Efficient Adaptive Local Thresholding for Improving Elevation Change Detection in Urban Environments Using Bi-temporal LiDAR Data

Three-dimensional changes in urban environments have become one of the most critical topics in urban environmental studies and structural change monitoring. As cities rapidly expand and the number of construction projects continues to increase, identifying and monitoring these changes have gained significant importance. Furthermore, the ability to simulate and predict urban changes can help urban planners and decision-makers make informed choices. In this regard, the use of three-dimensional data, such as LiDAR point clouds, is an effective method for accurately simulating and detecting three-dimensional changes at the urban level. The primary objective of this research is to develop an algorithm for identifying and analyzing elevation changes in urban areas using bi-temporal LiDAR data processing techniques. These elevation changes can result from various factors, such as the construction and demolition of buildings, changes in land use, or variations in vegetation cover. In this study, LiDAR point cloud data from the years 2014 and 2019, collected from the coastal region of Duck in North Carolina, were utilized. These datasets contain three-dimensional points with X, Y, and Z coordinates, along with intensity values recorded from the Earth's surface using a LiDAR sensor. To identify elevation changes, the distance between the two point clouds was computed. The relative distance between corresponding points was determined using Delaunay triangulation, which created an irregular triangular network from one set of points and measured the distance between this network and the corresponding points in the other dataset. Following this step, a local adaptive thresholding method was applied to detect elevation changes at various scales. This method has the advantage of identifying localized changes more effectively than global detection techniques, which may overlook smaller variations. In this research, the two point clouds were merged into a single dataset to analyze elevation changes in urban environments. This process resulted in a unified point cloud with higher density in unchanged areas. In contrast, in areas where elevation changes occurred, multiple elevation surfaces emerged, leading to an increase in elevation variance. This effect was particularly noticeable in cases where land use had changed or where buildings had been demolished and reconstructed. The variance differences were clearly visible, providing strong indicators of real changes in the urban landscape. In regions experiencing changes, the elevation variance of the combined point cloud increased significantly, highlighting structural modifications. Conversely, in areas without elevation changes, the elevation variance remained low. This method proved especially effective in detecting minor changes that are often overlooked by conventional global change detection methods. To evaluate the accuracy of the proposed approach, key performance metrics such as completeness, correctness, overall quality, and the F1 score were computed. The results demonstrated that the proposed method performed exceptionally well in detecting elevation changes in urban areas. Specifically, in regions where changes in building heights or land use had occurred—such as the conversion of land into buildings or vegetation into urban structures—the algorithm successfully identified changes with high precision. In particular, for two subsets of the study area where building demolition and construction activities took place, the completeness metric exceeded 98%, while accuracy in other metrics ranged between 86% and 98%. For the third subset, where vegetation was converted into residential land use, completeness was measured at 85%, with accuracy in other metrics ranging between 83% and 98%. The local adaptive thresholding method introduced in this study effectively identifies elevation changes in complex urban environments. This technique is particularly efficient in detecting small-scale, localized changes that global methods may overlook. The results of this research have significant implications for urban planning, infrastructure monitoring, and disaster management, as they can enhance decision-making processes in these domains. Future studies should focus on applying this method to various urban environments while considering its computational efficiency and scalability for processing large-scale datasets. The integration of advanced machine learning models with this approach could further improve change detection accuracy and automation, leading to more efficient monitoring of urban transformations.