample values and locations of a variable. Then, GHM estimates simultaneously the spatial variations of the variable within and between the spatial strata. Finally, GHM interpolates unbiased estimates and uncertainty at unsampled locations. We demonstrated the GHM by predicting the spatial distributions of marine chlorophyll in Townsville, Queensland, Australia. Results show that GHM improved both the overall interpolation accuracy across the study area and along strata boundaries compared with previous interpolation models. GHM also avoided bull’s eye patterns and abrupt changes along strata boundaries. In future studies, GHM has the potential to be integrated with machine learning and advanced algorithms to improve spatial prediction accuracy for studies in broader fields.

Simulating urban morphology with location attributes is a challenging task in urban science. Recent studies have shown that Generative Adversarial Networks (GANs) have the potential to shed light on this task. However, existing GAN-based models are limited by the sparsity of urban data and instability in model training, hampering their applications. Here, we propose a GAN framework with geographical knowledge, namely Metropolitan GAN (MetroGAN), for urban morphology simulation. We incorporate a progressive growing structure to learn hierarchical features and design a geographical loss to impose the constraints of water areas. Besides, we propose a comprehensive evaluation framework for the complex structure of urban systems. Results show that MetroGAN outperforms the state-of-the-art urban simulation methods by over 20% in all metrics. Inspiringly, using physical geography features singly, MetroGAN can still generate shapes of the cities. These results demonstrate that MetroGAN solves the instability problem of previous urban simulation GANs and is generalizable to deal with various urban attributes.

Spatial interpolation is a traditional geostatistical operation that aims at predicting the attribute values of unobserved locations given a sample of data defined on point supports. However, the continuity and heterogeneity underlying spatial data are too complex to be approximated by classic statistical models. Deep learning models, especially the idea of conditional generative adversarial networks (CGANs), provide us with a perspective for formalizing spatial interpolation as a conditional generative task. In this article, we design a novel deep learning architecture named conditional encoder-decoder generative adversarial neural networks (CEDGANs) for spatial interpolation, therein combining the encoder-decoder structure with adversarial learning to capture deep representations of sampled spatial data and their interactions with local structural patterns. A case study on elevations in China demonstrates the ability of our model to achieve outstanding interpolation results compared to benchmark methods. Further experiments uncover the learned spatial knowledge in the model’s hidden layers and test the potential to generalize our adversarial interpolation idea across domains. This work is an endeavor to investigate deep spatial knowledge using artificial intelligence. The proposed model can benefit practical scenarios and enlighten future research in various geographical applications related to spatial prediction.

Spatial interactions underlying consecutive sequential snapshots of spatial distributions, such as the migration flows underlying temporal population snapshots, can reflect the details of spatial evolution processes. In the era of big data, we have access to individual-level data, but the acquisition of high-quality spatial interaction data remains a challenging problem. Most research has been focused on distributions of movable objects or the modelling of spatial interaction patterns, with few attempts to identify hidden spatial interaction patterns from temporal transitions of spatial distributions. In this article, we introduced an approach to infer spatial interaction patterns from sequential snapshots of spatial population distributions by incorporating linear programming and the spatial constraints of human movement. Experiments using synthetic data were conducted using four simple scenarios to explore the characteristics of our method. The proposed method was used to extract interurban flows of migrants during the Chinese Spring Festival in 2016. Our research demonstrated the feasibility of using discrete multi-temporal snapshots of population distributions in space to infer spatial interaction patterns and offered a general analytical framework from snapshot data to spatial interaction patterns.