Street identification on historical maps is essential for investigating urban historical landscapes and advancing heritage conservation. Conventional methods rely on manual feature engineering to extract street characteristics, which limits their accuracy and generalizability. Although deep learning automates high-dimensional embedding extraction, it performs poorly under spatiotemporal variations and results in reduced ineffectiveness and limited interpretability. To address this limitation, this study proposes a novel multi-modal historical street identification network (MHSI-Net) which integrates street-related semantics to identify streets from specific construction periods in historical maps to achieve fine-grained point-to-line identification. It comprises three primary modules: multi-modal representation learning, multi-modal interactive fusion and interpretable attention-enhanced modules. These modules are designed to extract features of street morphology from street maps and street structures from street graphs, facilitate cross-scale semantic fusion, capture local-global connections and improve the analysis of street morphology and distribution. In the task of identifying Yuan Streets (1267–1368) on the 1937 Beijing map, our MHSI-Net-based model achieved superior performance compared with a series of benchmarks. The interpretability analysis further quantified the contributions of multi-modal interactive effects and revealed the influence of street structure and morphological features on the identification process. This analysis provides valuable insights into model decision-making.

Human mobility prediction is a crucial task in various domains, including urban planning, traffic management, disaster risk reduction, and public health. However, accurately predicting intercity mobility remains challenging due to the complex correlations among flows and inherent spatiotemporal uncertainties. Recent advances in deep learning have improved prediction accuracy but often at the cost of increasing algorithmic complexity. In this paper, we propose a simple yet effective approach that leverages large language models (LLMs) with supervised fine-tuning (SFT) only. By formulating human mobility prediction as a sequence generation task, our method efficiently adapts pre-trained LLMs to the competition dataset. Despite its simplicity, the proposed approach achieved competitive performance across all four regions, surpassing baseline methods. Experimental results demonstrate that LLMs fine-tuned with SFT can effectively capture spatiotemporal dynamics and generalize across cities with minimal effort, providing a promising direction for efficient and transferable urban mobility modeling.

Understanding the response of human mobility to disruptive weather events is beneficial for the development of urban risk mitigation and emergency response policies, thus enhancing urban resilience. Most human mobility studies relying on aggregate flow data inevitably neglect the heterogeneity of disaggregate travel patterns with distinctive spatiotemporal characteristics, causing the uncertainty problem for identifying meaningful travel behaviors. Moreover, there is a lack of robust methodological approaches to extracting stable and genuine travel patterns under normal or disruptive situations. To address these issues, this study proposes a data-driven approach to spatiotemporal flow decomposition based on non-negative matrix factorization. With sparseness factored in the decomposition, stable disaggregate travel patterns can be extracted from origin-destination mobility flows. By combining temporal, spatial, and urban functional perspectives, heterogeneous travel behaviors can be analyzed and inferred. With a case study of the Zhengzhou ‘7.20’ heavy rainfall in 2021, the most extreme rainfall ever recorded in China, this study validated the effectiveness of the proposed approach and managed to identify representative and interesting travel patterns and behaviors, facilitating a better understanding of human travel behaviors under external impacts. In practice, this study can provide valuable insights for coping strategies in the face of increasingly frequent disruptive events.

Building function identification (BFI) is crucial for urban planning and governance. The traditional remote sensing approach primarily focuses on extracting the physical features of buildings, overlooking their functional uses. Recently, progress has been made in urban functional area identification through multi-modal representation learning from multi-source spatial big data. However, the two approaches are disconnected, and each approach is inadequate to tackle the fine-grained BFI problem solely. To address this challenge, this study proposes a multi-modal foundation model for BFI, called BF-SAM, by fine-tuning a large visual model, Segment Anything Model (SAM), with multi-modal features related to urban functions. This model harnesses the segmentation capability of SAM for building delineation and fuses it with multi-modal representation learning for functional identification through a novel multi-modal fine-tuning method for SAM. Modality-dedicated feature extraction methods are devised to learn geographic features from road networks, population density, and points of interest. The validity of BF-SAM was evaluated on datasets from Munich, Beijing, Suzhou, and Hefei, and the importance of multi-modal geographic features was examined through extensive experiments. BF-SAM achieved a superior performance compared to a series of benchmarks. The potential of model transferability of BF-SAM was further explored under different spatial contexts.

