Evidence has suggested that built environments are significantly associated with residents’ health and the conditions of built environments vary between neighborhoods. Recently, there have been remarkable technological advancements in using deep learning to detect built environments on fine spatial scale remotely sensed images. However, integrating the extracted built environment information by deep learning with geographic information systems (GIS) is still rare in existing literature. This method paper presents how we harnessed deep leaning techniques to extract built environments and then further utilized the extracted information as input data for analysis and visualization in a GIS environment. Informative guidelines on data collection with an unmanned aerial vehicle (UAV), greenspace extraction using a deep learning model (specifically U-Net for image segmentation), and mapping spatial distributions of greenspace and sidewalks in a GIS environment are offered. The novelty of this paper lies in the integration of deep learning into the GIS decision-making system to identify the spatial distribution of built environments at the neighborhood scale.
8 August 2023 in Journal of Geovisualization and Spatial Analysis
From April 14 to 18, 2024, the United Arab Emirates (UAE) experienced its heaviest rainfall in 75 years, resulting in widespread flooding across multiple emirates, including Dubai. This study utilizes high-resolution PlanetScope imagery and a U-Net deep learning model to assess the flood impact and analyze post-rainfall recovery patterns in Dubai’s urban landscape. By integrating Sentinel-2-derived land use and land cover (LULC) data to refine the training dataset, a high-accuracy U-Net model was developed through transfer learning that effectively classified pre- and post-rainfall LULC. Post-rainfall LULC change detections indicate that 23.8 km² of land was flooded, which is equivalent to about 10 times the area of Downtown Dubai. Bare ground (66% of the flooded land) and built area (32% of the flooded land) were the most affected, while vegetation (2% of the flooded land) presented greater flood resilience. Post-event monitoring revealed that 95% of flooded areas remained submerged after three days and 37% were still underwater even ten days post-event. These findings highlight the prolonged impact of extreme rainfall on urban infrastructure in an arid environment. This study contributes to remote sensing-based flood impact assessment by leveraging high-resolution PlanetScope imagery and deep learning techniques. It demonstrates the effectiveness of transfer learning in improving LULC classification, particularly for minority classes such as small-patch vegetation. The results provide critical insights into urban flood recovery dynamics and offer valuable information for disaster management, flood resilience planning, and future urban adaptation strategies.
3 October 2025 in IEEE Geoscience and Remote Sensing Letters
Urban flooding in arid regions threatens infrastructure and public safety. Fine-scale mapping of flood extents is vital for effective emergency response and resilience planning, but limited spectral contrast, rapid hydrological changes, and heterogeneous land covers make this task challenging. High-resolution, daily PlanetScope imagery provides the temporal and spatial detail needed. In this work, we introduce FM-LC, a hierarchical framework for Flood Mapping by Land Cover identification, for this challenging task. Through a three-stage process, it first uses an initial multi-class U-Net to segment imagery into water, vegetation, built area, and bare ground classes. We identify that this method has confusion between spectrally similar categories (e.g., water vs. vegetation). Second, by early checking, the class with the major misclassified area is flagged, and a lightweight binary ‘expert’ segmentation model is trained to distinguish the flagged class from the rest. Third, a Bayesian smoothing step refines boundaries and removes spurious noise by leveraging nearby pixel information. We validate the framework on the April 2024 Dubai storm event, demonstrating average F1-score improvements of up to 29% across all land-cover classes and notably sharper flood delineations. Compared to conventional single-stage U-Nets, FM-LC achieves over 12% higher mean F1, significant gains for vegetation classification, and more reliable temporal tracking of flood dynamics. These results highlight FM-LC as a practical and scalable solution for high-resolution flood mapping in complex urban and arid environments.
This study proposes a streamlined workflow for producing historical digital elevation models (hDEMs) from scanned 1950s aerial photographs using structure-from-motion and multi-view-stereo (SfM-MVS) techniques along with co-registration methods. We also conducted a sensitivity analysis to assess the impact of DEM references with varying spatial resolutions on the SfM-MVS process and co-registration accuracy. The DEM references included a 30 m SRTM DEM (low resolution), a 12 m TanDEM-X DEM (medium resolution), and a 5 m DEM provided by the General Directorate of Mapping of the Ministry of National Defense, Republic of Türkiye (high resolution). Results indicate that higher resolution reference DEMs lead to improved accuracy and precision of the final hDEMs, particularly evident in reduced errors and finer spatial resolution. This study contributes to streamlining the hDEM construction process and offers a nuanced understanding of the significance of elevation and reference DEM selection in ensuring hDEM accuracy.