Sat2building: Lod-2 Building Reconstruction from Satellite Imagery Using Spatial Embeddings

The main objective of the study is Digital Terrain Model (DTM) generation from aerial photogrammetric data and identify and map the potential avalanche prone zones in Manali region. Avalanche is a dynamic hazardous phenomenon in the snow-bound mountainous terrain. Mapping of avalanche prone terrain is crucial to minimize the avalanche hazard. Nowadays, airborne data capturing technology, such as large-format Photogrammetry, has opened new vistas for the mapping of complex and inaccessible mountainous areas. In the present study, large format digital Photogrammetry data of 20 cm ground sample distance (GSD) have been used to generate high-resolution and accurate Digital Elevation Model and ortho-images. Digital terrain model along with its derivative terrain products and land cover map generated from land cover classification of derived ortho-photo is analyzed to locate the probable avalanche zone. The terrain characteristics, snow-pack condition and prevailing meteorological conditions are the groups of variables that influence the occurrence of avalanche. Amongst these, the terrain characteristics is the most influencing factor, and easier to map due to its stable nature along the time. Therefore advanced geo-informatics techniques have been used by mixing terrain property, Digital Elevation Model (DEM) and satellite imagery to determine the different geographical factors that affect the avalanche triggering. Also the derived information was combined in Analytic Hierarchy Process to extract a map of the avalanche prone zones in the study area standard mapping techniques as coarse-resolution data are not very appropriate for such studies.

Urban waterlogging is a serious problem in fast urbanizing regions, exacerbated by climatic variations and poor drainage facilities. This paper situates Unmanned Aerial Vehicle (UAV) and Light Detection and Ranging (LiDAR) remote sensing methods as viable tools for assessing waterlogging in the Dwarka-Khaira corridor, New Delhi, India. Conventional terrestrial field monitoring is restricted by resource constraints and scale, whereas high-resolution aerial methods provide detailed hydrological information. UAV and LiDAR deliver excellent spatial resolution and vertical accuracy necessary for the identification of microtopographic features critical for detecting hydrological restrictions. The RMSE of the UAV-derived DEMs was highly reduced to 4.6 m through post processing, with an increased accuracy of 83.66% and was conducive for a good performance in hydrological modeling. UAV data, which had ground sample distance less than 5 cm, facilitated urban feature classification with an accuracy of 98%) and helped to mitigate spectral confusion and misclassifi cation observed from Landsat (30 m) data. High-resolution aerial data therefore minimized false-positives and enhanced network extraction quality as compared to their satellite based counterparts. But while satellite imagery is not well matched for development level analysis, it continues to offer useful potential for regional assessment of vulnerability. Combining data from both sources is consistent with evidence from similar research, in support of models based on Sustainable Development Goal resilience targets. The application of these enhanced methods ed to a 30-40% decrease in identified waterlogged areas. So, the study offers a scientific basis for urban water management policies and planning.

Automatic 3D reconstruction of urban scenes from stereo pairs of satellite images remains a popular yet challenging research topic, driven by numerous applications such as telecommunications and defense. The quality of reconstruction results depends particularly on the quality of the available stereo pair. In this paper, we propose an operational pipeline for large-scale 3D reconstruction of buildings from stereo satellite images. The proposed chain uses U-net to extract contour polygons of buildings, and the combination of optimization and computational geometry techniques to reconstruct a digital terrain model and a digital height model, and to correctly estimate the position of building footprints. The pipeline has proven to be efficient for 3D building reconstruction, even if the close-to-nadir image is not available.

Visual navigation based on image matching has become one of the most important research fields for UAVs to achieve autonomous navigation, because of its low cost, strong anti-jamming ability, and high performance. Currently, numerous positioning and navigation methods based on visual information have been proposed for UAV navigation. However, the appearance, shape, color, and texture of objects can change significantly due to different lighting conditions, shadows, and surface coverage during different seasons, such as vegetation cover in summer or ice and snow cover in winter. These changes pose greater challenges for feature-based image matching methods. This encouraged us to overcome the limitations of previous works, which did not consider significant seasonal changes such as snow-covered UAV aerial images, by proposing an image matching method using season-changing UAV aerial images and satellite imagery. Following the pipeline of a two-channel deep convolutional neural network, we first pre-scaled the UAV aerial images, ensuring that the UAV aerial images and satellite imagery had the same ground sampling distance. Then, we introduced attention mechanisms to provide additional supervision for both low-level local features and high-level global features, resulting in a new season-specific feature representation. The similarity between image patches was calculated using a similarity measurement layer composed of two fully connected layers. Subsequently, we conducted template matching to estimate the UAV matching position with the highest similarity. Finally, we validated our proposed method on both synthetic and real UAV aerial image datasets, and conducted direct comparisons with previous popular works. The experimental results demonstrated that our method achieved the highest matching accuracy on multi-temporal and multi-season images.

