GIScience 2025

A1. Shared micromobility and equity: A comparison between station-based, hybrid, and dockless models

Scarlett Ting Jin

Abstract

This study compares the equality and equity of four shared micromobility models, including station-based bikesharing, hybrid e-bike sharing, hybrid e-scooter sharing, and dockless e-scooter sharing, with a case study in Minneapolis, Minnesota. Equality assessment is made using Lorenz curves and Gini coefficients, which reveals that hybrid and dockless models are relatively more equal than the station-based model. Equity assessment is conducted using regression analysis. Results on supply equity show that station-based and hybrid models are more equitable than the dockless model, as they allocate more devices to economically disadvantaged areas. Regarding utilization equity, regression models yield mixed results: depending on the disadvantaged indicators used to identify disadvantaged communities, both the station-based and hybrid/dockless models can exhibit characteristics of equity or inequity. Overall, our findings suggest that the hybrid model shows the most promising potential for improving both the equality and equity of the spatial distribution of shared micromobility services.

A2. USGT: An unsupervised Spatial Graph Transformer framework for detecting spatial communities in human mobility data

Meihua Chen

Abstract

Identifying spatial communities from human mobility data is vital for enhancing the understanding of spatial interaction patterns and urban structures. However, spatially embedded graphs derived from movements are often sparse and noisy, posing challenges for spatial community detection. While various methods have been proposed, most rely on statistical characteristics of graphs and struggle with sparse, noisy mobility data. Recently, graph embedding techniques have shown promise in learning low-dimensional representations for community detection. Yet, existing techniques often fail to preserve modular structures and rarely incorporate spatial constraints for spatially embedded graphs. To address these challenges, we developed an unsupervised spatial graph Transformer framework for spatial community detection. To preserve the modular structure in graphs, we modified the self-attention mechanism to capture both global and local information of nodes. To incorporate spatial constraints into the model, we develop a spatial encoding scheme that explicitly encodes multi-order adjacency relationships between spatial units. To train the proposed model in an unsupervised manner, we generate multiple subgraphs according to a data masking strategy. Experiments on synthetic datasets demonstrate that the proposed method outperforms five representative methods in detecting spatial communities from movements with noise. When applied to taxi GPS trajectory data in Beijing, the spatial communities identified by the proposed method exhibit greater geographical continuity and stronger internal interactions. These communities not only enhance the understanding of travel patterns revealed by taxis but also provide valuable insights into the polycentric urban structure.

A3. A Data Cleaning Method for Identification of Outliers in Fishing Vessel Trajectories Based on Geohash Geocoding algorithm

Weifeng Zhou and Li Zhang

Abstract

In modern fisheries management, trajectory data of fishing vessels are utilized to monitor and analyze vessel activities. However, such data are frequently impacted by environmental factors, equipment malfunctions, signal loss, and human errors, leading to a significant number of outliers. These outliers not only diminish the credibility of the data but also adversely affect subsequent data mining and decision-making processes. This study presents a data cleaning method for identifying outlier points in the trajectory of a fishing vessel based on the Geohash geocoding algorithm. The method entails the following steps: obtaining and pre-processing the trajectory data of a fishing vessel; generating the corresponding Geohash codes for each ship position based on its latitude and longitude; calculating the reachable distance according to the time interval between the current point and the following points as well as their speeds; querying the neighborhood of the current point based on the reachable distance, and obtaining all the Geohash codes of the reachable areas of the fishing vessels within the time interval as the reachable range grid set of the current position; comparing the reachable range grid sets of the current position with those of its preceding and succeeding points in the fishing vessel trajectory respectively. If there is no intersection with either of them, the current ship position is identified as an outlier point and removed. The method proposed in this study can effectively identify outliers in the trajectory data, achieving efficient and effective trajectory data cleaning, and enhancing the accuracy and reliability of the data.

A4. Immersive virtual reality and spatial navigation: The impacts of ADHD

Susannah Cox

Abstract

Spatial navigation is a complex task that is dependent on several cognitive processes that can be hard to measure and replicate in traditional in-lab experiments. In comparison, immersive virtual reality proves to be a means to simulate real world contexts with enhanced ecological validity and spatial embodiment. As a result, immersive virtual reality continues to prove to be a useful tool to study spatial navigation. This research seeks to understand the impacts of executive dysfunction, present in ADHD adults, on navigational abilities and spatial learning. This study consists of two experimental tasks. The first is a wayfinding task, where participants are asked to traverse an urban virtual geographic environment and locate waypoints in a specified order according to a provided map. This task consists of ten levels that progressively get more difficult. The second task drops participants at a semi-random point in the environment they just traversed and asks them to orient themselves to the starting location; this heading orientation task tests participants’ spatial learning. Unlike traditional assessments, xR allows researchers to observe navigation behaviors dynamically, capturing how individuals with ADHD manage challenges like spatial working memory, attention, and cognitive load capacity during complex spatial tasks. This is particularly important for understanding how executive dysfunction affects real-world functioning and for tailoring environments or interventions to better support individuals with neurodivergent cognitive profiles. We hypothesize that as the wayfinding task increases in difficulty, we will see a divergence in participant performance with ADHD adults performing worse on all measures. This research will contribute to countering spatial deskilling and the development of neuro-adaptive navigational aids that support safety, efficiency, and spatial learning.

A5. Mapping Mismatch: Perceived vs. Measured Diversity in Urban Activity Spaces

Olena Holubowska, Ate Poorthuis, I-Ting Chuang, Katarzyna Sila-Nowicka

Abstract

The study of segregation in activity spaces is undergoing a renaissance, driven by the increasing availability of geospatial data from mobile phones and social media. These new data sources have facilitated high-resolution spatial analysis of individual mobility patterns, offering insights into the diversity of visitors in different urban spaces. A key motivation behind this research is to foster social cohesion by designing inclusive spaces that encourage intergroup contact. However, amidst the enthusiasm for analyzing large datasets, an important nuance may have been overlooked—perception of diversity often plays a more significant role than objective measurements in shaping urban experiences. Yet, we know very little about the relationship between perceived and measured diversity in activity spaces.

A6. Predictive Geomasking: Leveraging Machine Learning for Optimized Privacy in Point-based Geospatial Data

Pankajeshwara Sharma

Abstract

As the use of location-based services and geospatial analysis continues to grow, ensuring the privacy of individuals in shared spatial data has become more important than ever. Geomasking, a technique used to obscure precise geographic points, is commonly employed to protect privacy. However, traditional methods often rely on random processes or manual adjustments, which can be time-consuming and imprecise, requiring multiple iterations to reach the desired privacy level, which could be a barrier to masking many datasets simultaneously. Mechanisms that simplify geoprivacy application would significantly advance research and mark a key step forward in its practical adoption. In this talk, we discuss on a more efficient and intelligent solution, PrediMask, through the use of machine learning. Our approach trains models on a variety of geospatial datasets to predict the exact amount of displacement needed to achieve a specific privacy rating. By analyzing key factors such as point density, the average distance between neighboring points, and the spread of data, our model aims to accurately forecast the necessary masking distance to maintain privacy, based on Spruill’s geoprivacy measure. This predictive geomasking method aims to not only save time but ensures that privacy standards are consistently met, without compromising the quality of the data. Our research aims to demonstrate how machine learning can transform privacy protection in geospatial data by offering automated and reliable solutions for protecting sensitive information.

A7. Recharging the 15-Minute City: How e-micromobility is expanding access

Priyanka Verma, Dan Qiang, Grant McKenzie

Abstract

The concept of the 15-Minute City has emerged as an idealized framework for sustainable, human-scale urban planning. It promotes access to essential services, such as schools, parks, grocery stores, and hospitals, within 15 minutes of active travel (walk or bicycle) from a resident's home. Much of the current scholarly and policy focus has centred on walkability, with limited attention to how new mobility modes such as e-micromobility services (e-scooters and e-bikes) are reshaping this accessibility discussion. In this work, we address an important gap in existing literature by evaluating the role of shared e-micromobility in extending amenity access across five international cities: Berlin, Germany; London, UK; Paris, France; Wellington, New Zealand; and Washington DC, USA. Analyzing 24 million real-world e-micromobility trips collected between 2020 and 2024 and their relationship to OpenStreetMap points of interest (POI), we identify spatiotemporal patterns of access. The methodology involves calculating trip durations for walking and e-micromobility between trip origins/destinations and five categories of amenities (hospitals, parks, schools, grocery stores, entertainment venues). Using an Inverse Distance Weighting interpolation approach, we generate fine-resolution duration surfaces for each transport mode, city, and POI type. The accessibility surfaces are then reclassified into three temporal thresholds, under 15 minutes, 15-30 minutes, and over 30 minutes, with the goal of evaluating and comparing mode-specific accessibility. Our results reveal that e-micromobility significantly enhances 15-minute accessibility, particularly in lower-density, car-centric cities. For instance, in Wellington, e-micromobility triples the percentage of the city that is reachable within 15 minutes for school access. Washington DC sees similar improvements across all POI types, with accessibility increases ranging from 10-20% compared to walking. In contrast, European cities such as Berlin and London, which already feature high baseline walkability, report modest gains, especially for shorter trips. Importantly, we find that accessibility enhancements are not uniform across amenity types. Schools and parks benefit most from e-micromobility, while access to hospitals and entertainment venues shows greater variability depending on spatial distribution. Furthermore, improvements are most noticeable in peripheral and under-served areas, demonstrating e-micromobility's potential to address spatial inequities and fill access gaps in the urban periphery.

A8. Under Pressure: The Impact of Psychological Stress on Usability in Cartographic Interaction

Brynn Patrello

Abstract

From escaping a natural disaster to finding a dentist's office on time, people often use maps while under stress. Yet, little research has explored how stress affects users during cartographic interaction. This study addresses that gap by asking: What is the role of psychological stress in cartographic interaction? To investigate, participants first undergo a modified Trier Social Stress Test, a proven method in psychology to induce stress through social-evaluative and cognitive challenges. Then, they complete mapping tasks using maps of varying visual complexity. Their performance is observed to assess how stress influences map-based decision-making. The study mirrors real-life scenarios where users approach maps already stressed—not because of the map itself, but due to external pressures. This stress may still impact their decisions, behaviors, and interaction with the map. The research evaluates both quantitative and qualitative metrics. Quantitative measures include task completion rate, speed, and accuracy. Qualitative insights focus on user experience, perceived usability, and interaction pragmatics. Visual complexity is a key variable due to its known effects on user stress. By applying different complexity levels across stressed and non-stressed participant groups, the study isolates the role of both stress and design in user performance. Ultimately, this research emphasizes that users are not always neutral or consistent—psychological states matter. Understanding how stress affects map use can help cartographers design with empathy, creating tools that better support users in high-stress situations. It also encourages a shift in user studies toward considering the user's mental state as a key design variable, not just the interface itself.

A9. Protecting sensitive locations in urban movement data through behaviour-preserving map-matched trajectories

Maike Gatzlaff

Abstract

Movement trajectory anonymisation remains an ongoing topic as privacy requirements become stricter, and the surplus of data generated daily makes anonymisation more complex. Currently, the utility of securely anonymised trajectory data remains low for most use cases as information such as traffic characteristics or drivers behaviour get lost due to generalisation. In our solution, we aim to preserve information on movement behaviour, such as speeding or dangerous driving, while hindering the profiling of individuals. We propose building on the promising approach of interchanging user information to preserve an individual’s privacy. Yet, most well-established concepts rely on pre-defined geometries to connect one user's trajectory to another. These so-called mix-zones are often masked - meaning no movement data is available at certain locations. To address this gap, we propose a map-matching-based framework that generates synthetic trajectories by mimicking the movement characteristics of nearby users and connecting segments via shortest-path algorithms. This approach allows for bigger gaps between sub-trajectories and enhances the potential for linking them. Our approach enables splitting and reconnecting trajectories based on spatial and temporal proximity to another user’s (sub-)trajectory, whether real or artificially generated. Furthermore, sub-trajectories that could disclose sensitive personal information, such as stops at semantically important locations, can be replaced with synthetic sub-trajectories. Using map-matching algorithms, we modify trajectory points related to these stops to simulate continuous travel. The resulting synthetic trajectories reflect realistic traffic patterns by drawing on the movement behaviour of nearby users or from movement behaviour immediately before/after the sensitive stop. By interchanging user information of sub-trajectories at random road segments, a maximum number of raw trajectory points mapped to the road network is maintained - a crucial factor for high utility. Sensitive locations are protected by synthetic trajectory points, and no regions are obscured by cloaking or mix-zone geometries.

A10. A hermeneutic review of how spatial weights are operationalized in GIScience

René Westerholt, Fillipe Oliveira Feitosa

Abstract

Spatial weights are often treated as instrumental tools in spatial analysis. Yet, the ideas and assumptions shaping such weights often go unquestioned and are not always fully explicated. This abstract shares early insights into an ongoing systematic study exploring how empirical researchers frame and define spatial weights. At the center of our research is one key question: What meanings do researchers ascribe to spatial linkages when they formalize them into spatial weights? We address this question utilizing a hermeneutic approach, which enables researchers to interpret underlying positions. Drawing from a structured review of more than 16,000 articles indexed in Scopus and Web of Science, we use a filtering and clustering approach to identify a manageable group of 30 empirical studies for detailed qualitative analysis. The selection process begins by narrowing the dataset down to 638 abstracts, a number that ensures representativeness for the overall corpus. Since this number is too extensive for our intended qualitative analysis, two-step screening is then applied using an inclusion/exclusion protocol and expert agreement to settle on 30 studies. Our methodological innovation lies in combining structured corpus selection with qualitative hermeneutic inquiry, allowing for a deeper look into the ideas and assumptions that guide methodological decisions involving spatial weights. Our so far obtained early findings already show repeating patterns in how scholars operationalize spatial linkages through their chosen methods. This work opens space for broader conversations about what formal spatial concepts like spatial weights actually mean in practice. By uncovering and interpreting the assumptions embedded in methodological choices, we aim to promote more reflexive spatial modeling practices in addition to technical rigor. Our work contributes to growing efforts to include deeper interpretive practices within spatial analysis and foster the community's self-awareness and critical engagement.

