Abstracts Track 2026

Area 1 - Artificial Intelligence and Learning

Nr:	85
Title:	Detection of Drainage Ditches from LiDAR DTM Using U-Net and Transfer Learning
Authors:	Holger Virro, Alexander Kmoch, William Lidberg, Merle Muru, Wai Tik Chan, Desalew Meseret Moges and Evelyn Uuemaa
Abstract:	Accurate mapping of ditches is essential for effective hydrological modeling and land management. Traditional methods, such as manual digitization or threshold-based extraction, utilize LiDAR-derived digital terrain model (DTM) data but are labor-intensive and impractical to apply for large-scale applications. Deep learning offers a promising alternative but requires extensive labeled data, often unavailable. To address this, we developed a transfer learning approach using a U-Net model pre-trained on a large high-quality Swedish dataset and fine-tuned on a smaller localized Estonian dataset. The model uses a single-band LiDAR DTM raster as input, minimizing preprocessing. We identified the optimal model configuration by systematically testing kernel sizes and data augmentation. The best fine-tuned model achieved an overall F1 score of 0.766, demonstrating its effectiveness in detecting drainage ditches in training data-scarce regions. Performance varied by land use, with higher accuracy in peatlands (F1 = 0.822) than in forests (F1 = 0.752) and arable land (F1 = 0.779). These findings underscore the model’s suitability for large-scale ditch mapping and its adaptability to different landscapes.

Area 2 - Interacting with Data

Nr:	82
Title:	Effectiveness of Military Geoinformation Infrastructures Based on the Example of the "Geoserwer 2" System
Authors:	Tobiasz Wieczorek and Agnieszka Zwirowicz-Rutkowska
Abstract:	One of the characteristics that describes any information system is effectiveness, which is the most natural criterion for analyzing and evaluating systems in terms of their purpose and ability to meet specific needs in a field of application, as well as being a function of economy. The basic place of access to geospatial data for the Polish Armed Forces is the Geoinformation Portal of the Polish Armed Forces and the related Geographic Information and Services Server – "Geoserwer 2.0". This solution provides access to the resources of the Geospatial Reconnaissance Agency and Satellite Services (ARGUS). The goal of the entire system is to maintain a repository of digital geospatial information. The system provides digital geospatial products for all authorized users who conduct and coordinate military operations. The study presents the results of the system’s infrastructure evaluation from the users’ perspective in the following aspects: 1) use of infrastructure resources, 2) impact of the application on the tasks performed by users, 3) goals and activities of organizational units using infrastructure resources, 4) spatial data sets and services and technical support available in the geoportal. A multi-criteria method of assessing the effectiveness of spatial information infrastructures, based on an IT project, was used for the assessment, which is in line with the methodology of. SDI Multi-View Assessment Framework. Over 100 representatives of organizational units subordinate to the Minister of National Defense or supervised by him participated in the survey. The total "Geoserwer 2.0" project weighted score was 72.75 %. According to the established score interpretations for the method used in the study, projects with scores between 51 and 75% are good and quite effective, although their artifacts may need some improvements and additional analyses in the future. The results of the conducted research may contribute to the further development of the system as a tool supporting military geoinformation infrastructure. The collected information allows for adapting the functional layout to current needs and improving work ergonomics. Comments pointing to specific deficiencies in the functionality and availability of data will allow their correction and enable the range of available analytical modules to be expanded in the future. The results of the research are also presented in the broader context of building operational capabilities of the Polish Armed Forces.