In recent years, research on increasing the spatial resolution and enhancing the quality of satellite images using the deep learning-based super-resolution (SR) method has been actively conducted. In a remote sensing field, conventional SR methods required high-quality satellite images as the ground truth. However, in most cases, high-quality satellite images are difficult to acquire because many image distortions occur owing to various imaging conditions. To address this problem, we propose an adaptive image quality modification method to improve SR image quality for the KOrea Multi-Purpose Satellite-3 (KOMPSAT-3). The KOMPSAT-3 is a high performance optical satellite, which provides 0.7-m ground sampling distance (GSD) panchromatic and 2.8-m GSD multi-spectral images for various applications. We proposed an SR method with a scale factor of 2 for the panchromatic and pan-sharpened images of KOMPSAT-3. The proposed SR method presents a degradation model that generates a low-quality image for training, and a method for improving the quality of the raw satellite image. The proposed degradation model for low-resolution input image generation is based on Gaussian noise and blur kernel. In addition, top-hat and bottom-hat transformation is applied to the original satellite image to generate an enhanced satellite image with improved edge sharpness or image clarity. Using this enhanced satellite image as the ground truth, an SR network is then trained. The performance of the proposed method was evaluated by comparing it with other SR methods in multiple ways, such as edge extraction, visual inspection, qualitative analysis, and the performance of object detection. Experimental results show that the proposed SR method achieves improved reconstruction results and perceptual quality compared to conventional SR methods.

The purpose of this article is to investigate modern approaches, methodologies, and principles in the application of geoinformation systems (GIS) in urban planning decisions. The study provides an in-depth analysis of the functional capabilities of GIS and emphasizes their integration with complementary technologies such as remote sensing, global positioning systems (GPS), and building information modeling (BIM). By consolidating diverse spatial data sources, GIS facilitates comprehensive analysis and visualization of urban environments, thereby supporting informed decision-making in urban development. Recent technological advancements have enabled the creation of dynamic digital urban models that simulate various planning scenarios. These models allow urban planners to assess infrastructure requirements, optimize land use, and predict future growth trends. The integration of real-time data via cloud-based platforms further enhances the responsiveness and accuracy of urban planning processes, contributing significantly to sustainable urban development. The integration of data from numerous sources - satellite imagery, aerial photography, cadastral registers, GPS measurements, and remote sensing results - allows us to create a comprehensive digital model of urban space that provides a sound analysis of spatial processes. Integrated geographic information systems open up new horizons for urban planning and management, providing operational monitoring and accurate forecasting of changes in the urban environment. Further research should focus on optimizing data analysis algorithms and developing flexible information platforms with the introduction of artificial intelligence and the Internet of Things technologies, which will significantly expand the capabilities of modern urban planning solutions. Such a synergistic approach allows identifying potential problem areas, optimizing the location of infrastructure facilities, and developing recommendations for modernizing urban planning policy. The research findings confirm that geoinformation systems are an essential tool in modern urban planning. They not only improve the precision of spatial analysis but also contribute to the efficient allocation of resources and the formulation of proactive strategies to tackle urban challenges. Consequently, the use of GIS is instrumental in promoting sustainable growth and enhancing the quality of life in urban areas.