A11. MatLoc-NeRF: Scalable and Accurate Neural Geolocalization via Selective NeRF Feature Matching

Changhao Chen, Huaiji Zhou, Bing Wang

Abstract

Accurate large-scale geolocalization and 3D scene reconstruction are fundamental problems in geographic information science. Neural implicit representations, such as Neural Radiance Fields (NeRF), have recently enabled photorealistic 3D modeling. However, visual localization using NeRF remains limited by poor scalability and high computational cost, especially in large-scale environments. We propose MatLoc-NeRF, a novel matching-based localization framework that efficiently estimates camera poses using selected NeRF features. At its core, MatLoc-NeRF introduces a learnable feature selection module that identifies the most informative NeRF features for matching with query images. This avoids reliance on dense scene descriptors or exhaustive comparisons, substantially improving localization accuracy and efficiency. To support large-scale deployment, MatLoc-NeRF incorporates a pose-aware scene partitioning strategy. This method activates only the relevant NeRF sub-block based on the predicted query pose, enabling localized feature extraction and reducing memory and computation overhead. Additionally, we integrate a scene segmentation module and a lightweight place recognition network to deliver fast and robust coarse pose estimates, effectively narrowing down candidate regions and accelerating the overall localization process. We evaluate MatLoc-NeRF on two public large-scale benchmarks and compare it against state-of-the-art NeRF-based localization methods, including iNeRF and Curriculum iNeRF. Experimental results show that MatLoc-NeRF achieves superior accuracy and efficiency, demonstrating its robustness across diverse scenes and lighting conditions. Unlike prior methods, it scales gracefully to complex environments without compromising localization quality.

A12. Participatory GIS: Uncovering Hidden Data for a Social Digital Twin

Batel Yossef Ravid and Hadas Chassidim

Abstract

This research presents an applied, multidisciplinary collaboration between urban design and human-computer interaction, focusing on the integration of civic knowledge into digital planning tools. It builds on the framework of the Social Digital Twin and incorporates Participatory GIS (PGIS) methods to address the spatial invisibility of unrecognized Bedouin villages in Israel's Negev region. Unrecognized villages are home to tens of thousands of residents, yet remain absent from official government maps and databases. Lacking basic services, infrastructure, and formal planning recognition, these communities operate largely outside the formal spatial system. This research asks how Digital Twin technology — typically used for buildings and infrastructure — can be redefined as a social tool, shaped by local knowledge and lived experience. The case study focuses on the village of 'Al Naam'. In the first phase, we designed and implemented a digital survey using ESRI Survey123, collecting spatial data from residents. Participants were asked to describe locations, mark routes, and give directions using local landmarks and terminology. In the second phase, we began developing a PGIS tool based on this data, enabling residents to map their environment collaboratively. Our findings reveal that residents possess rich spatial knowledge based on culturally embedded reference points such as “the white mountain,” “the green caravan,” and local mosques. These verbal maps fill the gaps left by the absence of formal geospatial data. By transforming these narratives into spatial layers, we demonstrate how community-driven data collection can inform planning, enhance visibility, and support more inclusive governance. The project offers a model for using digital twins not only as technical systems but as tools for civic empowerment in marginalized areas.

A13. The BioWhere Gazetteer: A Culturally Enriched Place Name Knowledge Base for New Zealand

Aneesha Fernando, Kristin Stock, Hone Waengarangi Morris, Kalana Wijegunarathna, Pragyan Paramita Das, Jonathan Procter, Surangika Ranathunga, Raj Prasanna, Christopher B. Jones

Abstract

Gazetteers link place names to their geographic footprints and associated contextual knowledge. While traditionally used in cartography and geography, modern gazetteers have evolved into extensive sources of geographic intelligence. However, within Aotearoa New Zealand's rich cultural and historical landscape, existing gazetteers often lack the capacity to integrate conventional geographic datasets with indigenous, narrative-rich place name resources. We propose a culturally enriched gazetteer model for Aotearoa New Zealand, extending the Alexandria Digital Library (ADL) Gazetteer Content Standard (GCS) to integrate geospatial and cultural dimensions. The proposed model consolidates data from authoritative datasets such as the Land Information New Zealand (LINZ) gazetteer and volunteered geographic information (VGI) platforms like GeoNames and OpenStreetMap. A key innovation is the inclusion of Māori cultural knowledge through whakapapa, the lineage, stories, and histories associated with place names, enabled by extending the etymology component of the ADL model. Structured as a relational database, the gazetteer supports multiple names, feature types, and geometries per place, while maintaining temporal information and source provenance. It accommodates bilingual names in Māori and English, as well as alternative spellings, abbreviations, and variants, reflecting the linguistic diversity of New Zealand's toponymy. Another distinctive feature of this gazetteer is the integration of georeferenced locality descriptions derived from biodiversity collection records. These often refer to unnamed or informally described places, and their inclusion facilitates georeferencing tasks by centralising both named and descriptive spatial data. The database is populated via an Extract, Transform, Load (ETL) pipeline in PostgreSQL, using PostGIS for spatial data maintenance. We also plan to incorporate self-learning capabilities to infer coordinates for previously unrecorded place names. Upon completion, this gazetteer aims to serve as a repository of New Zealand's geographic features and cultural heritage, encompassing more locations than existing gazetteers for the region, and supporting GIS, ecology, digital humanities, and indigenous knowledge systems.

A14. Common errors in geo-ontology engineering

Nenad Radosevic, Matt Duckham, Mohammad Kazemi Beydokhti, Yang Chen, Katherine Williams, Allison Kealy, Yaguang Tao, and Qian Chayn Sun

Abstract

Ontology engineering—the systematic design of machine-readable vocabularies or ontologies—has a long history in GI science. The promise of such “geo-ontologies” is to provide solutions to some of the most long-standing challenges in the field, such as semantic interoperability; tracking provenance and metadata; and capturing enormous richness of geospatial data and phenomena with simple and consistent representations. Recent rapid advances in large language models (LLMs) have provided renewed impetus for geo-ontology engineering, because knowledge-based representation and reasoning offers structural and logical characteristics that complement the strengths of LLMs for pattern-matching unstructured data. However, geo-ontology engineering is fraught with pitfalls and snares. Errors are frequent, especially for less-experienced ontology engineers. This paper argues that three key errors are most commonly encountered, including in papers in the domain of geographic information science. The paper explores in more detail these three types errors, illustrated through the example of engineering a geo-ontology for update and revision of foundation spatial data (SMURF, the spatial management, update, and revision framework) within a recent project in partnership with state-based mapping agency (the “Dynamic Vicmap” project). In brief, the three key errors relate to ontology reuse, clarity, and consistency. Reuse is a core principle of ontology engineering, which enables ontologies to be more modular and standardized, and data to be more easily linked together. Clarity relates to avoiding errors in definition, such as misclassifying instances as classes. Consistency relates to avoiding internal contradictions, and most concerns the domain-specific judgements embodied in a geo-ontology, such as deciding whether an entity is best represented as an instance or attribute. Examples of each type of error are illustrated with reference to the SMURF geo-ontology, engineered to represent and link multiple statewide hydrology, property, and flood event data sets reusing established ontology standards, including GeoSPARQL, PROV-O and DCAT.

A15. A deep learning urban development type classification system: an application to millions of Australian property parcels

John Duncan, Bryan Boruff, Warin Chotirosniramit, Sally Thompson, Joe Hurley, Marco Amati, Chayn Sun

Abstract

Monitoring parcel-level urban development is necessary to characterise how cities are changing to accommodate growing populations. Multi-temporal datasets of parcel-level development type are required to understand how development affects socio-economic, environmental and climatic outcomes and to guide planning policy and analyses that promote sustainable and liveable cities. However, generating consistent classifications of parcel-level development across time and jurisdictions remains challenging. To address this gap, a multi-modal deep learning model is presented and validated that converts urban land cover and built form data into classifications of a parcel’s development type. The classes under consideration are single unit, multi-unit, low-rise apartments, cluster dwellings, high-rise apartments, other urban, urban vegetation, rural and bare earth. This system is based on the SWIN-V2 Transformer architecture but is adapted to also consider surface cover, built form and spatial metrics encoded as attribute tables associated with each parcel. This model is trained and tested using Nearmap AI data covering five Australian cities and multiple time-points (Perth, Sydney, Melbourne, Brisbane and Adelaide). A ground truth dataset of over 70,000 parcels were labelled across Perth and a Perth-specific model was trained and evaluated. The Perth model was adapted to other Australian cities using a transfer learning workflow. A fine-tuning training dataset was generated by detecting parcels in other cities that were out-of-distribution from the Perth training dataset. The Perth model was fine-tuned using LoRA with a sample of out-of-distribution samples from other cities and was evaluated on representative city-specific test datasets. This model was applied to every property parcel in each of the five cities in 2019 and 2023 to generate complete city-wide maps of parcel-level development type. These maps are used to characterise the nature, extent and spatial patterns of different types of development occurring in Australian cities.

A16. Coordinating Future Economic Growth and Heatwave Exposure Mitigation in Global Coastal Cities: A GIScience Framework Integrating Machine Learning, Spatial Modeling, and Climate Projections for Sustainable Urban Development

Yuwei Wang and Na Zhao

Abstract

The growing interdependence between urban economic development and heatwave risks demands new analytical frameworks, particularly for coastal cities facing dual pressures of climate change and rapid urbanization. We propose a novel GIScience framework integrating Heat Index (HI) calibrated with local humidity patterns, machine learning, and spatial-statistical modeling to quantify bidirectional relationships between economic trajectories and population exposure to heatwaves across global coastal cities by 2100. Our approach uniquely combines CMIP6 climate projections, high-resolution urban expansion projections, and SSPs socioeconomic scenarios through a three-stage analytical framework: (1) dynamic calculation of humidity-adjusted heatwave exposure using spatially varying HI thresholds, (2) Mann-Kendall trend analysis of long-term heatwave intensification, along with correlation and ANOVA analyses to preliminarily explore economic-heatwave exposure relationships, (3) machine learning-driven identification of non-linear economic-exposure linkages via random forest regression with SHAP interpretability, and generalized nonlinear modeling to capture city-specific response curves between GDP growth and exposure intensification. Results demonstrate significant spatial heterogeneity in economic-heatwave exposure interactions, and coastal cities under current low- and high-income levels will face relatively more severe and rapidly intensifying heatwave exposure. Three distinct risk categories emerge through cluster analysis: Development Imbalance Risk, Economic Lag Risk, and Potential Risk Alert, leading to targeted strategies for sustainable urban development in the face of heatwave risks. This research advances GIScience through three principal contributions: (1) an open-source HI processing toolkit enabling humidity-aware exposure mapping in coastal microclimates, (2) demonstration of machine learning's capability to decode complex economic-climate interactions beyond traditional linear assumptions, and (3) a transferable risk classification framework and modeling workflow bridging climate projections with economic decision-making, directly supporting SDG 11 (Sustainable Cities) and 13 (Climate Action). The findings underscore the necessity of context-specific strategies for harmonizing economic development with heat resilience in coastal urban systems, offering significant advancements for sustainable urban planning and climate action.

A17. Transparent and reusable semantic enrichment of place name data in Australia

Nayomi Geethanjali Ranamuka, Matt Duckham, Dana McKay, Nenad Radosevic, Alexis Horde Vo, Yaguang Tao, Prawal Lohani, Amina Hossain, Mohammad Kazemi Beydokhti

Abstract

Semantic enrichment is the process of augmenting data with additional contextual information that helps disambiguate its meaning. In practice, the semantic data enrichment process typically involves uplifting traditional data sources, such as tabular or object-relational, with connections to standard published ontologies. Conventional semantic data enrichment approaches are often based on ad hoc scripting and coding pipelines. Consequently, these bespoke approaches result in substantial development and maintenance costs, time, and manual effort. They may also undermine the quality and scalability of the enrichment applications, give rise to significant duplication, and negatively impact their long-term usability. To address these challenges, our approach explores the use of declarative mapping languages to enrich spatial data. The approach is founded on separating the semantic mapping logic from the mechanism of performing the mapping, two aspects that are conflated by scripting and coding pipelines. Separating these aspects is expected to lead to the development of more transparent and reusable semantic data enrichment techniques. However, to date, most such explorations have focused solely on non-spatial data. We employed Australian place name data as a case study to evaluate the effectiveness of semantic spatial data enrichment using declarative mapping. RDF Mapping Language (RML) was used to semantically enrich place names across each state in Australia, resulting in a national placename knowledge graph. Place names were enriched with metadata and contextual information using standard ontologies, including Place Names Ontology, Data Catalog Vocabulary (DCAT), Data Quality Vocabulary (DQV), and GeoSPARQL Ontology. The proposed transparent and reusable approach enhances the quality of insights derived from the knowledge graph while minimizing the overall cost of the transformation process. It results in a more efficient and effective semantic enrichment process. The place name knowledge graph can quickly adapt to new modifications with minimal resources and ensure long-term usability and maintainability of the application.