The rapid development of generative artificial intelligence (GenAI) presents new opportunities to advance the cartographic process. Previous studies have either overlooked the artistic aspects of maps or faced challenges in creating both accurate and informative maps. In this study, we propose CartoAgent, a novel multi-agent cartographic framework powered by multimodal large language models (MLLMs). This framework simulates three key stages in cartographic practice: preparation, map design, and evaluation. At each stage, different MLLMs act as agents with distinct roles to collaborate, discuss, and utilize tools for specific purposes. In particular, CartoAgent leverages MLLMs’ visual aesthetic capability and world knowledge to generate maps that are both visually appealing and informative. By separating style from geographic data, it can focus on designing stylesheets without modifying the vector-based data, thereby ensuring geographic accuracy. As a result, the proposed CartoAgent could effectively produce maps that are not only visually appealing but also accurate and informative. We applied it to a specific task centered on map restyling, namely, map style transfer and evaluation. The effectiveness of this framework was validated through extensive experiments and a human evaluation study. CartoAgent can be extended to support a variety of cartographic design decisions and inform future integrations of GenAI in cartography.

Intercity human dynamics has involved increasing interactions between the physical world and cyberspace with the boom of the Internet and social media, expanding traditional intercity activities from a single dimension of physical space to a multidimensional hybrid space. Revealing the dynamics of intercity activities in a hybrid physical–virtual space requires considering interactions between human activities in both the physical world and cyberspace. However, existing studies often overlook cross-space interactions, failing to adequately model interactions between virtual and physical spaces in terms of the network structure and spatial effects. To address these gaps, this study investigates intercity human dynamics through travel movement–intention interactions. A multilayer network framework is proposed to represent the hybrid space, where travel movement–intention interactions can be conceptualized and measured. Using travel and search flows in China during the ‘Labor Day’ holiday periods in a three-year period of the pandemic, we demonstrate the proposed approach’s utility through an analysis of city centrality of multilayer networks. Results find that travel movement–intention interactions exhibit dynamic patterns that vary with city size and geographic distance. This approach is broadly applicable to studies of hybrid physical–virtual spaces, aiding the evaluation of human dynamics beyond intercity activities.

Urban land use is central to urban planning. With the emergence of urban big data and advances in deep learning methods, several studies have leveraged graph convolutional networks (GCNs) with local functional characteristics from points of interest data and spatial features from flow data to infer urban land use. However, these studies cannot distinguish spatial interaction and spatial dependence in terms of conceptualization and modeling mechanisms and overlook the inadequacy of GCNs in modeling spatial interaction. This study proposes a novel framework—a heterogeneous graph convolutional network (HGCN)—to explicitly account for the spatial demand and supply components embedded in spatial interaction data. Several experiments, including 19 different models and datasets from Shenzhen and London, were conducted to validate the proposed framework and its generalizability within the same and different spatial contexts. The HGCN can distinguish heterogeneous mechanisms in supply- and demand-related modalities of spatial interactions, incorporating both spatial interaction and spatial dependence for urban land-use inference. Empowered by HGCN, we found that spatial interaction features play a distinctively crucial role in urban land-use inference compared to local attributes and spatial dependence features. In addition, our findings highlight the superiority of HGCN-based models in boosting performance and enhancing model transferability.

The computational approach of agent-based models (ABMs) supports the representation of interactions among spatially situated individuals as a decentralized process giving rise to space–time complexity in geographic systems. To cope with the computational complexity of these models, this article proposes a parallel approach that leverages the power of multicore systems, as these architectures have quickly become ubiquitous in high-performance and desktop computing. An ABM of individual-level spatial interaction that simulates information exchange, spatial diffusion of opinion development, and consensus building among decision makers is proposed to demonstrate the advantages of the parallel approach against its sequential counterpart. This study focuses on two key spatial properties of the interaction system of interest, the extent and range of interaction, and examines their influence on the computing performance of the proposed parallel model and the performance scalability of the model as more computing resources are added. Significant influence from these two properties is found and can be attributed to three possible sources of effects, namely the model level, the parallelization level, and the platform level. It is suggested that these effects should be taken into consideration when leveraging multicore computing resources for the development of parallel ABMs.