Introduction:The observed presence and orientations of fractures on boulders on the asteroids  (101955) Bennu and (65803) Didymos-Dimorphos have been instrumental in establishing the role of thermal fatigue and fragmentation on asteroids (Molaro et al., 2020, Lucchetti et al. 2024). With cratering processes, thermal fatigue is hypothesized to be one of the primary mechanisms forming regolith on small asteroids (Delbo et al., 2014).  However, lighting and viewing conditions can introduce ambiguity in the orientation and number of surface lineaments  that are identified on an asteroid’s surface (Buczkowski et al., 2008). Similarly, lighting conditions can affect the identification and orientation of fractures observed on boulders. Therefore, any unidentified bias in determining the orientation of these boulder fractures could influence any conclusions on the importance of thermal fatigue relative to cratering.To understand what imaging bias might exist, and to determine a way to correct for it, we explore systematically how lighting influences measurement of fracture orientations on asteroid boulders. We render several local digital terrain models (DTMs) of boulders on Bennu generated using data collected by the OSIRIS-REx Laser Altimeter (OLA; Daly et al., 2020, 2017). The modeled boulders were chosen because they possess numerous linear structures on their surfaces that are well captured by the high-resolution, 5 cm per pixel ground sample distance (GSD) of the OLA-derived DTMs. The boulders considered in this study range in size from 6–33 m in diameter.Figure 1. Setup used for rendering surface boulders observed by OLA on Bennu. Camera is looking straight down on surface with an elevation on 90°Methodology:In this analysis, each local boulder DTM had its slope de-trended, such that the boulder was sitting on a flat-horizontal plane.  Images of the boulder were rendered using a ray tracing algorithm and a modified Lommel-Seeliger photometric function identical to what was done in Daly et al., (2022). The camera was always placed at an elevation angle of 90° (0° emission) and a range of 1 km from the surface, looking directly down at the boulder (Figure 1). The modeled camera had 1024x1024 pixels with a focal length of 500 mm and an instantaneous field of view (IFoV) of 4x105 radians, resulting in images with a ground sample distance of 4 cm/px. These camera properties were picked so that the resulting image pixel scales were similar to the GSD of the DTMs. The boulder DTMs were rendered using a range of solar azimuths (α of 0° to 360°) and elevations (β -30° to 30° via 90°) (Figure 2). Images were rendered every 30° in α and 20° in β. Figure 2. Boulder images rendered for differing sun azimuths α and elevations β. The image with the yellow outline indicates the observing conditions that existed during the DART impact (Daly et al., 2023).Members of our team were then tasked with mapping lineaments on a subset of the synthetic images. Each investigator was only given images for one lighting condition when they undertook their lineament assessment, in order to minimize any bias that might have arisen by having more than one viewing orientation. They were also given a downsampled and smoothed version of the boulder DTM so that they could map lineaments on the rendered images in the Small Body Mapping Tool (SBMT, Ernst et al. 2018) if they so desired. Once all the rendered images were mapped and the result turned in, each investigator was given a full resolution boulder DTM with all of its renderings, so that a complete set of lineaments could be identified. These were then compared to the lineaments obtained from the individual rendered images. The results for any biases were then assessed.Preliminary results:The mapping is ongoing, and a full report of our findings will be presented at EPSC 2024. However, some preliminary findings can be mentioned. As is already well-known, the morphology of surface features, including boulder fractures, are not easily recognized when the sun elevation is high (immediately overhead): features get washed out. Images taken with low sun β (

<p>Microcatchments (<10 ha) are often used to monitor the effect of disturbances or restoration on the hydrological functioning of peatlands. Catchment areas serve as the spatial limits within which fluvial processes are studied and topographic parameters are derived. Digital Elevation Models (DEMs) and Digital Surface Models (DSMs) are used in standard practice to delineate the watershed boundary of microcatchments. These digital representations of topography contain errors, meaning the derived catchment areas have inherent uncertainty. The low-slope and hummock/hollow nature of peatland surfaces mean catchment delineation is particularly sensitive to vertical errors, so understanding the potential effect of uncertainty on the accuracy of catchment delineation is essential to providing a reliable account of peatland microcatchments hydrology.</p><p>This paper investigates the sensitivity of catchment delineation to DEM/DSM error for 30 peatland microcatchments across the Peak District National Park, UK. To evaluate the suitability of DSMs for hydrological applications in peatlands, a 0.25m photogrammetric DSM is used and evaluated against a 1m LiDAR DEM. Monte Carlo simulation is applied to produce a range of realisations of the DSM and DEM within their vertical error margins, from which a range of catchment areas are calculated. The variability of the watershed boundaries of each catchment is evaluated in the context of local gradient, difference from mean elevation and extent of gullying, to determine the relationship between terrain characteristics and variability in catchment delineation. Findings will have implications for the generation of catchment areas in peatland hydrology.  </p>

One of the most important parameters when formalising the specifications for a photogrammetric project is the ground sample distance (GSD). While there are a variety of methods for the computation of the mean GSD for an aerial image, the aim of this research paper is to propose a method for the computation of the GSD that balances both precision and speed of computation. The rigorous computation of the mean GSD on an image requires the evaluation of the ground projection for each pixel over the digital terrain model. Three large photogrammetric projects within the Spanish National Plan for Aerial Orthophotography (PNOA) are evaluated to determine the degree of fulfilment of the required GSDs. The statistical distribution of the GSD is analysed to confirm its (lack of) normality. The rigorous approach to compute the mean GSD was presented, tested and compared with ten approximate solutions based on regular patterns of GSDs. The results achieved with the mean GSD value based on two object points with maximum and minimum ground elevations per image are also presented as well as the base/height ratios achieved on the three large PNOA projects. It is empirically demonstrated that the computation of the mean GSD based on a pattern of 3 × 3 GSDs is recommended. An error better than 1 cm is achieved for 99.8 % of the images, independent of the topography of the mission.