A18. Assessing impact of community mapping

Diego Pajarito Grajales and Anne Schauss

Abstract

Participatory mapping has gained visibility within geographic information sciences in recent decades. Alongside community engagement, it can complement traditional data production to better capture local knowledge. Geographic data production has advanced in line with the underpinning technologies in the same period. However, those participatory exercises do not always positively impact the intended populations. This paper compiles and analyses a series of participatory mapping cases presented in the “mapping with communities” workshops organised at the GIScience 2023 and AGILE 2024 conferences. The authors use five categories to understand the case studies contributions: i) What and Why, or the purpose and elements mapped; ii) Who, or the stakeholders involved in the mapping activities; iii) Where, or the location of the mapping activities; iv) How, or the methods applied and v) Impact, or the personal and territorial changes that the mapping initiated. To assess the last category, the team screened different general impact assessment frameworks. An adapted framework helps assess the impact of all the case studies in terms of Higher-level, Instrumental, Conceptual, and Low-Level. Conversely, the research question: How can participatory mapping drive impactful changes or benefits for society? drives the discussion. By compiling, discussing and publishing this impact assessment, the authors acknowledge contributions from workshop participants to enable advocacy of vulnerable communities' livelihoods. The identified research focus of the studies in countries of the global south indicates the particular data requirements of those regions and the researchers' interests in working with historically marginalised communities. Moreover, the role of external parties and researchers initiating most of the mapping procedures highlights that these are still complex processes that require stronger engagement, especially from government agencies. Among other outcomes, the analysis claims that in order to strengthen the credibility of participatory methods, better impact assessment frameworks are needed while this first attempt served to identify key impact dimensions. This assessment highlights the importance of embedding impact throughout the entire process. While high-level impact (i.e., societal change) is desirable, it often requires additional time to materialise.

A19. Evaluating Diversity of Staying People by Using High-resolution Gridded Population

Ginta Iida and Wataru Nakanishi

Abstract

Many urban planning master plans by local governments in Japan consider “interactions among diverse individuals” as a key factor in promoting urban vibrancy. Such interactions are more likely to occur when people from diverse backgrounds stay within the same urban spaces. However, while much research had addressed these interactions themselves, the quantitative evaluation of the diversity of the staying population in urban spaces has received limited attention. To support effective urban measures such as organizing events in open spaces, the diversity evaluation therein in advance is important. This study proposes a method for such diversity evaluation at high spatial resolution. Although the diversity in individual's values and knowledge is difficult to observe directly, this study assumes that such diversity can be inferred from characteristics of their residential areas, for which data is often available. Specifically, we prepare 18 variables for each 1-kilometer grid such as age composition, family composition, employment status, and education level from national census data. These are aggregated into one-dimensional values via principal component analysis. We also utilize 500-meter grid-based population data derived from mobile phone location records, which are linked to users’ places of residence in 1-kilometer grid. Consequently, for each 500-meter grid, we can determine the composition of the staying population based on residential characteristics. The diversity of the staying population is then calculated using differential entropy, a metric from information theory. The results for two prefectures in Japan showed that high levels of diversity were found not only in central urban areas, large hospitals, and suburban shopping malls, but also in certain grids without any notable facilities. In these cases, the areas attracted people from residential areas with diverse characteristics. Furthermore, we identified several interesting cases where the level of diversity differed significantly among grids with similar facilities.

A20. Watching the wetlands: a visibility analysis of the Ruamahanga basin, Aotearoa New Zealand using human movement data

Mairead de Roiste

Abstract

Wetlands play an important role in the ecosystem. Wetland loss has been estimated at over 90% in Aotearoa New Zealand. Restoring these important environments can bring significant benefits such as flood protection, runoff mitigation, and support for wildlife. Identifying restoration sites with high public visibility may enhance awareness and appreciation of these landscapes. This paper explores the use of viewshed analysis to calculate visibility across the Ruamahanga basin in the Wairarapa to better identify areas where former wetlands can be restored. To enhance this method, we combine visibility analysis with mobile phone tracking data to determine the weighting associated with different locations, reflecting human movement patterns. For studies where such data is unavailable, we developed a simulation model using spatial data from the Ruamahanga and two other similar-sized catchments to predict areas of high human movement and their visibility weighting. This scalable approach can guide decision-making for wetland restoration efforts, ensuring ecological gains are paired with improved public awareness. Such a method will be of interest to planners, conservationists, and local councils seeking to maximize the social and ecological benefits of wetland restoration projects. By identifying locations where restored wetlands are both ecologically viable and highly visible, this approach could help increase public engagement and support for conservation initiatives.

A21. Predicting Forest Fire Refugia Using Machine Learning: The Role of Topography and Microclimatic Variables

Sven Christ, Helen de Klerk

Abstract

Forest fire refugia are areas within fire-prone landscapes that remain fire-free or experience lower fire frequency and severity. These refugia are crucial for biodiversity, supporting species, maintaining mature vegetation, and aiding post-fire recovery. They enhance forest resilience by preserving genetic diversity and facilitating regeneration. Understanding how topography influences fire behaviour is key to identifying and conserving refugia, informing forest management to protect them from logging and disturbances. Machine learning models using wind patterns and altitude can effectively predict refugia in localized areas, aiding conservation efforts and promoting ecosystem resilience in fire-prone regions. Forest fire refugia can be accurately predicted using aspect, surface wind direction and speed (derived from computational fluid dynamics), topographic roughness, and temperature in machine learning algorithms (Random Forest, XGBoost; 2 ensembles models) and K-Nearest Neighbour (all run with and without ADASYN over-sampling). Six iterations were run per algorithm to assess the impact of leaving variables out. Among these variables, aspect is the most influential across both feature importance (Random Forest) and feature gain (XGBoost), as it aligns fire refugia with the leeward slopes of prevailing fire winds. Surface wind speed and direction, and global irradiation are also key predictors, with significant drops in model accuracy when these features are excluded. Temperature and topographic roughness show context-dependent importance. Temperature was significant in XGBoost but diminished after ADASYN oversampling, while topographic roughness increased in importance when elevation was excluded. Elevation did not significantly enhance model performance, and its exclusion had minimal impact on predictive accuracy. Ensemble models consistently produced the most accurate results, although accuracy metrics across all experiments where high and averaged 0.96 (±0.2), indicating robust predictive performance. These findings highlight the importance of topographic and microclimatic variables in fire refugia prediction, with machine learning providing reliable forecasting frameworks.

A22. Integrating Multi-Source Population Data for Spatiotemporal Mapping of Human Mobility in Urban Areas

Toshihiro Osaragi

Abstract

Observing and analyzing human movement on a large spatial scale is crucial for managing crowding, guiding emergency responses, and supporting commercial and tourism planning. In dense urban areas, where population distribution fluctuates rapidly due to public transportation, traditional static datasets provide limited insights. This necessitates methods for dynamically mapping population distributions in real time. However, existing datasets on static and mobile populations vary in temporal resolution, spatial coverage, and sampling fraction, imposing constraints on research and analysis. This study proposes a method for constructing a spatiotemporal data structure of moving individuals by integrating multiple population datasets while leveraging their strengths. The primary dataset, derived from mobile phone base station coverage, estimates total population distribution, including transient and static individuals. A secondary GPS-based dataset, transmitted every five minutes, provides detailed movement patterns despite a lower sampling rate and is used to estimate inter-grid-cell movement fractions. A maximum likelihood estimator quantifies grid-cell entries and exits, with validation against precisely measured flow data confirming the approach’s accuracy. To further assess the method's applicability, we analyze multiple urban regions, quantifying transient population numbers and their movement patterns over time. This approach revealed notable differences in movement dynamics between weekdays and weekends, as well as distinct travel patterns in commercial and office districts that appeared similar in static datasets. The proposed methodology allows for the identification of transient population numbers and movement directions at any given time across large spatial scales. By constructing comprehensive spatiotemporal datasets for both static and transient urban populations, this study introduces new analytical perspectives for understanding urban dynamics. Future research will focus on developing models to evaluate the impact of large-scale public events and natural disasters on population movement. These models will offer valuable insights for crowd management, risk mitigation, emergency response planning, and evacuation strategies, ultimately enhancing urban resilience.

A23. Leveraging Topological Data Analysis for disaster preparation and response in GIScience

Jim Thatcher, Chad Giusti, Connor Progin, Courtney Thatcher, Carolyn Fish, David Retchless

Abstract

Disasters often cause road network degradation resulting in reduced traffic capacity, which directly impacts emergency response. In GIScience, it is common to model road networks as directed graphs, which provide a formal framework for analyzing traffic flow. These analyses tend to focus on computing graph-based metrics as a series of edges are removed from static graphs (Knoop et al., 2012, Loretti et al., 2022). For static graphs, there are alternative algorithms that can determine the maximum flow between intersections across the graph in polynomial time. For infrastructure networks which contain a very large number of nodes and edges, these algorithms become computationally expensive. Further, in dynamic models, where edges are constantly added and removed, recomputing the maximum flow value becomes highly inefficient. We propose an algorithm to efficiently analyze and visualize road network capacity under changing conditions using techniques from Topological Data Analysis. The algorithm can also be used for determining the maximum flow value between two geographic regions (i.e. from one collection of nodes to another). It starts by computing a maximum flow on the original graph. If the new edge goes from node u to node v, the new maximum flow value is determined by solving two smaller independent maximum flow problems: from the source to u and from v to the sink. The increase in maximum flow value is equal to the smaller value of these two subproblems. These computations are performed in parallel with information recycled across iterations to reduce redundancy. This algorithm results in a maximum flow on the graph, ensuring immediate iterability and analysis as edges are added or removed. Future work will improve the algorithm for dynamic infrastructure models with a long-term goal of providing a comprehensive tool for understanding the stability of road networks in geographic regions susceptible to degradation.

A24. Foundation Spectral Model-Based Detection of Parasitic Plants in Urban Environments: A Case Study of Struthanthus interruptus

Paola Andrea Mejia-Zuluaga, Hugo Carlos Martinez, Juan Carlos Valdiviezo-N, León Dozal

Abstract

Struthanthus interruptus, a parasitic plant, poses a rising threat to Mexico City's urban areas. This hemiparasite infests a wide variety of native tree species, lowering their health, hastening mortality, and eventually diminishing the ecosystem services they provide. Traditional detection is based on costly fieldwork and expert visual recognition, which limits the frequency and scope of monitoring activities. Although UAV-based multispectral photography has made data collection more effective, finding mistletoe species, particularly those that visibly blend in with their hosts, remains a substantial issue due to their spectral and morphological similarities. In this study, we explore how foundation models trained on spectral data can detect S. interruptus in drone-acquired multispectral imagery. We specifically assess the performance of a spectral transformer architecture (SpectralGPT) that has been pre-trained on several spectral datasets in order to encode image patches and detect parasitic signals. Unlike traditional deep learning algorithms that require huge labeled datasets, foundation models can generalize from small amounts of annotated data using spectral priors and transfer learning. We trained and evaluated our technique on photos from urban parks with proven infestations, utilizing expert-curated segmentation masks for validation. To achieve uniform spectral representation across samples, the picture data was preprocessed using a comprehensive pipeline that included radiometric correction, geometric alignment, and band coregistration. Preliminary findings show that spectral embeddings derived from foundation models capture significant differences between infested and non-infested canopy patches, even under changing phenological conditions. Our ongoing study attempts to improve the classification method by including few-shot learning and interpretability tools to understand the model's decision criteria. This study advances the field of remote sensing and ecological monitoring by proving the viability of using foundation models for species-level plant detection in urban environments. It brings up new possibilities for low-cost, high-precision monitoring of parasite infestations in tree canopies. This work contributes to the growing body of research using foundation models to solve geographic ecological challenges, with a focus on species detection under high spectral similarity.

A25. Jointly spatial-temporal representation learning for individual trajectories

Fei Huang, Jianrong Lv, Yang Yue

Abstract

Individual trajectories, capturing significant human-environment interactions across space and time, serve as vital inputs for geospatial foundation models (GeoFMs). However, existing attempts at learning trajectory representations often encoded trajectory spatial temporal relationships implicitly, which poses challenges in learning and representing spatiotemporal patterns accurately. Therefore, this paper proposes a joint spatial- temporal graph representation learning method (ST-GraphRL) to formalize structurally-explicit while learnable spatial-temporal dependencies into trajectory representations. The proposed ST-GraphRL consists of three compositions: (i) a weighted directed spatial-temporal graph to explicitly construct mobility interactions over space and time dimensions; (ii) a two-stage joint encoder (i.e., decoupling and fusion), to learn entangled spatial- temporal dependencies by independently decomposing and jointly aggregating features in space and time; (iii) a decoder guides ST-GraphRL to learn mobility regularities and randomness by simulating the spatial-temporal joint distributions of trajectories. Tested on three real-world human mobility datasets, the proposed ST- GraphRL outperformed all the baseline models in predicting movements' spatial-temporal distributions and preserving trajectory similarity with high spatial-temporal correlations. Furthermore, analyzing spatial-temporal features in latent space, it affirms that the ST-GraphRL can effectively capture underlying mobility patterns. The results may also provide insights into representation learnings of other geospatial data to achieve general- purpose data representations, promoting the progress of GeoFMs.

A26. Understanding Uncertainty in Informally Sourced GIS Data: A Case Study of Historical POIs in China

Jiahua Chen

Abstract

Informally produced GIS datasets are increasingly prevalent in China, providing critical geographic information through online platforms, social media, and e-commerce websites. These sources fill critical gaps left by official data platforms and mainstream map service providers like Baidu or Amap, which frequently do not support granular geographic or historical data queries. Despite their widespread use, these datasets frequently lack adequate metadata and documentation, raising significant concerns about data quality, uncertainty, and provenance. This research systematically investigates the production, use, and uncertainties of these informal GIS datasets, with a particular focus on historical points-of-interest (POIs) in China. Through preliminary interviews and comprehensive surveys, I first identify who creates these datasets, their processing methods, and the motivations driving user reliance on such informal sources. To quantify spatial uncertainties, I collect historical POI datasets from multiple channels, including open data portals, social media, e-commerce websites and mapping service API. Historical street-view imagery is used to ground-truth the positional and attribute accuracy of the sampled POIs. Additionally, drawing methodological parallels with bias-correction techniques developed for volunteered geographic information, this study integrates neighborhood-level variables to systematically characterize and mitigate spatial biases. Ultimately, this research advocates for recognizing informal GIS datasets as valuable sources. By systematically understanding their uncertainties and provenance, GIScience can better harness these data for accurate, transparent, and equitable spatial decision-making.

