Building footprint polygons that represent the extent of individual building structures serve as the foundation for 3D city modeling and play a crucial role in decision-making issues such as city planning and development. Due to the poor contrast or shadow in orthoimages, extracting precise building footprints from them can pose a significant challenge. In this paper, we intend to use LiDAR point clouds alongside orthoimages to overcome this challenge and automatically detect building footprints by leveraging multimodal deep learning techniques. In the proposed network, these two modalities are combined in different levels of fusion including early (data) fusion, middle (feature) fusion, and late (head) fusion. The results show a 4% improvement in the proposed multimodal fusion network, compared to orthoimage or LiDAR-based baseline networks.

One of the most important disaster management requirements is accurate damage map generation to support rescue and reconstruction efforts. In this application, remote sensing images play a significant role because of the great details provided by their high spatial, spectral, and temporal resolutions; thus, the literature is rich with studies that use pre- and post-event images along with geospatial machine learning techniques for automatic damage mapping. However, acquiring proper pre-event data can be challenging due to the unpredictable nature of hazards. In this paper, we customize a pre-trained version of the residual neural network with 34 layers (ResNet-34) to identify damaged buildings by using only post-event high-resolution remote sensing images. For evaluating the damage detection framework efficiency, airborne orthophotos of the 2010 Haiti earthquake and the 2018 Woolsey fire are utilized. The network identified damaged and non-damaged buildings with over 91 % overall accuracy.

Synthetic Aperture Radar (SAR) imagery plays a vital role in flood mapping due to the day/night, almost all-weather, and cloud penetration capabilities. Although SAR backscatter intensity can accurately identify flooded areas on bare soil, it is still challenging to classify flooded urban areas due to the complexity of urban structures. Polarimetric SAR (PolSAR) and Interferometric SAR (InSAR) can provide us with a robust identification of backscatter patterns in urban areas, including single-bounce and double-bounce backscatters. In this study, we explore the potential of PolSAR and InSAR in urban flood mapping using a Random Forest model. The study area is located in Fredericton, New Brunswick, along the Saint John River, which has a long history of flooding. We examined various combinations of PolSAR and InSAR features, derived from Sentinel-1A images, along with four other features that are well-known to contribute to flooding, to select the best features for the model. The results showed that employing Polarimetric and Interferometric SAR (PolInSAR) features together with land-use/land-cover, altitude, slope, and aspect layers, reached an 88.6% flood classification accuracy in urban areas.

Recently developed deep learning networks along with advances in remotely sensed data have considerably broadened change detection applications. While tracking changes in urban areas manually is a laborious and time-consuming procedure, the recent improvements in deep learning have enabled researchers to use base and target images and update building footprint layers automatically with high accuracy. However, combining off-nadir satellite and airborne images for automatic change detection is still an ongoing issue in the literature. In this research, we used Patch-wise Co-registration (PWCR) and Mask R-CNN to implement building change detection over off-nadir very high-resolution satellite images taken in 2011 and 2013 from Fredericton, NB, Canada. Then, the new/demolished constructions were detected. The results showed that the model was able to detect buildings with nearly 85% overall accuracy compared to ground truth data.

The frequency and magnitude of natural disasters, especially floods, has been increasing in recent years and climate is expected to exacerbate these events. The increasing availability and number of satellites with continually improving spatial and temporal resolutions can provide efficient data sources to help with real-time disaster relief efforts from hazard assessment to rescue operations. Analyzing these volumes of data requires highly efficient algorithms that can produce fast and accurate results across the complex landscape. Artificial Intelligence (AI) provides an opportunity for repeatable, timely and reliable data processing. Once mapped, getting the spatial information to the users/decision-makers rapidly and in a format that is easily understandable and accessible has been a challenge. New methods of data sharing are emerging and gaining popularity. The widespread stability and availability of the internet has led to a surge in direct data access via Application Programming Interface (API) and interactive dashboards. Since the declaration of a global pandemic, there has been dozens of COVID-19 dashboards developed for tracking new cases. The overwhelming response to these dashboards has led to growth in the development of dashboards focused on assisting the emergency management community and first responders, as these dashboards can provide a common operating picture for multi-users operating from different locations.

Automatic detection of buildings out of urban objects is not a straightforward task due to the existing spectral and textural similarities. The problem gets even worse in buildings with pitched roofs. Pitched roof buildings receive dissimilar amount of solar radiation on their different faces causing different brightness values for a single roof. Thus, in object based classification methods, each side will probably be assigned to different segments preventing proper building boundary detection. In this study, in order to detect the proper building boundaries through image segmentation, IHS (Intensity, Hue, and Saturation) color system is used. Then, to detect buildings out of the segmented image, elevation information extracted from stereo satellite imagery is benefited. The presented method was tested on GeoEye stereo imagery and 92% of the image buildings were detected precisely.