Area 3 - Our Changing Planet

Nr:	75
Title:	Mapping Climate-Driven Dynamics of Atitudinal Vegetation Zones in the Southern Carpathians Using Multi-Source Remote Sensing and Topographic Climate Downscaling
Authors:	Iosif Lopatita, Patrick Chiroiu, Florina Ardelean, Olimpiu Pop, Andrei Ioniță and Petru Urdea
Abstract:	Mountain environments are particularly sensitive to ongoing climate change, as steep altitudinal gradients generate marked spatial contrasts in temperature, precipitation, radiation, and snow-cover persistence. In the Southern Carpathians, these gradients strongly influence the distribution of altitudinal vegetation zones, from deciduous and mixed forests to coniferous, subalpine, and alpine communities. Tracking recent changes in these zones is therefore essential for understanding how climate change is reshaping high-mountain ecosystems. This study develops an integrated GIS-based framework for analysing climate-driven dynamics of altitudinal vegetation zones in the Southern Carpathians. The approach combines supervised classification of multispectral satellite imagery with topographically explicit climate modelling to link vegetation zonation to fine-scale topoclimatic variability. PlanetScope is used as the main high-resolution dataset for vegetation-zone mapping, while Sentinel and Landsat imagery provide complementary multi-temporal observations. In parallel, spatially detailed climatic fields are derived from ERA5 reanalysis through topographic downscaling with TopoPyScale, allowing mountain-scale thermal and pluviometric heterogeneity to be represented more realistically. The classification framework distinguishes five main classes: deciduous forests, mixed forests, coniferous forests, subalpine vegetation, and alpine open areas. Preliminary results indicate that high-resolution multispectral imagery can reliably delineate the expected altitudinal structure of vegetation zones and capture ecotonal transitions with high spatial detail. Comparative tests using Support Vector Machine, Maximum Likelihood, and Random Trees classifiers suggest that Support Vector Machine produced the most coherent and least fragmented class patterns, particularly in transition zones where spectral separability is reduced. Random Trees showed strong flexibility in handling complex non-linear relationships and multi-source predictor sets, but, when applied to high-resolution Planet imagery, it may also increase local class heterogeneity and confusion in ecotonal areas characterized by strong within-class spectral variability. Preliminary downscaled climate results reveal pronounced altitudinal gradients and multi-decadal warming trends, providing the environmental basis for interpreting vegetation distribution under changing mountain climate conditions. Accuracy assessment is being carried out using confusion-matrix-based metrics, including overall, producer’s, and user’s accuracies. By integrating multi-source remote sensing with downscaled reanalysis data, the proposed framework provides a spatially explicit basis for detecting climate-sensitive vegetation patterns and ecotonal shifts in mountain environments. The results highlight the role of climate change in the ongoing reorganisation of altitudinal vegetation zones in the Southern Carpathians, with particularly evident effects in the climatically sensitive upper forest-limit ecotone.