A27. Modeling Spatial-Social Heterogeneous Graphs: A Framework Emphasizing Metapath-Guided Representation Learning

Qian Cao, Gengchen Mai, Xiaobai Angela Yao, Christopher C. Whalen

Abstract

Human mobility is influenced by both spatial preferences and social interactions. Spatial-social networks provide a unified framework to integrate these dimensions, capturing how social ties influence movement patterns across space. However, most existing work in this area focuses on social network analysis and visualization, while rarely leveraging deep learning (DL) for predictive modeling. To fill this gap, we propose a framework that models the human-place interaction with a spatial-social heterogeneous graph for human movement prediction. Based on Call Detail Records of 280 volunteers in Kampala, Uganda (May 2019-May 2022), we construct a heterogeneous graph with two types of nodes (Human and Place) and three types of edges: (i) Visit links connecting humans to visited places; (ii) Move-to links between sequentially visited places, capturing mobility patterns; and (iii) Precede links connect humans who visit the same place within a time window, with edge direction indicating visit order, capturing both co-presence and temporal sequence. Based on the constructed graph, we perform link prediction to discover the potential human-place, place-place, and human-human relations. We evaluate three types of graph representation learning models: a metapath-based model (Metapath2Vec), a schema-based model (HeteroGraphSAGE), and a contrastive model combining both views (HeCo). Results show that Metapath2Vec consistently outperforms the others, achieving AUC scores of 0.894 (Human–Human), 0.986 (Human–Place), and 0.936 (Place–Place). While schema-based models perform moderately on homogeneous edges, they underperform on heterogeneous links. Contrastive models fail to generalize to homogeneous edges and offer only intermediate performance on heterogeneous ones. Our findings emphasize that domain-guided metapath design significantly improves model performance in spatial-social settings, whereas schema-level and contrastive learning models may struggle with graph heterogeneity. This research contributes a reproducible framework for spatial-social graph construction and provides empirical insights into selecting appropriate DL models for spatial-social graph inference tasks such as mobility prediction and infectious disease transmission risk modeling.

A28. Cause and effect - Boat traffic and declining water quality in Lake Wörthersee

Franziska Hübl and Philipp Berglez

Abstract

Carinthia, Austria, a popular holiday destination known for its clear water is currently facing a weakening of lake ecosystems, which is particularly evident in the bioindicator aquatic plants (macrophytes). Studies from 2021 show a significant decline in underwater plants and reeds, particularly in Lake Wörthersee, whose ecological status has dropped from “good” to “moderate” according to the European Union's Water Framework Directive. Boat waves are thought to be one of the main causes, along with factors such as intensive tourist use and climate change. The aim of the FFG-funded project WAMOS (Wave MOnitoring System based on Global Navigation Satellite System aided Inertial Navigation System (GNSS/INS) integration, project number FO999900575) is now to answer the question of whether and to what extent boat traffic influences macrophyte vegetation. To investigate interactions between wave dynamics, sediment turbulence, and macrophytes in shore zones, we are developing a novel interdisciplinary monitoring system. A key innovation is the design and development of measuring buoys to record and quantify the incoming wave energy in near real-time. Wave heights are derived from inertial data recorded with low-cost micro-electro-mechanical system (MEMS) inertial measurement units (IMUs) and precise point positioning (PPP) positions based on correction data provided by the Galileo High Accuracy Service. A dedicated filtering algorithm was developed to distinguish boat-induced waves from natural wave patterns. Initial field deployments at Lake Wörthersee demonstrate the system's ability to capture fine-scale wave events relevant to ecological processes. We present the measurement setup, the methodology and preliminary results linking wave energy to biological indicators. This approach offers a cost-effective and scalable solution for monitoring wave impacts in lake ecosystems. By the end of the project in October 2025, findings will support evidence-based decision-making and environmental management at the national level.

A29. Modelling Social Mobility Networks During Natural Disasters: A Mobile Phone Data Approach

Minh Kieu, Alexis Comber, Quang Thanh Bui, Nicolas Malleson

Abstract

Understanding human mobility patterns during disasters is critical for effective emergency response and resilience planning. While traditional mobility models provide aggregated movement insights, there remains a significant research gap in analysing the complex interdependencies between individuals and point-of-interests (POIs) within social-spatial networks during crisis events. This research proposes a novel social mobility network framework leveraging anonymised mobile phone data to model mobility patterns during natural disasters, considering co-visitation. By conceptualising mobility as a dynamic social phenomenon rather than merely spatial displacement, our approach captures emergent structures that form when individuals converge at critical locations during emergencies, revealing previously unexamined mobility dependencies that influence both individual decision-making and system-wide resilience. Our methodology employs graph-theoretical techniques to analyse mobile phone datasets from three recent disaster events. We develop a projected network model where POIs are connected through shared visitation patterns, quantifying dependencies based on co-visitation frequency and duration. The mathematical framework utilises spatial-temporal point process models—specifically adapted Hawkes processes—to capture cascade effects in human mobility during disasters, modelling how visitation to one location influences subsequent movements elsewhere. The findings demonstrate that social mobility networks exhibit distinct characteristics during disasters, with critical POIs emerging through their functional role in facilitating social connections during crises. Our approach identifies vulnerable network components through novel centrality measures specifically designed for co-visitation networks, capturing both direct dependencies and emergent relationships that prove crucial for system resilience. This research contributes a transformative framework that integrates individual behaviour, social connections and spatial dependencies, enabling more accurate prediction of population movements during emergencies and identifying overlooked vulnerabilities in transport systems. These insights provide emergency managers with evidence-based strategies to enhance disaster preparedness by strengthening critical nodes in social mobility networks.

A30. Tackling the Ambiguity of Stop Detection in Human Mobility Data: A Model-Agnostic Evaluation Framework

Kamil Smolak, Katarzyna Sila-Nowicka

Abstract

Human mobility trajectories hold immense potential for understanding various phenomena, from improving transportation networks to modelling infectious disease spread. This potential comes from the ability to extract insights from individual movement trajectories, making high-quality mobility data essential. A critical first step in ensuring such quality is data pre-processing, which transforms raw tracking records into meaningful representations of human movement for analysis. As human mobility research expands, calls for methodological standardisation grow. Without clear benchmarks and common practices, the results of different studies can be difficult to compare or replicate. Stop detection is a crucial phase in the preprocessing pipeline, identifying locations where individuals remain stationary and extracting valuable information from raw trajectories. Numerous stop-detection algorithms have been proposed, each with the aim of handling noise and capturing only the most relevant locations. However, these methods are difficult to compare directly, especially without ground truth, as mobility patterns vary widely between individuals and environments. Different applications can define a ‘correct’ stop differently, making it difficult to select the best approach. This variability and ambiguity mean that no single method can be universally optimal. To address these foundational challenges, we propose a model-agnostic framework to evaluate stop-detection methods using interpretable metrics. Specifically, stop-detection methods can produce either predictable and simplified trajectories or, conversely, rich and chaotic ones. By balancing individual-level predictability with trajectory complexity, our framework identifies a set of Pareto-optimal solutions, allowing analysts to tailor stop-detection algorithms to their needs. Furthermore, using a manually labelled dataset of more than 1,400 stops from New Zealand, we demonstrate that Pareto-optimal solutions are also more closely aligned with ground truth, suggesting that human trajectories naturally balance richness and predictability. Our framework has the potential to enhance mobility data processing pipelines by offering an independent mechanism to evaluate and tune stop-detection performance.

A31. STARS: A novel gap-filling method for SDGSAT-1 nighttime light imagery using spatiotemporal and spectral synergy

Congxiao Wang, Bailang Yu, Wei Xu, Zuoqi Chen

Abstract

The Sustainable Development Goals Satellite 1 (SDGSAT-1), equipped with the Glimmer Imager (GLI), provides high-resolution nighttime light (NTL) data across multiple spectral bands. Thus, it can notably monitor human dynamics and light pollution with enhanced spectral and spatial resolution. However, cloud cover and low-quality observations often contaminate the SDGSAT-1 GLI NTL data, limiting its effectiveness. This challenge is addressed by the development of a novel method, namely the SpatioTemporal And spectRal gap-filling method for Sdgsat-1 (STARS) GLI NTL images, which combines spatiotemporal and spectral information to generate cloud-free NTL images with satisfactory pixel brightness and continuity. STARS is the first method to effectively address gap-filling in multiband NTL data using RGB spectral information, even with irregular time intervals and limited image inputs. Compared with traditional methods such as the temporal gap-filling method and the mean-weighted gap-filling method, the Cloud Removing bY Synergizing spatioTemporAL information (CRYSTAL) method, and the spatial and temporal adaptive reflectance fusion model (STARFM), which do not specifically account for the differences in light source variations in multi-band NTL data, STARS demonstrates superior performance (higher R-squared (R2) and lower root-mean-square error (RMSE)) in simulations across seven global cities, demonstrating its effectiveness in filling cloud-induced gaps in multi-band NTL data. On average, STARS achieves R2 values for the gap-filling results compared to the actual values of 0.79, 0.78, and 0.70 in the RGB bands, respectively. The cloud-free images produced by STARS extend the time series of the SDGSAT-1 GLI NTL data, supporting multitemporal quantitative analysis. In cloudy regions like Tianjin, China, STARS effectively captures dynamic changes in NTL before and after the Spring Festival, closely matching human activity patterns from Baidu Maps, both spatially and temporally. Overall, STARS offers an innovative and effective approach for gap-filling multiband NTL data, with potential applications in similar datasets.

A32. Developing a walking accessibility index in urban areas of New Zealand: An Enhanced Two-Step Floating Catchment Area model approach

Jessie Colbert, Ed Randal, Conal Smith, Tom Stewart, Daniel J Exeter

Abstract

Good access to commonly utilised community resources is key when implementing urban regeneration, and the uptake of active transport as a means of accessing these resources. Walkability in an urban area can impact on the social, economic, and environmental wellbeing of the area and the people who reside within it. We developed an area-based urban walking accessibility index to measure walking access to commonly used, publicly accessible community facilities, services and amenities, with nationwide coverage of New Zealand. We obtained xy location data for 39 different amenities, such as Hospitals, Supermarkets, Schools, and Libraries. We calculated accessibility scores for each type of point location using the Enhanced Two-Step Floating Catchment Area method, accounting for distance decay, with scores output at the Statistical Area 1 2018 (SA1) spatial units as ‘neighbourhoods’. Travel times for each origin-destination pair were calculated using Open Street Map walking network dataset. Scores were combined into nine domains using ranking and weights generated by factor analysis: Education, Employment, Environmental ‘Bads’, Health, Māori Culture, Public Transport and Communications, Recreation, Social and Cultural, and Shopping. The domains are combined to create the overall accessibility index score for each urban SA1, which are visualized using static and dynamic mapping. Results show geographic variation across New Zealand, and within urban areas. We apply the resulting index to case studies to explore relationships between accessibility to community resources and other datasets relating to urban regeneration, such as health and wellbeing. We have developed the most detailed and comprehensive urban walking accessibility index to date for New Zealand, using accurate and up-to-date datasets for many commonly used and publicly accessible facility types. We hope that the index we produced will be widely used in policy development, future research and resource allocation, to inform effective urban regeneration policies.

A33. Context over convenience: a critical perspective on spatial weights in spatial-statistical research

Fillipe Oliveira Feitosa and René Westerholt

Abstract

In spatial analysis, the use of geometric distance and physical adjacency as the basis for spatial weights matrices (SWMs) has long been a quasi-standard. These approaches are rooted in a geometrical perspective and offer computational convenience, yet they often oversimplify how spatial relationships actually function. Many spatial processes are shaped by more than just physical proximity, making conventional SWMs insufficient in certain contexts. This work critically examines the over-reliance on geometry-derived spatial weights and offers recommendations for recognizing cases where alternative models should be considered. We highlight a range of contexts where spatial dependence defies the logic of geometric distance or adjacency. These include settings shaped by institutional connections, digital systems, economic interactions, and other partially or non-physical networks. In such cases, dependencies may be unequal, evolve over time, or emerge from feedback mechanisms and spillover effects. Applying standard, off-the-shelf SWMs in these contexts can lead to misleading interpretations, masking the actual forces shaping spatial dynamics, and weakening the credibility of empirical findings. Rather than endorsing a universal technical fix, we advocate for a modeling approach grounded in the specific features of the spatial phenomena under investigation. Our main argument extends beyond methods, fostering conceptual and critical analysis. Spatial structures should emerge from the actual relationships and mechanisms present in the data, not be imposed through predetermined geometric assumptions, where justified by the phenomenon. Our goal is to encourage a paradigm shift from defaulting to proximity-based models toward more nuanced, theory-informed modeling practices. This involves critically assessing when traditional SWMs suffice and when they fail. By doing so, researchers can build spatial models that are not only methodologically sound but also better reflect the lived and relational geographies shaping today’s spatial realities.

