Coastal regions, integral to human economic and social frameworks, face increasing threats from disasters and climate change. This situation has made it necessary and a priority to study these areas at an international level. To be able to take precautions, protect, and manage coastal areas, it is essential to identify their coastal vulnerability. In this study, a coastal vulnerability analysis was conducted using best–worst method (BWM) in the Marmara Gulf Region, Türkiye. Moreover, comparison was made with the analytic hierarchy process (AHP) method in the subject of assessment of coastal vulnerability. According to the vulnerability map obtained with AHP, 17% of the region was determined to have a moderate vulnerability risk, 54% had a high risk, and 29% had a very high risk. On the other hand, the analysis results in BWM calculated the vulnerability percentages as follows: 18% moderate vulnerability, 57% high vulnerability, and 25% very high vulnerability. Visual and statistical comparisons revealed that the BWM method provides more consistent results and involves fewer pairwise comparisons than the AHP method. Thus, it offers ease of use and convenience to decision maker while maintaining relatively same level of weights for criteria. This study aims to lay the foundation for a dynamic system designed to assess coastal vulnerability, emphasizing usability for policymakers in decision-making and flexibility across various scenarios. Particularly, the adoption of the BWM provides notable benefits due to its direct approach and ability to yield more uniform and dependable evaluations, thereby efficiently addressing intricate decision-making challenges.

The need for accurate and up-to-date spatial data by decision-makers in public and private administrations is increasing gradually. In recent decades, in the management of disasters and smart cities, fast and accurate extraction of roads, especially in emergencies, is quite important in terms of transportation, logistics planning, and route determination. In this study, automatic road extraction analyses were carried out using the Unmanned Aerial Vehicle (UAV) data set, belonging to the Yildiz Technical University Davutpasa Campus road route. For this purpose, this paper presents a comparison between performance analysis of rule-based classification and U-Net deep learning method for solving automatic road extraction problems. Objects belonging to the road and road network were obtained with the rule-based classification method with overall accuracy of 95%, and with the deep learning method with an overall accuracy of 86%. On the other hand, the performance metrics including accuracy, recall, precision, and F1 score were utilized to evaluate the performance analysis of the two methods. These values were obtained from confusion matrices for 4 target classes consisting of road and road elements namely road, road line, sidewalk, and bicycle road. Finally, integration of classified image objects with ontology was realized. Ontology was developed by defining four target class results obtained as a result of the rule-based classification method, conceptual class definition and properties, rules, and axioms.

Undoubtedly, Unmanned Aerial Vehicles (UAVs) are one of today’s most functional technology products. In recent years, UAVs integrated with different sensors and transformed into harmony with advanced technology are developing rapidly and used in various applications to obtain spatial data. With UAV photogrammetry, images of target areas can be obtained quickly, at low cost, with high accuracy, and up to date. In addition to the advantages and success of existing methods for orthophoto production in large areas with classical photogrammetry, it is observed that the accuracy of these methods decreases in detecting changes in geometric properties, especially in small-scale areas. For this reason, obtaining geometric accuracy with the desired precision, which is of great importance in orthophoto production with UAV photogrammetry, has made it the basis for preference over classical methods. In this study, autonomous flights were carried out with DJI Mavic-2 Pro UAV in the selected pilot region, and orthophoto, Digital Elevation Model (DEM), and Digital Terrain Model (DTM) were produced as a result of processing the images obtained. To determine the geometric accuracy of the orthophoto, its coordinates were measured by the CORS-GPS method, and ten ground control points (GCP) were used. As a result of the accuracy analysis of the produced orthophoto, the spatial accuracy in the easting (ΔE) direction is ±6.9 cm, the spatial accuracy in the northing (ΔN) direction is ±7.8 cm, and the spatial accuracy in the height (ΔH) direction is ±10.3 cm.

Point cloud segmentation underpins various applications in geospatial analysis, such as autonomous navigation, urban planning, and management. Kernel Point Convolution (KPConv) has become a de facto standard for such tasks, yet its fixed geometric kernel limits the modeling of fine-grained contextual relationships—particularly in heterogeneous, real-world point cloud data. In this paper, we introduce the adaptation of a Local Contextual Attention (LCA) mechanism for the KPConv network, with reweighting kernel coefficients based on local feature similarity in the spatial proximity domain. Crucially, our lightweight design preserves KPConv’s distance-based weighting while embedding adaptive context aggregation, improving boundary delineation and small-object recognition without incurring significant computational or memory overhead. Our comprehensive experiments validate the efficacy of the proposed LCA block across multiple challenging benchmarks. Specifically, our method significantly improves segmentation performance by achieving a 20% increase in mean Intersection over Union (mIoU) on the STPLS3D dataset. Furthermore, we observe a 16% enhancement in mean F1 score (mF1) on the Hessigheim3D benchmark and a notable 15% improvement in mIoU on the Toronto3D dataset. These performance gains place LCA-KPConv among the top-performing methods reported in these benchmark evaluations. Trained models, prediction results, and the code of LCA are available in a GitHub opensource repository.