Area 4 - Sensing

Nr:	74
Title:	Microtopographic Controls on Snow Depth Spatial Distribution across the Galeșu Rock Glacier (Southern Carpathians, Romania) Revealed by UAV-SfM Photogrammetry and Terrain Morphometry
Authors:	Andrei Ioniță, Flavius Sîrbu, Iosif-Otniel Lopătiță, Oana Berzescu, Nicolas Radu, Florina Ardelean, Petru Urdea and Alexandru Onaca
Abstract:	Accurate snow depth mapping in mountain terrain remains challenging because snow distribution is strongly controlled by wind exposure, slope configuration, surface roughness, and microtopographic complexity. These controls are especially important on rock glaciers, where ridges, furrows, depressions, steep fronts, and coarse blocky surfaces create strong small-scale variability in snow deposition, redistribution, and retention. In such environments, snow cover patterns depend not only on seasonal snowfall amount, but also on the interaction between local terrain morphology and snow-transport processes. This study presents a multi-temporal UAV photogrammetry workflow for monitoring snow depth over the Galeșu Rock Glacier, Retezat Mountains, Romania, based on repeated surveys acquired between 2023 and 2026. Multiple winter campaigns, mainly conducted in February and March, were processed using Structure-from-Motion techniques to generate high-resolution orthomosaics and digital surface models of snow-covered terrain. Snow depth was derived by differencing these models against a snow-free reference surface, enabling detailed mapping of snow distribution and its seasonal evolution across the rock glacier. The workflow was further designed to investigate how terrain morphology controls snow accumulation patterns. UAV-derived snow depth maps were coupled with morphometric analysis based on terrain metrics including slope, aspect, curvature, roughness, and topographic position. Particular attention was given to the influence of characteristic rock glacier microforms, especially furrows and ridges, on preferential snow deposition and storage. These elements act as local traps or exposed zones, generating marked contrasts in snow accumulation through wind redistribution and differential melt. Preliminary results from the first two surveys in 2023 reveal pronounced snow-depth heterogeneity, with values ranging from 0 m to over 3 m across significant portions of the landform, together with a clear increase in snow accumulation from winter to spring. Areas characterized by sheltered depressions, higher roughness, and negative topographic position generally retain deeper and more persistent snow, whereas exposed ridges and convex surfaces show reduced accumulation. In addition, this study also presents the UAV surveys acquired in 2024, 2025, and 2026, providing a multi-year perspective on snow deposition and accumulation under contrasting winter conditions. The UAV-based analysis is further complemented by correlation with a time-lapse camera installed on site, which offers continuous visual monitoring of snow deposition, persistence, and melt evolution throughout the snow season. This integration helps link discrete UAV surveys with the temporal dynamics of snow cover development and disappearance. Initial validation against in situ measurements indicated mean errors of -0.24 m in winter and -0.14 m in spring, with the highest accuracy observed over vegetation-free surfaces unaffected by canopy-related elevation bias. Statistical comparison between furrows and ridges demonstrates a clear microtopographic signal, with mean snow-depth differences of about 1.2-1.5 m between these landform types. Overall, the new results highlight the value of UAV-SfM photogrammetry for repeated snow monitoring in complex periglacial terrain and support its integration with terrain morphometry for high-resolution analysis of snow-topography interactions.

Nr:	89
Title:	K-Means Exploration of Desert Boundary in Google’s Alphaearth Embeddings for the South-Eastern Mediterranean
Authors:	Maxim Shoshany
Abstract:	Mapping desert boundaries is important for assessing their stability, expansion or contraction under the current Global climate and anthropogenic changes. While there is a distinctive number of studies which assessed changes in the extents of deserts utilizing climatic data or its NDVI surrogates, there are very few studies which aimed at detecting eco-geomorphic desert boundaries, as a step required prior to monitoring their areal change. Employing traditional eco-geomorphic surveys for this purpose over wide regions is time consuming, expansive and labour intensive. Recent developments in intelligent data generation offers global data bases synthesizing diverse spatial information sources. One of the leading examples concern the development of AlphaEarth self-supervised embeddings. These embeddings were extracted from optical, thermal, SAR, DEM, LIDAR, climate, gravity, land-cover, and geocoded text, thus represent unprecedent synergetic patterns of most diverse terrain and climate information for wide regional/ Global extents. This data base may facilitate empirical exploration of desert border phenomenology which had not reached yet common conceptualization and agreed description. While there are scholars who claim that deserts are border less and bounded by wide transition zones, there are phytogeographic mappings of distinctive plant zones between sub-humid and arid regions. A desert border line if exists is an environmental threshold which may evolve due to the presence of sharp transition in eco-geomorphic extrinsic factors, a non-linear change in such extrinsic factors or by the formation of tipping points. The multi-source patterns emerging in the AlphaEarth embeddings represent implicitly different combinations of plant and soil properties at high spatial detail which may potentially allow detection of such threshold zones. However, exploration for a geographical entity which was not described yet for wide desert extents carry significant methodological challenges. Hypothesizing the existence of a border line imply a partition between desert and non-desert regions. Implementing unsupervised classification technique may be instrumental for revealing such partition. However, classification parameterizations and especially the number of classes may shift border delineations. Our operational hypothesis is that analysis of results from multiple classifications with different number of classes may allow detection of consistent partition between desert and followingly the identification of a desert border line. The approach was implemented for the Mediterranean to arid transition across the South-Eastern Mediterranean. We tested K-Means classifications with 6, 9 and 12 classes and found that the classifications utilizing 9 and 12 classes yielded almost the same boundary line between the desert and non-desert terrain of this region.