A34. Advancing equity of health service accessibility through spatial analysis: Applying the 3-step floating catchment area method for examining physiotherapy in Aotearoa New Zealand

Miranda Buhler, Tayyab Shah, Meredith Perry, Marc Tennant, Estie Kruger, Stephan Milosavljevic

Abstract

Unequal access to health care contributes to disparities in health outcomes. Physiotherapy services can play a major role in conditions that excessively burden Aotearoa New Zealand (NZ) Māori, Pacific, high socioeconomic deprivation and rural populations. However, the distribution of the NZ physiotherapy workforce relative to these populations is not known. This study demonstrates a novel application of the distance-based 3-step floating catchment area (3SFCA) method to examine the distribution of NZ physiotherapy workforce spatial accessibility at a high acuity level stratified by demographic indicators of health need (Māori or Pacific ethnicity, socioeconomic deprivation, rurality). Geocoded physiotherapy workforce data for 5,582 physiotherapists (92% of the 6,038 registered physiotherapists at March 2022) were integrated with 2018 NZ Census data to generate 'accessibility scores' (weighted practitioner: population ratio) for each Statistical Area 2 using the 3SFCA method. Demographic characteristics of rurality, Māori ethnicity, Pacific ethnicity, and socioeconomic deprivation were categorised based on pre-defined rationale, cross tabulated with accessibility scores, and thematically mapped using ArcGIS software. Trends in relationships were explored statistically using correlation techniques. The study identified specific geographic locations of inequity where health need is high, yet accessibility of physiotherapy is low (<0.94 to 9.06 per 10,000). In particular were areas in the mid-central North Island, west and northern Northland, and the East Coast. In comparison, areas of highest accessibility included the South Island lakes district and central Canterbury (up to 26/10,000). The mean accessibility score was 11.88 per 10,000. At a national level, low physiotherapy accessibility was statistically associated with high proportion population Māori ethnicity and more rural location. The methods developed in this study can be implemented regionally to help establish profiles of health services that meet local population needs.

A35. From Brainwaves to Cityscapes: Exploring the Neurophysiological Dimension of Place Perception

Jiaxin Feng, Bo Zhao, Chao Qin

Abstract

GIS researchers have shown interest in incorporating the human brain to learn place perception and psychological interaction with space. Nowadays, advances in neural and location tracking technologies open up opportunities for inferring ever-changing place perception and emotion from brain activities. This has the potential to advance dynamic investigation and offer neuroscientific evidence to inform urban policy. However, research that leverages these opportunities to rigorously examine brain activities in an interconnected geospatial context is still rare. This paper explores if ongoing psychological processes about urban spaces can be inferred from brain activities using electroencephalography (EEG) and machine learning. It further examines if this approach can gain meaningful insights into place perception from a dynamic perspective. To demonstrate this approach, an empirical study investigated the fluctuating emotions of individuals in three urban settings in Seattle. Emotions were inferred from instantaneous, continuous neurophysiological indicators and then compared with self-report measurements. The results reveal that neurophysiological indicators of emotional responses could detail what emotions were felt over time in relation to which urban scenes, when EEG was used with video recording and location tracking. This study makes a significant methodological contribution to GIS and geographic research. Through understanding the dynamic place perception, policy about place can be made better for the cognitive and emotional well-being of people.

A36. Using support vector machine and decision tree to predict mortality related to traffic, air pollution, and meteorological exposure in norway

Cong Cao, Jan Morten Dyrstad, Colin P. Green

Abstract

Cardiovascular and respiratory disease (CPD) is a leading cause of death worldwide. There is increasing evidence that air pollution and exposure to extreme weather conditions have important contributory roles. In practice, understanding the interaction of these factors is difficult due to the complexity of the relationship between CPD, air pollution, and environmental factors in general. This paper returns to this point and uses machine learning approaches to explore these relationships focusing on four cities in Norway, as well as investigating whether meteorological factors and air pollution have a synergistic effect on CPD. We demonstrate that machine learning outperforms regression models in terms of the accuracy of predicting CPD mortality, as regression models are prone to overfitting with the increase in variables. We show the importance of the interaction between weather and air pollution. We demonstrate that extreme weather is associated with higher CPD mortality, as is exposure to air pollution in the form of NOx and particulate matter. These effects are most pronounced for older 75-year-old individuals. Our results suggest policy responses for mitigating the negative health impacts, especially for vulnerable age subgroups.

A37. GIscience, AI/ML, and community-based approach: An innovative approach to flood mapping using static cameras and custom CRIS-HAZARD Flood App Data

Barnali Dixon

Abstract

Extreme weather events are causing flooding around the world, and coastal counties in Florida are no exception. Hydrological and hydrodynamic models can benefit from detailed flood data, i.e., flood depth and extent during extreme rainfall and storm surges. The use of static (in-situ) cameras that monitor real-time water level changes, coupled with data extraction from a custom flood reporting app and social media, can offer a cost-effective, viable option for expanding data on flood depth and extent. In 2024, Pinellas County (PC), a coastal county in Florida, experienced Tropical Storm Debby, Hurricane Helene (a storm surge event), and Hurricane Milton (a rainfall event). It is estimated that these three storms caused $2.5 billion in damage to the PC. Storm surge for Helene was ~7.1 ft, and the highest rainfall captured for Milton was ~16 inches. We will demonstrate 3 case studies: i) the use of static camera-derived flood depths and flood inundation maps to predict building damage due to flooding from Helene, ii) the use of CRIS-Hazard Flood App to predict Flood depth using AI, and iii) the use of social media based data for flood inundation mapping. For detailed flood data, 26 static solar-powered Wi-Fi-enabled cameras with reference poles and night vision capabilities (for day and night time data collection every 15-minute interval) were installed in Phase I, with plans to install nine more. We also created a custom CRIS-Hazard Flood App to source geotagged flood photos that are used to extract flood depth using a Convolutional Neural Network (CNN). We utilized Nextdoor, a community-based app, to collect ground-truth data from community members to calibrate and validate the flood map created for this project.

A38. Using foundation model embeddings to map Colorado's Build Hazard Interface for Wildfire

Aleksander Berg and Stefan Leyk

Abstract

Since 1980, the United States alone has seen 403 “billion dollar” weather and climate related disasters, accounting for 16,918 deaths and over $2.9 trillion dollars in damages. In the past three years, the U.S. has seen 73 of these billion-dollar disasters with 1,534 total deaths and $461.6 billion in damages. Climate and weather hazards such as fire, landslides, floods, heat waves, and cold snaps are accelerating in their destructiveness and deadliness due to climate change, development patterns, differential vulnerability, and numerous other human-environmental factors. In this project we propose a new framework and methodology for dynamically mapping spaces where the built environment and hazard prone areas meet, which we term the Built Hazard Interface or BHI. We use the Clay geospatial foundation model to generate embeddings of Landsat imagery to build a semantic picture of land cover in our study area. We use these embeddings in concert with the highest resolution data available on the built environment (parcel and building footprint data) in a LightGBM classifier to map Colorado’s BHI for wildland fire. We present a testbed analysis generating these maps for the wildland urban interface (WUI), a space interaction between flammable vegetation and the built environment, for the state of Colorado, USA, from 2010-2020. Our workflow using a foundational geospatial model presents a start on how to map hazard prone spaces in a way that better represents the complex climatological, environmental, and social processes that drive heightened exposure in the first place at the spatial and temporal scales at which they operate.

A39. Spatially explicit machine learning for assessing equality in urban green space accessibility

Zih-Hong Lin, Shawn Laffan, Graciela Metternicht

Abstract

Urban green spaces (UGS) offer multiple benefits to residents, including improved air quality, heat mitigation, and enhanced opportunities for exercise. For urban planners, aside from providing sufficient UGS, ensuring equitable access to these spaces is crucial for promoting environmental justice in cities. This study uses spatially explicit machine learning methods to help planners understand how socio-economic development affects equitable access to UGS and to predict the levels of equality. We investigate the potential driving factors behind UGS equality within the Taipei metropolis from 2019 to 2023. The Gini coefficient is utilised to assess equality levels of UGS access at the village level. Spatial lagged variables are integrated into tree-based boosting models to reflect various spatial econometric specifications, allowing comparison of their performance with traditional spatial and non-spatial regression models. SHapley Additive exPlanations (SHAP) are applied to identify feature importance and explore non-linear relationships between socio-economic indicators and the calculated Gini coefficients. Results indicate that machine learning algorithms outperform traditional spatial statistical models. Furthermore, incorporating spatial lagged variables into machine learning models enhances their predictive accuracy compared to models lacking spatial consideration. The findings inform urban planners about spatial disparities in green space distribution, highlighting the need for location-specific interventions. This research supports inclusive urban green space development in Taipei, aligning with the Sustainable Development Goals, particularly target 11.7, which emphasises providing universal access to safe, inclusive, and accessible green spaces.

A40. Improving indoor 3d semantic segmentation with multi-view geometric constraints

Ding Youli

Abstract

Indoor 3D semantic segmentation with jointly 2D-3D deep neural networks is a popular yet challenging way for stereoscopic scene understanding through cross-modality information fusion. Challenged by low fidelity of 3D reconstructed models and complexity of indoor scenes, the semantic segmentation accuracy of current methods is still limited by the unstable scene information and ineffective fusion strategies. To this end, this paper introduces a multi-view geometric constraints-based 2D-3D jointly learning method for robust indoor 3D scene semantic segmentation (termed as MVGeo). Instead of directly using pixel-point mapping features, we construct a multi-view geometric constraints that can guide a triangular feature interaction between multi-view images and 3D geometry. In this way, MVGeo can absorb the complementary context information both cross-view and cross-dimension, and thus alleviate the noisy influence accused by poor indoor conditions and provide a stable information enhancement for 3D semantic learning. Accordingly, a cross-view context aggregator and a cross-dimension feature lifter are developed to strength the accuracy of 2D feature representation and guide a geometry-enhanced 3D training respectively. Extensive experiments demonstrate conducted on the challenging ScanNetv2 dataset show that the proposed MVGeo can effectively enhance the accuracy of cross-modality semantic learning, yielding outstanding 3D segmentation performance with 77.6% mIoU.

A41. Integrating the Geographic Dimension into Information-Epidemic Co-evolution: A Spatially Explicit Agent-Based Model

Yepeng Shi, Ling Yin, Kemin Zhu, Zhengan Xiong, Kang Liu

Abstract

With virtual and physical spaces intertwined by advancing information technology, understanding information–epidemic co-evolution is crucial for improving epidemic control. Researchers often use microscopic Markov chains in multilayer networks to study information–epidemic co-evolution. However, those models neglect the geographical perspective, constrained by two main issues:1）Spatial Representation Deficiency: They rely on abstract networks that omit the critical geographic dimension necessary for realistically depicting information–epidemic interactions; 2）Network Linkage Deficiency: They overlook that individuals who have physical contact are highly likely to form virtual social links—indicating that these two network layers are not independent. To address this, we propose the Agent-based Spatially Explicit Info-Epidemic Model (ASEI) incorporating geographic space into information–epidemic dynamics. First, ASEI remedies the spatial representation deficiency by constructing spatially explicit multilayer networks, embedding interactions within realistic geographic contexts in households, workplaces, schools, and communities at a city scale. Second, drawing on a questionnaire-based study, we demonstrate and model the reciprocal influence between physical interactions and virtual links. Building on these findings, we then establish a spatially explicit information–epidemic network that integrates the network structures of both physical and virtual interactions. Finally, we employ an agent-based simulation at the city scale to capture individual interactions across these spatial layers, modeling the spatiotemporal co-evolution of information–epidemic. Our simulation results reveal: 1. Information dissemination exhibits distinct spatial clustering patterns in its early stages, further reinforced as the epidemic unfolds. 2. This interdependent network significantly amplifies the influence of information on epidemic spread, demonstrating its far-reaching real-world impact. 3. Targeted information dissemination within specific urban geographic communities validates the model’s utility for spatially targeted control strategies. Overall, ASEI is capable of offering new insights into information–epidemic co-evolution, emphasizing geography's critical role in network interactions and guiding effective control strategies for urban environments.

A42. Incorporating Geolocation into Deep Learning for Dynamic Air Pollution Estimation: Performance and Generalizability Impacts

Morteza Karimzadeh, Zhongying Wang, James Crooks

Abstract

Deep learning models have demonstrated success in geospatial applications, yet quantifying the role of geolocation information in enhancing model performance and geographic generalizability remains underexplored. A new generation of location encoders have emerged with the goal of capturing attributes present at any given location for downstream use in predictive modeling. Being a nascent area of research, their evaluation has remained largely limited to static tasks such as species distributions or average temperature mapping. In this paper, we investigate the impact of incorporating geolocation into deep learning for a real-world application domain that is characteristically dynamic (with fast temporal change) and spatially heterogenous at high-resolutions: estimating surface-level daily PM2.5 levels using remotely sensed and ground-level data. We build on a recently published deep learning-based estimation model that achieves state-of-the-art on data observed in the contiguous United States. We examine three approaches for incorporating geolocation: excluding geolocation as a baseline, using raw geographic coordinates, and leveraging pre-trained location encoders. We evaluate each approach under within-region (WR) and out-of-region (OoR) evaluation scenarios. We conduct feature importance analysis using interpretability methods to assess the relative importance of location in the models’ predictive performance. Our findings indicate that while naïve incorporation of geographic coordinates improves within-region performance, it can hinder generalizability across regions. In contrast, pre-trained location encoders like GeoCLIP enhance performance and geographic generalizability for both WR and OoR scenarios. However, our results also show varying performance among location encoders such as SatCLIP versus GeoCLIP, various patterns in urban and rural areas, potentially affected by the pre-training data used for each location encoder. To the best of our knowledge, this is a first quantification of the impact of location using interpretability methods, and a first evaluation of location encoders in a complex, temporally dynamic estimation scenario.

A43. Unveiling image-based geo-localization model: interpretable geolocation through conceptual understanding

Lanxin Liu and Furong Jia

Abstract

Image-based geo-localization models, such as GeoCLIP, enable image localization for any location on Earth through a multimodal alignment mechanism. While these models show great promise, both their interpretability and the research into how they understand geographic concepts—such as regional landscapes and climate—remain limited. Interpretability is essential for validating their reliability and ensuring transparency in their predictions, especially in applications such as urban planning, environmental monitoring, and disaster response. This study investigates whether image-based geo-localization models inherently develop a generalizable understanding of geographic knowledge, enabling them to perform image geolocation. Specifically, we examine if these models can, like humans, infer a city from an image by recognizing distinctive natural or cultural features. To this end, we develop a geographic knowledge concept bank that systematically links visual cues to location hypotheses while simultaneously converting state‑of‑the‑art geographic vision models into interpretable counterparts, revealing a positive interplay between transparency and predictive accuracy. This work advances the interpretability of image-based geo-localization models, providing an efficient, semantically rich framework to unravel their reasoning processes and multimodal alignment mechanisms. Our approach allows geographers to analyze and engage with these models in an intuitive, human-centric way.