Photogrammetric digital surface models and orthoimagery for the continuous coastline, Wales to Platinum, Alaska

Abstract. Landscapes have been photographed dozens of times at scales ca. 1/25,000 and better since World War II. Scans are distributed freely online (e.g. remonterletemps.ign.fr). In parallel, Structure-from-Motion (SFM) software made photogrammetric processing easy to non-specialists. Yet puzzling questions crop up to use both: (i) Can raw scans be used as is? (ii) Can Ground Control Points (GCP) and checkpoints be safely collected from a web portal? (iii) How many parameters are sufficient for camera interior orientation? (iv) Are single flight camera networks sufficient to constrain camera models compared to multiple flights? (v) Are photogrammetric Digital Surface Models (DSM) fit for quantifying landslide activity? Processing of scanned black-and-white 1/27,000 photographs from IGN flown in May 1978 over Cirque de Salazie in La Réunion Island answer these questions. We find that raw scanned photographs need translation, rotation and cropping to match the camera reference frame. GCP and Check point coordinates collected on geoportail.gouv.fr with assumed accuracy of 10 m, achieved ca. 7 m accurate SFM registration. The optimal camera model uses only 4 parameters: f, cx, cy and K1. Compared to a 2015 lidar Digital Terrain Model (DTM), the 0.66 m/pixel DSM of 1978 has a median deviation of −1.39 m ± 3.34 m (Median Absolute Deviation) which is comparable to GCP quality. Elevation difference more importantly reveals, for the first time, the 37 years and 13 cyclones cumulated landslides pattern on Cirque de Salazie. Photographic archives hold decades-long 3D history. SFM is a game changer for landslide risk mitigation planning.

Forest variable estimation using a high-resolution digital surface model

Aerial photographs of the coastline from Nightmute to Kipnuk were collected on August 11, 2015 and September 9, 2015. The photographs were processed using Structure-from-Motion (SfM) photogrammetric techniques. Global Positioning System (GPS) checkpoints were collected via a Global Navigation Satellite System (GNSS) survey conducted between August 15 and September 14, 2015. For the purposes of open access to elevation and orthoimagery datasets in coastal regions of Alaska, this collection is being released as a Raw Data File with an open end-user license. Orthoimages contain 4-band, 8-bit, unsigned raster data (red/green/blue/no data; file format-GeoTIFF; source-Fairbanks Fodar) with 20 cm GSD. No-data values are found in band 4 or shown as transparent in red/green/blue images. The file employs Lempel-Ziv-Welch (LZW) compression. Light exposures in the orthoimages are a result of daily weather conditions, which ranged from low cloud cover, rain, and full sun.

Aerial photographs of the coastline from Newtok to Toksook Bay were collected on September 1, 2015. The photographs were processed using Structure-from-Motion (SfM) photogrammetric techniques. Global Positioning System (GPS) checkpoints were collected via a Global Navigation Satellite System (GNSS) survey conducted between August 15 and September 14, 2015. For the purposes of open access to elevation and orthoimagery datasets in coastal regions of Alaska, this collection is being released as a Raw Data File with an open end-user license. Orthoimages contain 4-band, 8-bit, unsigned raster data (red/green/blue/no data; file format-GeoTIFF; source-Fairbanks Fodar) with 20 cm GSD. No-data values are found in band 4 or shown as transparent in red/green/blue images. The file employs Lempel-Ziv-Welch (LZW) compression. Light exposures in the orthoimages are a result of daily weather conditions, which ranged from low cloud cover, rain, and full sun.

This study presents a Level of Detail 2 building reconstruction approach for open and occluded areas from stereo-satellite imagery. The approach combines deep learning techniques, and digital surface models with model-driven methodology. The best performance of deep learning algorithms (U-Net, FCN, and Mask R-CNN) for building boundary segmentation was selected and then integrated with model-driven technique for the purpose of accurate geometric building fitting employing digital surface model (DSM) generated by semi global matching. The Reconstructed model was refined by utilizing OpenStreetMap library and graph cut optimization method. The suggested methodology is tested on the GeoEye-1 satellite imagery dataset for Erbil City, which is validated with ground truth data. The proposed algorithm presented promising results, it is shown that the model can predict building heights for ridge and eave to a mean absolute error of 0.70 m, and in the occluded area was approximately 1.0 m. Meanwhile, the computed root mean square error are shown to be within 0.9 m for the ridge and eave, which is essentially small. While for occluded area it was approximately 1.2 m and 0.8 m for ridge and eave heights, respectively. This indicates that the predicted values are close to real values. Furthermore, most of the building’s roofs were correctly classified in both open and occluded areas. These findings underline the effectiveness of the model-driven deep learning approach in producing reliable and accurate LoD2 building reconstructions, a precondition for detailed urban analysis and 3D city modeling.