Area 5 - Spatial Data Analysis and Management

Nr:	92
Title:	Modeling Spatial Variation across Data Distribution: A Moran Eigenvector-Based Spatially Varying Quantile Regression Approach
Authors:	Zhan Peng and Ryo Inoue
Abstract:	Spatial relationships between observed phenomena and their driving factors often vary not only across geographic space but also across different levels of the conditional distribution of the response variable. Existing approaches usually address only one of these dimensions or rely on restrictive assumptions, such as a common spatial scale for all coefficients, while local-regression and Bayesian quantile methods can become computationally demanding for large spatial datasets. This study proposes a Moran eigenvector-based spatially varying quantile regression (MSVQR) model to analyze spatial heterogeneity across data distribution. MSVQR combines conditional quantile regression with Moran eigenvector spatial filtering so that regression coefficients can vary simultaneously by location and quantile. Each coefficient consists of a global effect and a spatially varying component represented by Moran eigenvectors. An L1-regularization scheme selects the eigenvectors relevant to each explanatory variable at each quantile. This enables the model to determine whether spatial variation exists and, when it does, to identify its spatial scale. A bootstrap procedure is further introduced for statistical inference on location-specific estimates. The proposed model is evaluated through simulation experiments and an empirical application to rental apartment data from the Tokyo 23 Wards, Japan. In the simulations, MSVQR is compared with an existing approach under settings with different spatial scales and distributional variation. The results show that MSVQR attains comparable accuracy in recovering spatially varying coefficients, while more reliably identifying spatially invariant effects and offering better computational efficiency for large samples. The experiments also indicate that using a more parsimonious candidate set of Moran eigenvectors can improve performance at the tails of the distribution by reducing over-localized estimates. In the case study, MSVQR is applied to 20,029 rental apartment observations to model rent per square meter as a function of property size, age, floor level, and walking time to the nearest station. Compared with conventional conditional quantile regression, the proposed model improves goodness of fit and reduces residual spatial autocorrelation, demonstrating the importance of accounting for spatial heterogeneity in quantile-based spatial analysis. The results further reveal meaningful spatial and distributional differences in the housing market: the relationship between property size and unit price is negative in lower-priced segments but turns positive in some high-end areas; age has a consistently negative and floor level has a broadly positive association with unit price; and accessibility shows generally negative yet locally unstable effects. Overall, MSVQR provides a scalable tool for spatial data analysis and is applicable to other environmental, transportation, and socio-economic problems in which relationships vary across both geography and data distribution.

Nr:	40
Title:	Modelling Historical Events with Geographic Information Technologies and Archaeological Evidence: GIS-Based Multi-Criteria Analysis and Automatic Route Tracing
Authors:	Pascual Perdiguero Asensi, Alfredo Ramón Morte, María Estela García Botella, Fernando Llorens Cobos, José Luis Martínez Boix and Feliciana Sala Sellés
Abstract:	The Roman general Scipio began a seven-days forced march in 209 BC, the objective was the fast conquest of the enemy Barcid capital, Qart Hadasht (present-day city of Cartagena – Spain). This military mission was definitive for the victory in the Second Punic War, for the domination of the Iberian Peninsula and the supremacy of the Roman Empire in the known world. Traditionally, it was thought that the beginning of this campaign took place at the mouth of the Ebro River, but the archaeological evidence and the use of GIS-based multi-criteria analysis for processing terrain data, ancient landscapes, historical communication routes, historical maps, and archaeological evidence has allowed us to place the starting point at the mouth of the Sucro River (present-day Júcar River). This was achieved by using Dijkstra's algorithm to trace the route followed by the Roman general's troops and even locate their marching camps. This summary explains how the multi-criteria analysis GIS-based has been carried out, through weighted direct linear combination (MCDA), whose objective is to obtain a final raster layer, with the integrated valuation of the mobility costs and the low-cost candidate pixels, with which to calculate the layout of the Roman troops, the forced marching itineraries and possible locations of the marching camps, with the least route calculation (LCP) add-on from QGIS. The initial results of this research have allowed for the modelling of the final forced march, prior to the capture of Qart Hadasht, and to locate the last marching encampment of the Roman legions before the assault. The methodology applied confirms its usefulness and could be easily implemented to resolve historical or archaeological hypotheses, with potential applications in other scientific fields.