A44. Assessing Representativeness and Bias in Mobile Phone Locational Data

Hongyi Zou

Abstract

Locational data derived from mobile phone devices offer significant opportunities for spatiotemporal analysis of human movement patterns. However, such data are rarely complete or socially neutral and always pose challenges related to completeness and representativeness. Completeness issues can arise from device inactivity, data affordability and network limitations. As a result, not all users’ locations are continuously tracked. These gaps affect the ability of mobile phone locational data (MPD) to represent human movement patterns accurately. This study evaluates the impacts of completeness and representativeness of MPD on their usability. We conceptualize completeness as the fraction of hourly intervals during which at least one location point is recorded by a mobile phone. Completeness is calculated across three temporal periods: whole-day, active-hour, and night-hour periods. We define active hours as 7 am to 10 pm based on exploratory analysis of 1,000 random users in Christchurch with at least seven consecutive days of data in 2022. During these hours, location data collected by mobile phones were consistently more frequent than during other times of day. We assess the numbers of users reaching completeness thresholds ranging from 0.1 to 0.9 over a range of consecutive days varying from two to seven days). We also examine how streaks of consecutive days with different completeness thresholds fluctuate throughout the year and relate to significant events, including holidays, COVID-19 restrictions and changes in the data collection process. We further investigate data usability and representativeness by detecting home location under different completeness thresholds and validating outputs with Census population data for New Zealand. Based on these findings, we identify appropriate completeness levels and consecutive-day indices for different analysis and propose reliability thresholds for MPD selection. This work contributes scalable methods for evaluating MPD quality, investigating hidden biases, and fostering more transparent, context-aware use of MPD across multiple contexts.

A45. Exploring and visualizing the edge effect in areal data analysis

Ikuno Yamada and Yukio Sadahiro

Abstract

The edge effect, a prevalent yet underexplored issue in spatial analysis, affects analytical results around the boundary of a study region due to missing information beyond its boundary. This challenge is particularly pronounced in areal data analysis within the local context, compared to point pattern analysis, which often benefits from established edge correction methods implemented in GIS and other analytical tools. In the absence of such readily available tools, researchers may overlook the potential risks associated with the edge effect in areal data analysis, implicitly assuming it does not pose a significant problem. This study delves into the characteristics and magnitude of the edge effect across different scenarios in areal data analysis, using local Moran’s I statistic as a representative analysis method. Furthermore, we propose a new edge correction method that adds a statistical perspective to existing approaches and evaluate its effectiveness against them. Given that the proposed method and the statistical testing via local Moran's I rely on random sampling and random permutation of observed values, respectively, the testing results naturally exhibit some degree of random variability. This study explores how to integrate such variability into the visualization of the testing results, allowing for a simultaneous assessment of their significance and potential instability. A preliminary analysis employing a regular grid system and a spatial distribution with no significant patterns was conducted to evaluate the edge effect concerning Type-I errors. The findings revealed that areal units near the study region's boundary exhibited a higher incidence of Type-I errors than those located well within the region. They also suggested that visualization that incorporates statistical variability aided in cautious interpretation. The proposed edge correction and visualization methods offer a new approach to handling the edge effect to enable more reliable interpretations of analytical results even with the presence of the edge effect.

A46. Assessing Spatial Patterns and Risk Factors for Preterm Birth in New Zealand Using GIS

Cristal Salatas, Matt Hobbs, Daniel J Exeter, Tanith Alexander, Frank H. Bloomfield, Clare R. Wall

Abstract

Background: Preterm birth (PTB) is a leading cause of neonatal morbidity and mortality worldwide. While PTB has multifactorial causes, modifiable factors such as nutrition, socioeconomic status, and environmental exposures provide opportunities for intervention. We applied national-level spatiotemporal approaches to identify geographic disparities and risk factors for PTB across New Zealand (NZ), integrating individual-level data with area-level environmental and socioeconomic metrics. Method: National data from 2003 to 2022 (n=1,189,244 births) were obtained from the NZ Integrated Data Infrastructure, a secure virtual environment that links multiple national datasets through individuals’ unique national identifiers. We examined these data to determine associations between PTB and socioeconomic characteristics (area-level deprivation deciles: D1=most deprived, income, education), micronutrient supplementation, dietary habits (fruit and vegetable servings), and environmental factors (Healthy Location Index [proximity to food, vape, alcohol, and physical activity outlets] and domicile [urban/rural using StatsNZ 2018 classification]). Multilevel spatial analyses (areas >6 PTBs per population) were conducted. Logistic and multilevel models were applied using RStudio Results: The final analysis included 318,180 births, with 23,705 (7.5%) PTBs. Spatiotemporal analyses revealed persistent PTB hot spots in rural areas, with five clusters identified, four in Counties Manukau and one in Bay of Plenty. Temporal analysis revealed peak cluster intensities in these regions in 2021 and 2010, respectively. Proximity to fast food outlets (OR 1.26, 95% CI: 0.19-1.44) and absence of daily fruit and vegetable servings (OR 1.78, 95% CI: 0.22-2.84) were associated with increased odds of PTB. Lower area-level socioeconomic deprivation (OR 1.12 per decile, 95% CI: 1.01-1.24) and tertiary education (OR 1.21, 95% CI: 1-1.46) were also associated with higher odds. Conclusions: This novel NZ-based study identifies key risk factors for PTB and highlights the need for geographically tailored interventions, particularly in rural areas and regions with limited access to healthy food options.

A47. Urbanization Unbound: VIIRS/DNB Insights into Thailand's Sprawl Beyond Municipal Borders.

Sirikul Hutasavi, Mohammad D. Hossain, Siam Lawawirojwong

Abstract

Thailand's rapid urbanization often extends beyond established municipal boundaries, creating gaps in governance that hinder sustainable development. This study utilizes VIIRS/DNB nighttime light data from 2013 to 2023, along with advanced space-time analytics, such as Emerging Hot Spot Analysis (EHSA) and local entropy mapping, to examine patterns of urban sprawl and their misalignment with administrative jurisdictions. The findings reveal that 84% of municipal areas showed no significant growth patterns in nighttime light during this period. Additionally, more than 18% of urban expansion in Thailand occurs outside or inconsistently with municipal boundaries, particularly in peri-urban areas, such as the outskirts of Bangkok and secondary cities like Nakhonsawan. Municipalities that exhibit stagnant growth patterns, like Bang Pramung Municipality, underscore the governance challenges associated with fiscal centralization. Moreover, a statistically significant negative correlation (R² = -0.34, p < 0.05) exists between nighttime light growth and municipal effectiveness, suggesting systemic inefficiencies, including rigid boundary definitions and uneven resource allocation. This study argues that Thailand's "unbound" urbanization necessitates dynamic boundary reforms, decentralized fiscal policies, and infrastructure investments that promote growth across jurisdictions. By connecting geospatial insights with governance frameworks, policymakers can enhance the management of urban sprawl, ensuring equitable development and aligning administrative borders with the actual conditions on the ground.

A48. A Novel Hybrid Cellular Automata-Deep Learning Framework for integrated 3D urban growth simulation

Farasath Hasan and Xintao LIU

Abstract

Urban growth modeling has conventionally struggled with unclear transition rules and a lack of rigorous uncertainty quantification, which hinders its capacity to accurately simulate and predict complex urban phenomena. In this study, we introduce a novel Hybrid Cellular Automata (CA)-Deep Learning framework designed for integrated 3D urban growth simulation. This framework innovatively captures both horizontal expansion and vertical intensification within a single modeling run, thereby bridging a significant gap in current GIScience research. The core of our approach lies in leveraging Shannon Entropy to establish robust transition rules that not only guide the simulation but also quantify uncertainty at the rule level. This quantification is instrumental in generating spatial uncertainty maps, which enhance the interpretability of the model outputs and provide decision-makers with critical insights into potential urban development scenarios. The CA component of our framework incorporates adaptive development pressure, stochastic variations, and explicit spatial constraints that reflect regulatory and physical limitations, ensuring that the simulated urban growth is both realistic and compliant with real-world conditions. A key innovation of our framework is the integration of deep learning techniques into the vertical simulation process. Specifically, a deep neural network is employed to predict building heights and volumes by utilizing kernel densities of commercial, industrial, and amenity factors. This represents a pioneering step in CA-based urban growth modeling by introducing an explicit vertical dimension to the simulation process, thereby capturing the dynamics of urban intensification alongside horizontal urban expansion. Furthermore, scenario simulations under varied policy constraints reveal diverse urbanization pathways, highlighting the framework’s potential to inform sustainable urban planning on a global scale. This research thus advances the state-of-the-art in urban growth modeling by providing a comprehensive, uncertainty-aware, and vertically integrated simulation tool for GIScience. Declaration: I used ChatGPT to refine my writing while keeping my original wording.

A49. Assessing cognitive load and temporal dynamics of brain activity during landmark-based navigation using mobile maps

Ioannis Delikostidis, Peyman Zawar-Reza, Abdul-azeez Bello and Binyang Han

Abstract

Studying brain activity and cognitive load during navigation with a mobile map in realistic virtual environments is a multifaceted endeavour. Cheng et al. (2023) addressed this issue by using spontaneous eye-blinks during navigation as event markers in electroencephalography (EEG) recordings. Prior research shows that eye-blink derived Event Related Potentials (bERP) are indicative of cognitive load. By analysing EEG data based on those markers during navigation in unfamiliar urban areas with a mobile map featuring 3, 5, or 7 landmarks, they identified an optimal number of landmarks for effective navigation. Building on their research methodology and experimental design, we adapted the approach to a 360-degree cylindrical projection immersive virtual environment, incorporating mobile EEG and realistic locomotion techniques. Our study investigates cognitive load during navigation with mobile maps showing 5, 6, or 7 landmarks, aiming to determine whether cognitive load increases when six landmarks are displayed and levels off at seven, or if it continues to rise with more than five landmarks. Additionally, we are employing microstate analysis, which identifies brief intervals of stable scalp potential fields (microstates) generated by brain networks. This method provides insights into the temporal dynamics of brain activity and large-scale network functions, helping us understand cognitive processes and differences in brain activity during navigation among individuals. By combining bERP and microstate analysis, our research contributes to advancing the field of spatial cognition and navigation. This integrated approach enhances our understanding of the relationship between the number of landmarks displayed on mobile maps and cognitive load, potentially guiding the development of neuroadaptive maps, as proposed by Cheng et al. (2023). Moreover, it enables the extension of existing research on individual differences—such as cognitive ability, experience, and impairment—in navigation performance, offering valuable insights that may inform new strategies to support individuals with cognitive decline in wayfinding tasks.

A50. Using Ridership Profile Clustering and Geographically Weighted Regression to Investigate Spatio-Temporal Patterns of Shared E-scooter Ridership in Christchurch, New Zealand

Goldie Leung, Katarzyna Sila-Nowicka, Lindsey Conrow, Grant McKenzie, Vanessa Brum-Bastos

Abstract

E-scooters have emerged as one of the most popular shared micromobility options. Yet, understanding their spatiotemporal usage patterns remains limited, partly due to the lack of available data beyond API-based trip endpoints used to derive origin-destination (OD) pairs for route estimation. As a result, it is challenging to address public concerns regarding infrastructure allocation, long-term sustainability, service regulation, and the safe integration of e-scooters into the broader urban transportation landscape. To address this gap, we present the first study to analyse shared e-scooter ridership at an hourly, street-segment level across an entire city using real trip count data. Our dataset, sourced from RideReport (https://www.ridereport.com/) includes hourly e-scooter counts for almost 30,000 street segments in Christchurch, New Zealand. From these data, we derived average hourly ridership curves for each street segment for Monday to Thursday, Friday, and weekends, reflecting differences in travel behaviour across the week. Then we applied Dynamic Time Warping and hierarchical clustering to group street segments with similar ridership curves. To investigate spatial variation in ridership volume, we used Geographically Weighted Poisson Regression (GWPR) to model the relationship between ridership and ten contextual variables, including population density and points of interest. Our clustering method identified groups of street segments with ridership patterns that match well-known types in transport research (e.g., utilitarian, leisure). This shows the method's potential as a data-driven way to classify usage types, without relying on assumptions about riders' demographics, trip purpose, or direction. Our analysis reveals that the proportion of cycleways, the number of bus stops, and the distance to car parks are the most significant factors driving shared e-scooter ridership at the local scale in Christchurch.

A51. Everyday mobility as an indicator of neighborhood change

Xiaoyue Xing

Abstract

Effective urban management requires timely assessment of neighborhood change. However, our ability to detect such fine-grained changes in urban space has been largely limited by the coarse spatiotemporal resolution of statistical survey data. Recent studies have used novel data sources, such as mortgage loan data, social media records, and housing prices, to extract timely-updated information on socio-economic activities and to capture neighborhood change. While contributing significantly to the study of neighborhood change, previous research often treats neighborhoods as isolated spatial entities, overlooking their interconnected nature. As Anthony Downs suggested in his 1981 theory, neighborhoods are inherently linked during urban development, influencing and being influenced by other areas. The growing availability of open-access mobility data presents an opportunity to analyze neighborhood change from an interconnectivity perspective. This study explores how large-scale daily mobility records can reveal shifts in neighborhood connectivity and further serve as indicators of neighborhood socioeconomic change. By constructing an aggregated mobility network in New York City, we quantify the topological centrality and functional centrality of neighborhoods and introduce a mobility-based index of neighborhood change. This index is then correlated with traditional indicators of neighborhood change, including the proportions of high-income individuals, elderly residents, and youth populations, to assess its effectiveness in capturing socioeconomic change. Using a case study of Manhattan and Brooklyn, we find that neighborhoods identified as declining or upgrading based on mobility dynamics are consistent with policy interventions during the study period and correspond to neighborhood changes observed in previous research using social media data. We further discuss the limitations of mobility data, including biases and validation challenges, while highlighting opportunities for integrating new mobility data sources to improve neighborhood planning. This study underscores the value of mobility-based metrics in monitoring and understanding neighborhood change within the interconnected urban system.

