Interactive visual query systems are essential for the exploration and analysis of geospatial data. However, developing such systems has become increasingly challenging in recent years due to the conflict between the unprecedented volume of data and the need for instantaneous feedback. To address this challenge, we present HiVQ, a High-performance Visual Query system for real-time interactive visual query of geospatial big data. HiVQ adopts an innovative “Query as Visualization” paradigm, transforming user interactions into pixel value queries which can be processed efficiently with specialized indices and optimization strategies. Unlike conventional solutions that query and visualize geospatial objects sequentially, HiVQ effectively omits most geospatial objects and unnecessary computations that do not affect the final visualization, ensuring minimal sensitivity to data volume. Experimental results show that HiVQ accelerates visual queries by at least seven times compared to SOTA methods. This demonstration enables users to interactively explore and analyze spatial data with billions of nodes at any scale, receiving responses in milliseconds as they dynamically adjust analysis parameters, query conditions, or map styling. The demonstration video is available at https://gitee.com/kyrie-Bang/HiVQ-Demo.

In the rapidly evolving global shipping industry, accurate vessel trajectory prediction is essential for effective maritime traffic management. However, the inherent uncertainties in vessel voyages pose significant challenges for existing methods in achieving precise long-term predictions. Inspired by the success of large language models (LLMs) in natural language processing, we propose SEMINT—a novel framework that integrates SEMantic cognition and INTent context-awareness for long-term vessel trajectory prediction. SEMINT leverages LLMs to semantically interpret vessel navigation behaviors and infer multiple voyage intents. The final intent inference is derived from the consensus among multiple reasoning chains, which are then combined with historical preferences to form an intent context. Guided by this context, a Transformer-based trajectory prediction model captures the dependencies between critical waypoints and voyage intent, thereby achieving stable long-term numerical predictions. SEMINT combines the advantages of LLMs and task-specific deep learning models, transforming voyage uncertainties into high-level intent diversity and simplifying the learning process for long-term movement patterns. Experiments on real-world datasets show that SEMINT outperforms previous methods, reducing average and final displacement errors by 28.47% and 31.59%, respectively.

Large Language Models (LLMs) demonstrate various capabilities previously considered unique to humans. However, current evidence is insufficient to determine whether LLMs have developed spatial cognition, a fundamental aspect of human cognition underpinning logical-mathematical reasoning and various other skills. Previous studies on this topic have primarily concentrated on small-scale perceptions, leaving the spatial cognition within the context of GIScience largely unexamined. We introduce a benchmark that evaluates spatial cognition abilities across seven categories to systematically assess how well LLMs process and generate three types of spatial knowledge: landmark, route, and survey knowledge. Furthermore, we propose a tool-augmented approach named Hybrid Mind, which integrates LLMs with deterministic GIS algorithms to enhance their performance in spatial cognitive tasks. The core idea involves the implementation of a mental map builder that generates a quantitative map based on segmented qualitative constraints, overcoming LLMs’ fallacies in synthesizing spatial information. Our experimental results revealed that although LLMs exhibited potential for spatial cognition, their performance was poor across most spatial cognitive tasks, particularly in constructing route and survey knowledge. The leading model, GPT-4-turbo, correctly answered fewer than one-fourth of the questions. In contrast, the Hybrid Mind approach significantly improved performance, correctly solving 70.48% of the questions.

We present DiSA, a Display-driven Spatial Analysis framework for interactive analysis of large-scale geographical vector data. DiSA calculates visualization of analysis results directly using a parallel per-pixel approach with efficient fine-grained spatial indexes. Compared with conventional object-based methods, DiSA can greatly reduce the computational complexity (from O(n) to O(log(n)) in some cases), making it less sensitive to data volumes. Experimental results verify that DiSA can provide analysis of billion-scale spatial objects in milliseconds. We demonstrate DiSA with various application scenarios including raw data exploration, spatial buffer and overlay analysis, and global cellular signal strength analysis. Users can explore 10 millions of spatial objects, adjust algorithm parameters, and always see the results in real-time on a personal computer.