Nr:	76
Title:	Making Geoinformation Accessible and Powerful: The Impact of New Standards
Authors:	Jordi Pallàs del Rio and María Amparo Nuñez Andres
Abstract:	The adoption of new standards defined by the Open Geospatial Consortium (OGC), particularly the OGC APIs, is transforming the way spatial information is accessed, shared and consumed across different platforms and applications. These interfaces are designed recently to follow web-native principles, enabling simpler integration with contemporary technologies and facilitating the development of interoperable geospatial services. They are recommended to access to the information from the European Union high data value (HDV) Regulation. This study presents the design and implementation of an architecture that integrates the OGC API – Features and OGC API – Processes standards to provide interoperable, scalable and web-accessible services for managing and processing geospatial data. The proposed architecture is based on a modular approach that allows the publication of spatial datasets through standardized APIs while enabling the execution of geospatial processing tasks in the cloud. Through an applied use case, the study demonstrates how these technologies enable users to perform spatial queries and geospatial processes without requiring specialized geographic information systems (GIS) software. This significantly broadens the accessibility of geospatial information and facilitates its use by a wider range of users, including decision-makers without advanced technical expertise. The developed solution includes a customized backend that incorporates additional functionalities, improving efficiency and enhancing the use of geoinformation to support decision-making in a more agile, transparent, and efficient manner. This approach is based on authoritative geographic data and promotes data reuse. The results highlight the strong potential of OGC API-based architectures to improve accessibility, automation, and efficiency in the management and exploitation of spatial data. Complementing material: https://ocs.editorial.upv.es/index.php/CIGeo/CiGeo2025/paper/view/19829.

Nr:	77
Title:	Predicting Cycling Infrastructure Development Based on Land Cover Forecasting and GIS Spatial Analyses: A Case Study of Lublin, Poland
Authors:	Wojciech Dawid and Bartosz Kubicki
Abstract:	Urban cycling infrastructure is increasingly recognized as a key component of sustainable city planning, yet it remains underdeveloped relative to road and pedestrian networks in many cities. The growing urgency to reduce greenhouse gas emissions, alleviate urban traffic congestion, and improve road safety has elevated cycling as a viable alternative to private car use and public transportation. However, successful expansion of cycling networks requires careful consideration of spatial, demographic, and environmental factors to ensure new routes are practical and aligned with long-term urban development trajectories. This study presents a methodology for predicting the spatial distribution of new cycling routes by 2030, integrating land cover forecasting, demographic trend analysis, and GIS-based network analyses, applied to Lublin, Poland — the largest economic, academic, and cultural centre in eastern Poland, with approximately 308,000 residents and an existing cycling network of 203.7 km as of December 2023. The research utilized open-source spatial data and free GIS software, including the MOLUSCE plugin in QGIS, employing an Artificial Neural Network – Multi-Layer Perceptron (ANN-MLP) to simulate future land cover changes. Explanatory variables included elevation, terrain slope, population density, proximity to roads, private land areas, and water bodies. Validated against CORINE Land Cover data, the model achieved 95.3% accuracy and a kappa coefficient of 0.94. Predictions for 2030 indicate urban expansion of 3.65 km², primarily at the expense of agricultural land decreasing by 3.54 km², while vegetation, wetlands, and water bodies remain relatively stable. Demographic trends across 27 administrative districts were assessed using a standardized registration index from 2018 to 2024, revealing an overall population decline of 5%, with divergent trajectories between individual districts. Data were normalized using a 50-metre hexagonal grid, enabling consistent spatial integration. Origin-Destination matrix analyses identified optimal primary and supplementary cycling corridors connecting residential, educational, commercial, and transport nodes. The results identified 5,683 hexagonal cells suitable for cycling infrastructure, representing nearly a quarter of the city's area. Proposed routes were prioritized into four categories based on projected urban growth and demographic potential. While 51% of primary routes are already constructed, only 13% of supporting routes are complete, highlighting connectivity gaps in growing areas. Urban expansion will require modifications to existing routes and development of new ones, with approximately 6.5 km² of primary and 23 km² of supporting paths proposed.