A52. A Multi-Algorithm GPS Error Analysis Framework (MAGAF) to assess positioning accuracy in complex urban environments

Tianyu Li, Yanjia Cao, Chi Yan Leung

Abstract

Understanding GPS positioning accuracy in complex urban environments is crucial for analyzing human mobility patterns and activity space. This study evaluates GPS device performance across diverse urban environments in Hong Kong, focusing on the relationships between urban morphology and GPS positioning errors. GPS data were collected across 32 routes ranging from high-density commercial centers to open space. We first evaluated the effectiveness of two common GPS data quality approaches: conventional quality metrics filtering and multiple-device deployment validation. Building upon these initial assessments, we propose a Multi-Algorithm GPS Error Analysis Framework (MAGAF) to evaluate urban morphology impacts on positioning accuracy. To capture both linear and non-linear relationships between urban morphological features and GPS errors, we implemented a comparative modeling approach. Starting with Ordinary Least Squares (OLS) as a baseline and incorporating regularized linear models (Ridge and Lasso), we implemented advanced machine learning algorithms, including Random Forest (RF), Support Vector Regression (SVR), and XGBoost, to capture complex non-linear patterns. The best-performing model was selected based on comprehensive evaluation metrics, and SHAP (SHapley Additive exPlanations) analysis was applied to interpret feature contributions to GPS positioning errors. This modeling framework provides robust interpretation of urban effects on GPS accuracy through model comparisons and feature importance analysis. Our analysis revealed limited effectiveness of traditional GPS data quality approaches, with quality indicators failing to reliably distinguish GPS positioning errors and multiple devices demonstrating statistically similar performance. The XGBoost model significantly outperformed the baseline OLS model (R² of 0.97 versus 0.06), with Floor Area Ratio (total floor area to plot area ratio, FAR) and Mean Building Height identified as the primary contributors to GPS accuracy. The findings established quantitative relationships between urban morphology and GPS positioning errors, providing evidence-based insights for GPS trajectory analysis in complex urban environments, particularly in high-density areas.

A53. Predicting Short-Term Urban Bike Sharing Demand in a Coupled Continuous and Network Space

Shen Liang, Yang Xu, Guangyue Li

Abstract

Bike sharing systems represent critical components of sustainable urban mobility infrastructures. While existing research has applied various geospatial methods to predict bike sharing demand, most approaches model spatial dependencies solely in Euclidean space or through station-based connections. Such methods often neglect the complex topological structures of urban transportation networks (e.g., metro, cycling networks) that influence bike sharing patterns. This study introduces a novel geographical artificial intelligence framework that addresses this gap by integrating both continuous space and network space in modeling bike sharing demand. Our approach leverages convolutional neural networks to model spatial proximity effects while simultaneously employing graph convolutional networks to capture the topological dependencies inherent in urban transportation networks. Through comprehensive evaluation using spatiotemporal bike sharing data from five global cities, we find that incorporating transportation network topology significantly enhances prediction accuracy, with improvements of up to 8.9% in RMSE. Spatial analysis reveals that prediction enhancements are most pronounced in areas with high demand levels and locations near metro stations. This research contributes to the advancement of geographic information science by demonstrating how explicitly modeling the coupled relationship between continuous and network spaces can improve our understanding of urban travel patterns and enhance the predictive capability of geospatial models. The proposed framework offers promising applications for spatial predictive systems in smart city initiatives and spatiotemporal forecasts of other urban phenomena that exhibit complex spatial dependencies.

A54. Assessing Spatio-Temporal and Process Controls on Anthropogenic Landforms with Landscape Evolution Models

Vinuri Piyathilake, Matthew Hughes

Abstract

Infrastructure systems upon which modern society depends are constructed within landscapes reshaped continuously by natural processes. Large-scale civil infrastructure works are also significant terraforming activities resulting in anthropogenic landscapes. Indigenous landscapes resulting from centuries of occupation also contain anthropogenic terraforms, and possess sacred sites and geological/geomorphic features considered fundamental to community cultural and spiritual identity. These anthropogenic landscapes are subject to long-term erosion and sedimentation processes that pose risks to the functioning of critical infrastructure and preservation of cultural heritage sites, and projected climate and land use changes may exacerbate these risks. Landscape evolution models (LEMs) numerically simulate Earth surface processes and are used to study long-term landscape change. Most applications project transformations over millennia, reconstructing past events or predicting future conditions. While these models identify geomorphic trends over a range of spatio-temporal scales, they rarely determine the timing erosion and sedimentation will impact engineered landforms or landscapes important to Indigenous peoples. Advancing LEMs to address these spatio-temporal and process-based uncertainties is essential for improving the predictive accuracy of geoforecasting, and understanding landscape responses to environmental change. This study employs LEMs using the Landlab platform, a Python-based framework designed for simulating Earth surface processes using Digital Elevation Models. By incorporating site-specific geomorphic knowledge, we model the unique surface processes influencing landscape evolution at each selected location. Our primary objective is to determine the timing at which specific points of interest will be affected by landscape change (erosion and sedimentation). To identify dominant processes, we systematically manipulate key parameters, holding certain variables constant to isolate their individual impacts. By iteratively adjusting these parameters, we aim to refine our understanding of the interactions between erosion, deposition, and hydrological dynamics, ultimately enhancing the predictive capabilities of landscape evolution models to determine the longevity of engineered landforms and inform long-term management of Indigenous landscapes.

A55. A Comparison of Two Steering Laws for Unconstrained Immersive Hybrid Navigation

Ronny Rowe

Abstract

Navigation within immersive environments is a fundamental yet often neglected human-computer interaction. Consequently, methods for designing and evaluating the effectiveness of immersive navigational interfaces remains at the forefront of HCI research. The Steering Law: a quantitative human performance model for trajectory-based HCIs, has been used as a usability tool to address this problem. However, the model is limited by the assumption of constrained path motion, and it is not known whether revised models which allow for unconstrained navigation (important for immersive analytics) can also be used for usability testing of immersive navigational interfaces. To address this research problem, we conducted a usability test which compared the traditional Steering Law with an unconstrained model (the Yamanaka Steering Law) for navigation in an immersive 3D VE with an HTC Vive controller and hybrid navigation (HTC Vive controller + subliminal teleportation). The use of the Yamanaka Steering Law was significant for subsets of the study populations identified with k-means clustering. These were meaningful groups with shared technological familiarity and spatial ability characteristics, based on quantitative self-assessment measures. This implies that the Yamanaka Steering Law may be limited by an assumption of interface proficiency, which is consistent with prior research that included populations of subjectively defined expert users. Overall, the control interface (HTC Vive controller) was more effective than hybrid navigation, which was supported by a fractal dimension analysis of participant path trajectories. Our approach to addressing this problem through a synthesis of usability testing, geovisualisation, and GIScience is a novel contribution to objective data-driven usability testing of immersive navigational interfaces. Future research should entail expanding the scale, and geospatial and cognitive scope of experiments to further investigate unconstrained immersive navigational interfaces.

A56. Mapping the contribution of community and citizen science: from global to local scales

Shane Orchard, Carolynne Hultquist, Holly Johnstone

Abstract

The field of participatory community and citizen science refers to a broad range of activities where information is collected by members of the public, often as volunteers. Supporting these initiatives and facilitating the re-usability of data has great potential to assist environmental and socio-cultural objectives worldwide. In response, many new technologies are being developed from bespoke Apps for discrete projects, to image classification methods for analysis, and global platforms for the storage and sharing of observations. At the same time, there has been much discussion around the further potential of participatory science to fill spatial and temporal information gaps, and yet relatively little attention to mapping the contributions of the field or understanding the drivers of participation upon which it depends. Doing so has the potential to assist the sector with its own needs, which may include engaging with the wider public and the communicating of results to maintain or secure scarce funding or attract new members to community groups. Data visualisation and analysis tools present further opportunities that might support the sector by providing interpretation and reporting functions that recognise and respect its intended use. Accounting for these bottom-up aspects is similarly needed to identify and map the contributions of community and citizen science. This requires a considerably more nuanced, contextualised and locally-ground approach than merely aggregation against pre-established indicators such as those used by the Convention on Biodiversity or Sustainable Development Goals. Catering for both local and global evaluation and reporting contexts therefore presents a further area of opportunity and innovation that is highly relevant to the standardisation of metrics discourse. In this presentation we illustrate these principles with examples from geographic information observatories and retrieval systems that support community and citizen science initiatives and conclude with an overview of recent initiatives to map the contributions of this sector in Aotearoa New Zealand.

A57. Robust and Scalable Spatial Predictions with Multi-Resolution Thin-Plate Splines: Bridging Deep Learning and Geostatistics

ShengLi Tzeng

Abstract

Accurate spatial predictions are essential in GIScience and numerous applications, including weather forecasting, disease mapping, and environmental monitoring. We introduce a novel framework for robust and efficient spatial predictions using multi-resolution thin-plate spline basis functions that address critical limitations in existing methods. Traditional geostatistical techniques, such as kriging, provide best linear unbiased predictions but face significant computational challenges with large datasets due to the cubic complexity of covariance matrix inversion. Fast Fourier convolution can handle large datasets efficiently but requires regularly spaced sampling locations. Both methods assume Gaussian processes and struggle with outliers and non-stationary characteristics, limiting their robustness. Even recent deep learning approaches, such as multilayer perceptrons, often yield suboptimal performance when directly using raw coordinates as inputs and observations as outputs. Our approach is an improvement of the recent DeepKriging framework, which uses neural networks to avoid matrix inversion bottlenecks and handle irregular sampling by transforming coordinates into basis functions. While DeepKriging represents a significant advancement, its performance heavily depends on the basis function selection, with improper choices potentially leading to unstable or inaccurate results. We adopt a novel family of multi-resolution thin-plate spline basis functions to eliminate the need for complicated basis function allocation. Additionally, we enhance robustness by incorporating Huber's loss function, which is proven to be resistant to the outliers common in real-world spatial data. Through simulations and applications to Taiwan air pollution data, our method consistently outperforms traditional kriging and DeepKriging, particularly for non-stationary, irregular, and noisy datasets. By avoiding computationally expensive matrix inversion, our approach achieves scalability for massive datasets while maintaining accuracy for non-Gaussian processes and non-stationary problems. This work advances geostatistics by bridging deep learning with spatial reasoning, offering a robust, scalable, and practical tool for applications such as weather forecasting, disease mapping, and environmental monitoring.

A58. Uncovering Freight Flow Dynamics: A Data-Driven Approach Using Edge-Convolutional model

Ziyi Wang and Meihan Jin

Abstract

Freight vehicle trajectories play a critical role in urban logistics systems, reflecting operational efficiency and exhibiting strong correlations with urban transportation functionality and safety. While trajectory mining for freight vehicles has gained considerable attention across multiple research fields, key challenges remain to be concerned. Existing studies often rely on trajectory origin-destination (OD) network structures but neglect the semantic features of interaction networks. Additionally, conventional methods are usually designed based on static network models, which inadequately capture temporal dependencies among logistics activities. To address these gaps, this study proposes a novel spatial-temporal-semantic integrated framework based on spatial interaction network analysis. After identifying freight vehicle dwelling locations using spatiotemporal DBSCAN clustering, we construct a temporally ordered freight vehicle flow network incorporating dwelling duration, land use attributes, and spatial distribution patterns. An enhanced edge-convolutional network model is developed for spatiotemporal trajectory representation learning, and combined with spectral clustering to infer semantic patterns. Interpretable machine learning method is further applied to uncover spatiotemporal dynamics in distribution of freight vehicle travel flow. The case study is conducted with 110,000 freight vehicles in Shenzhen during one week (270 million GPS records). Clustering results reveal distinct spatiotemporal distributions for long- and short-duration dwelling behaviors and identify diverse freight vehicle travel semantics. Contrary to human mobility networks, clusters of freight vehicle dwelling locations are not necessarily spatially contiguous, with certain clusters exhibiting cross-district interaction patterns. These patterns strongly correlate with urban functional structures, particularly logistics facilities and hubs distribution. The proposed framework aids in understanding freight flow network dynamics and offers insights for optimizing spatiotemporal resource allocation in freight systems. The findings provide empirical evidence to guide logistics operations and urban freight infrastructure planning.

A59. Reproducibility in GIScience: How we can leverage reproducibility to enliven research articles.