Nr:	80
Title:	A Reproducible GIS Framework for Nationwide PV Site Identification From Heterogeneous Spatial Data
Authors:	Sina Keller and Philipp Keller
Abstract:	Scaling up ground-mounted photovoltaic (PV) capacity is a central component of Germany’s energy transition. In 2025, the country reached a total installed solar capacity of 117 GW, with ground-mounted installations accounting for approximately half of the 16.4 GW added that year. Achieving the national target of 215 GW by 2030 will require an annual deployment rate of approximately 20 GW. Identifying suitable sites at the necessary scale poses a significant spatial planning challenge, as it requires integrating and consistently processing heterogeneous geospatial datasets spanning multiple administrative levels. This study presents a reproducible geospatial framework for the nationwide identification of potentially suitable areas for ground-mounted PV installations in Germany. The framework addresses two key challenges: first, the harmonization of heterogeneous spatial datasets, including land cover, topography, infrastructure, and environmental protection areas from federal and state authorities, into a unified geospatial database; and second, the automated implementation of a stepwise spatial exclusion methodology that systematically removes areas incompatible with PV development. The exclusion process integrates legal and planning constraints and applies category-specific buffer zones derived from regulatory requirements and scientific literature. The entire workflow is implemented in Python within a PostgreSQL/PostGIS environment and relies exclusively on openly available geospatial datasets and open-source software. This enables automated, scalable, and fully reproducible processing of large-scale spatial datasets at the national level. This work presents a fully reproducible nationwide workflow for PV site identification based entirely on open geospatial data. When applied to the entire German territory, the workflow reveals a considerable spatial potential for ground-mounted PV development. Preliminary analysis indicates that a significant share of these areas overlaps with agricultural land, highlighting the need for further evaluation considering land-use type, soil quality, and estimated energy yield. Additionally, proximity to infrastructure, particularly road access and grid connection points, plays a key role in determining both the technical feasibility and economic viability of potential sites. The proposed framework demonstrates how open data and open-source technologies can support transparent, scalable, and reproducible renewable energy planning. Its modular design allows the adjustment of exclusion criteria, buffer distances, and input datasets to different regulatory contexts, making the approach transferable to other national settings and adaptable to evolving policy and regulatory frameworks. Furthermore, the framework is designed to be extensible, enabling the integration of additional evaluation layers such as energy yield modeling, grid capacity constraints, or land-use conflict indicators.

Nr:	93
Title:	Construction of a Discrete Choice Model with Moran Eigenvector-Based Spatially Varying Coefficients
Authors:	Zhan Peng, Ryo Inoue and Yu Ito
Abstract:	Policy evaluation in urban planning requires quantifying how policies affect individual behavior. Traditional multinomial logit (MNL) models assume constant coefficients across space, failing to account for “spatial heterogeneity” where decision-making varies by region. While Geographically Weighted Multinomial Logit (GWMNL) models address this issue, they suffer from high computational costs and multicollinearity. This study proposes an MSVC-MNL model using Moran eigenvectors (ME) to represent spatial variation efficiently. The MSVC-MNL model incorporates spatially varying coefficients into a MNL framework. Based on random utility theory, the utility for an individual at a specific location choosing an alternative is expressed as a function of the alternative's attributes, the individual's attributes, locational characteristics, and random errors. The linear sum of ME is introduced to the regression coefficients to capture local spatial effects. The model was tested using Tokyo Metropolitan Area person-trip data in 2018, comparing it against MNL, Spatial Latent Class (SLC) models, and GWMNL models. In terms of accuracy, MSVC-MNL achieved the highest hit rate among all tested models. Additionally, the spatial distribution of rail distance coefficients estimated by MSVC-MNL reflects more regional variations than that shown in GWMNL. The results also revealed that mode choice in suburban areas is highly sensitive to travel distance and rail-related origin characteristics have lower impacts in suburbs where access to rail service is poor. The MSVC-MNL model provides superior predictive performance and highly interpretable results regarding regional transportation conditions. While effective, the improvement in accuracy for transport modes with low usage rates remains a challenge.

Area 6 - Spatial Data Mining and Computation

Nr:	87
Title:	Behavioral Interpretation of Unlabeled Destination Points in Emerging Residential Areas Based on Vehicle Trajectory Data Independent of POIs
Authors:	Hayato Yamaguchi, Hayato Yamaguchi, Kota Sata and Hayato Goto
Abstract:	Vehicle trajectory data collected through in-vehicle communication modules (DCM) and telematics systems provide fine-grained, long-term observations of real-world automobile use and have become an important data source for transportation analysis and carbon-neutral policy evaluation. A central challenge in utilizing such data is the lack of explicit trip purpose information. Most existing studies address this limitation by linking destinations to external Point of Interest (POI) databases and inferring purposes from POI attributes. However, in practice, a non-negligible share of destinations cannot be matched to any POI due to database incompleteness, spatial offsets between parking and activity locations, private residences, or newly developed facilities. These POI-unmatched destinations are typically labeled as ``unknown'' and excluded from further analysis, leaving their behavioral meaning largely unexplored. This study reframes these unlabeled (POI-missing) destination points not as noise, but as potentially meaningful activity locations that reflect everyday mobility behavior. We propose a POI-independent analytical framework that interprets destination usage patterns solely from behavioral features derived from vehicle trajectory data. Focusing on an emerging, car-dependent residential area in Mishima City, Japan, we analyze two months of DCM trajectory data collected in 2024 from 274 private passenger vehicles. For each destination cluster, we extract six spatiotemporal behavioral features: median dwell time, visit frequency, primary purpose rate, night arrival rate, weekend arrival rate, and single-purpose trip rate. Using these standardized features, we apply hierarchical clustering with Ward's method to identify common behavioral patterns among destinations. The analysis yields six distinct clusters representing qualitatively different destination usage types, ranging from short auxiliary stops and brief residential visits to routine single-purpose errands and high-frequency, long-duration work or school locations. Importantly, POI-unlabeled destinations are not randomly distributed across clusters. Instead, they are concentrated in specific behavioral categories, such as short neighborhood visits, brief evening stops, routine standalone trips, and even daily work or school destinations that are missing from POI databases. This demonstrates that many unlabeled destinations correspond to systematic and interpretable forms of daily activity rather than exceptional or anomalous behavior. By uncovering the latent behavioral structure of unlabeled destinations, this study complements conventional POI-based trip purpose inference and provides a more complete picture of everyday automobile use. The findings highlight the diversity of short-distance and routine car trips that are often overlooked in aggregate travel statistics, particularly in suburban contexts. From a policy perspective, the proposed framework enables behavior-specific insights that can inform targeted carbon-neutral mobility strategies, such as prioritizing electrification for high-frequency commuting trips or promoting alternative modes for short local movements. Overall, this work demonstrates the value of embracing POI-independent behavioral analysis to fully exploit vehicle trajectory data and to support more nuanced and effective transportation decarbonization policies.