Augustus Ellerm

Abstract

Paradoxically, as computational methods have been increasingly adopted in science, reproducibility has suffered. A Nature survey (Baker) revealed that over 70% of researchers failed to reproduce another’s experiments, and more than half couldn’t replicate their own work. In GIScience, reproducibility remains a challenge: a systematic assessment of AGILE conference papers found that over 80% lacked the data, code, or documentation necessary to support the reproduction of the research (Nüst et al.). In response to these challenges, various solutions have emerged. Literate Programming Environments like Jupyter, Quarto, and RMarkdown enable researchers to combine code, data, and methods with a research narrative. Workflow management systems such as CWL and Kepler enable the modular composition of computational tools. Containerisation technologies such as Docker and Kubernetes support the creation of portable computational environments. Collectively these technologies have moved computational science toward more robust and reproducible research. Despite these advances, opportunities remain to make reproducibility the default rather than an exception. Research artifacts such as code and data remain fragmented, and changes to these artifacts are not reflected in their associated article. LivePublication (Ellerm, Gahegan, and Adams) tackles these limitations by interfacing research articles with in-situ computational experiments. As new data is processed, results generated, and methods modified, so too is the article. In this talk I demonstrate how we can take a GIScience workflow, such as monitoring and predicting costal erosion, and interface it with a dynamic publication. The result is a publication container which must be reproducible, as the research article itself is an emergent property of executing the workflow. Baker, Monya. "Reproducibility crisis." nature 533.26 (2016): 353-66. Ellerm, Augustus, Mark Gahegan, and Benjamin Adams. "Livepublication: The science workflow creates and updates the publication." 2023 IEEE 19th International Conference on e-Science (e-Science). IEEE, 2023. Nüst, Daniel, Carlos Granell, Benedikt Hofer, Martin Konkol, Frank O. Ostermann, Rasa Sileryte, and Victor Cerutti. “Reproducible Research and GIScience: An Evaluation Using AGILE Conference Papers.” PeerJ, vol. 6, 2018, e5072. https://doi.org/10.7717/peerj.5072

A60. Cross-Disciplinary Opportunities for GeoAI and Human Geography

Song Gao

Abstract

The convergence science involves bridging different disciplines to tackle global challenges, such as climate change, urbanization, public health crises, disaster resilience, and ethical implications of technologies. Solving these challenges requires interdisciplinary knowledge, technologies, methodologies, and practices. This talk will present the synergies to identify and facilitate cross-disciplinary opportunities for GeoAI and human geography. A previous conversation about GeoAI, counter-AI, and human geography by Janowicz, Sieber, and Crampton (2022) discussed several opportunities and challenges in defining intelligence for machines, and the role of human in GeoAI development and decision-making processes, emphasizing the need for ethical considerations and accountability in the development of responsible GeoAI. The relationship between GeoAI and human geography is reciprocal. On one hand, the advancements of GeoAI technologies (especially spatially-explicit deep learning and generative AI) together with spatiotemporal big data have significantly expanded the views of human geography researchers on location, space, place, and complex human-environment interactions. On the other hand, human geography rooted in understanding of human behaviors in relation to the environment across spatial-socio-technical systems can provide valuable insights into the social, cultural, economic, and critical considerations that drive the development of ethical GeoAI in heterogeneous contexts. Co-design, also known as participatory design, is a cross-domain collaborative approach to convergent research that actively involves stakeholders, which offers diverse perspectives and strategies for integrative problem-solving across multiple disciplines. I will present a co-design framework of convergent research on GeoAI and human geography. The framework consists of four core components: (1) problem definition through stakeholder engagement, (2) knowledge and methodology integration, (3) prototype development and testing, and (4) knowledge dissemination and capacity building. Looking ahead, embracing human-centered approaches through the collaboration of human geographers and GIScientists will accelerate the design, development and adoption of advanced GeoAI technologies in exploring human-environment interactions and place-based decision support.

A61. Mitigating Perceptual Distortions in Virtual Reality Flood Geovisualizations

Andrew Mahisa Halim, Adelaide Genay, Martin Tomko

Abstract

Virtual reality (VR) is increasingly used to visualize and communicate characteristics of physical phenomena to non-expert audiences. Geovisualizations of water flow, in particular in flood studies, commonly assume that VR enables intuitive reconstructions that support understanding of flow scale and associated risks. However, VR environments are known to be prone to perceptual distortions. As a result, users’ perception of water flow velocity and depth in virtual spatial environments may not accurately reflect the characteristics of the real-world flow. A common approach to mitigating this issue is to display numerical values quantifying the 3D visualization, but such values require expertise to be meaningfully interpreted. Here, we explore an alternative approach based on the concept of concrete scales --- a technique which re-expresses physical measures as familiar objects (anchors) with well-known real-world dimensions. To inform the design of such re-expression for water flows, we first conducted semi-structured interviews to identify visual elements naturally used as anchors for water flows. Participants were shown videos of real water flows (including floods) and asked to estimate the depth and flow velocity of the water using only the visual cues present in the footage. We then replicated the identified anchors within a VR-based flood visualization prototype. Finally, we conducted a within-subject user study comparing the scale perception accuracy of anchor-based visualization against two baselines: (1) visualization without any additional contextual cues, and (2) with displays of raw numerical properties. Through this study, we demonstrate the potential of concrete scales to communicate flow scale more accurately in the presence of VR perceptual distortions. Additionally, we contribute a curated database of visual anchors as a practical design resource for future VR-based geovisualizations.

A62. Constructing a Sense of Place in Social Media

Sidney Wong

Abstract

While social media platforms are often perceived as placeless, recent research into social media platforms have identified place identities of a similar nature to real-world communities. Of interest to social scientists is how participants on social media platforms construct these place identities, also known as a sense of place, solely through their behaviour when spatial connections are lacking. In our investigation, we focus on one form of behaviour: language-use associated with a place (i.e., dialect). Our primary research question is how participants construct a sense of place on social media platforms through language-use. We took crowd-sourced dialect terms from Wiktionary for six inner-circle national varieties of English (Australia, Canada, Ireland, New Zealand, United Kingdom, and the United States). We then compared the distribution of these dialect terms on subreddits associated with these. Additionally, we analyse the distribution of these dialect terms associated with New Zealand English for six cities in New Zealand (Auckland, Hamilton, Tauranga, Wellington, Christchurch, and Dunedin). Our hypothesis is that if these place-based subreddits maintain a sense of place, then there should be a relationship between the presence of these dialect terms and their respective places. We take approaches from Natural Language Processing (NLP) to process and analyse our data. Our results show that there is a relationship between these dialect terms at both country-level and city-level geographies. Therefore, we can reject the null hypothesis that there is no relationship and that dialect terms partially contribute to the sense of place on social media platforms. Additionally, we show that the distribution of these dialect terms correlate with other demographic patterns when we combine our analysis with census and survey data. The findings from our study suggests that language-use can be used as a measure of place identity in the absence of other sociogeographic information.

A63. Semantic Graph-Driven Intelligent Recommendation of Similar Cases for Urban Regeneration

Shilong Wei, Shunli Wang, Minmin LI

Abstract

As cities transition from sprawling expansion to resource-constrained stock renewal, urban regeneration has become pivotal for enhancing the efficiency of urban resource utilization and unlocking latent urban potential. However, the spatial fragmentation and intricate interrelationships among current regeneration units hinder the availability of reference cases for informed decision-making. This study proposes a semantic graph-driven approach for the intelligent recommendation of similar cases in urban regeneration, grounded in the framework of ‘case representation – case extraction – case application.’ Anchored in the paradigm of geographical case-based reasoning, we develop a case representation model that captures both entities and their spatially constrained relationships, enabling a comprehensive expression of case features across relational and spatial dimensions. By integrating domain knowledge from urban planning and large language models, relevant case information is extracted from unstructured regeneration texts. This, in conjunction with multi-source geospatial data, supports the construction of a multidimensional knowledge graph for urban regeneration. A similarity computation model that incorporates temporal, spatial, and semantic attributes of cases is further designed to enable geographically grounded, quantifiable recommendations. Experimental validation based on urban regeneration documents and diverse geospatial datasets—such as land use, points of interest, and population distribution—demonstrates a joint entity-relation extraction accuracy of 0.874 and a recommendation precision of 0.81, substantiating the effectiveness of the proposed information extraction and multidimensional similarity matching method. Through systematic case modeling, efficient information extraction, and precise similarity assessment, this study significantly enhances the retrieval efficiency of urban regeneration cases, enabling intelligent recommendation of analogous cases and supporting spatial restructuring for high-quality urban development.

A64. A novel Geo-Bio Informatics framework to map Urban Stress: Integrating GPS Trajectories and digital Biomarkers

Yanjia Cao, Tianyu Li, Chi Yan Angus Leung

Abstract

Measuring and integrating GPS trajectories, digital biomarkers, and geographic context has been a persistent challenge in understanding human mobility and wellbeing patterns, both theoretically and technically. Current GIS methodologies demonstrate limited capacity to capture the dynamic interactions between urban environments and human stress responses across different spatial-temporal scales. To address these methodological gaps, we propose a novel Geo-Bio informatics framework that bridges geographic analytics with human stress responses. This framework extends traditional GIS-GPS analysis through three innovations: (1) the systematic integration of continuous stress biomarkers with GPS trajectories, (2) the development of multi-scale spatial-temporal analytical methods, and (3) the establishment of environmental exposure analytics through dynamic human-environment interaction modeling. This framework provides continuous monitoring of individual stress patterns and integrates with dynamic environmental characteristics, bridging the fundamental gap between stress responses and urban environmental exposures across multiple spatial-temporal scales. We conducted a case study in Hong Kong, involving 100 participants over a 7-day period. Data were collected by biosensor and GPS devices for biomarkers and GPS trajectories respectively. We first applied multi-dimensional Kernel Density Estimation (KDE) to generate initial spatial-temporal stress intensity surfaces, incorporating duration-weighted activity points to account for varying lengths of environmental exposure. Building upon these KDE results, we implemented ST-ResNet (Spatio-Temporal Residual Networks) to capture complex non-linear relationships and predict stress patterns. This deep learning approach integrated the KDE-derived stress surfaces with environmental characteristics, allowing us to identify significant geographic clusters of differential stress intensities across urban environments. The combined analysis revealed distinctive patterns of stress responses associated with various urban characteristics, particularly in high-density built environments. The proposed framework efficiently identifies stress-inducing locations and theoretical landscape, providing critical evidence for urban planning and public health interventions. It also lays the foundation for personalized exposure assessment and activity optimization, improving evidence-based stress management at individual level.

A65. Measuring the dynamics of healthcare accessibility by public transport in England: a comparison of scheduled and observed travel time variability

Zihao Chen and Federico Botta

Abstract

Fair access to essential services (e.g., healthcare) is fundamental to spatial justice and transport equity. Traditional accessibility studies typically focus on a single time point or part of the day and rely on average travel times derived from static schedules. This approach overlooks the temporal variability of travel time, and the uncertainty experienced in operation. Our previous work (Chen & Botta, 2025) developed a measure of public transport-based accessibility from the perspective of travel time variability (TTV) using nationwide bus timetable data (GTFS). We calculated hourly bus travel times from every small neighbourhood in England to the nearest hospitals and general practices (GPs) and summarised them using a TTV metric based on the standard deviation of hourly travel times. This revealed a geography of inequality characterised by a clear urban-rural divide, regional clustering of high and low TTVs, and spatial outliers. Here, we present new research that uses real-time bus operational data (GTFS-Realtime) to create empirical dynamic accessibility measures. Through a complex algorithm matching real-time vehicle positions to schedules, we construct a routable new schedule based on real bus departure/arrival times, which we can use to calculate observed travel times and TTV using an open-source routing engine (R5), accounting for delays and other operational issues. This enables a direct comparison between scheduled TTV and empirical TTV, offering a more realistic understanding of how observed TTV deviates from expectations. Our results show systematic gaps between theoretical and real-world TTV, often shaped by proximity to services. We identify clusters of underperforming areas where real-time variability significantly exceeds schedule-based estimates, and we investigate the inequalities in accessibility that arise from this. Our findings highlight the value of operational data in accessibility measurements, and their potential to improve accuracy. Finally, our results can be used for evidence-based policy making in transport and service planning.

A66. Promoting Citizen-Generated Data for Derived Products

Carolynne Hultquist

Abstract

Many national statistical offices struggle to track all of the 231 indicators of the United Nations Sustainable Development Goals (SDGs) and produce the underlying spatial data required to perform broad, consistent, and longitudinally reliable environmental monitoring. Opportunities arise to draw from citizen science data as communities connect to develop ways to monitor their environment. Technological advances enable citizen-led projects to collect environmental data from low-cost robust sensor networks and share data in real-time to monitor changing conditions. However, we are still in need of standards to promote the use of data generated by citizens for use in derived products. This talk on behalf of the Committee on Data (CODATA) Task Group on Citizen-Generated Data for the SDGs discusses our aims to amplify the values and contributions of communities for acknowledged use by a wide range of stakeholders including international organizations, humanitarian organizations, government agencies, businesses, and NGOs. We draw on experience in data science to develop reuseable and sustained data production from low-cost sensors, human digital actors, and community engagement.

A67. Measuring Regional Attractiveness in Europe Through Spatio-Temporal Mobility Patterns

Milad Malekzadeh, Tuomas Väisänen, Anastasia Panori, Olle Järv

Abstract

Regional attractiveness plays a crucial role in shaping spatial mobility, economic development, and population dynamics across Europe. Traditional approaches to measuring attractiveness rely on static indicators such as GDP, infrastructure, and amenities, providing an estimate of potential desirability rather than observed attractiveness. This study introduces a behaviorally grounded, spatio-temporal approach by analyzing realized human mobility flows across European NUTS2 regions, incorporating seven distinct types of movement: permanent migration, long-term and short-term student mobility, seasonal work, long-distance commuting, cross-border commuting, and multilocal living. By integrating heterogeneous data sources—including Labour Force Survey records, Erasmus mobility data, and Twitter-derived movement trajectories—we apply computational methods to quantify regional attractiveness. We operationalize this using two complementary approaches: an aggregate normalized net migration index and a Principal Component Analysis (PCA)-based attractiveness index. These spatially-explicit indicators capture both the intensity and diversity of regional inflows and outflows. Our results reveal strong spatial patterns, with urban centers consistently emerging as highly attractive, while regional anomalies, particularly in Nordic areas, highlight the significance of non-permanent mobility forms such as seasonal work and multilocal living. This emphasizes the limitations of potential-based indicators in understanding real-world mobility dynamics. By incorporating diverse mobility types and leveraging large-scale spatio-temporal data, this study advances movement analysis in regional science and supports more dynamic, data-driven approaches to modeling regional attractiveness. The findings have critical implications for spatial decision support, highlighting the need to integrate realized mobility data into regional planning and policy frameworks.