This session will feature academic-based presentations on image classification approaches including AI and machine learning.
2:00 – 2:15 PM – Wetland Classification With Multispectral and Lidar Data Using Deep Learning
Accurate and up-to-date wetland maps are critical for conserving wetlands on a landscape scale. The most effective methods for creating these maps are those that are automated, scalable, repeatable, and widely applicable. However, wetlands vary significantly across different regions and over time, making it challenging to map them using predictive models. Currently, most approaches are limited by geographical specificity, reliance on commercial data, and low resolution. This research aims to develop a deep learning model to evaluate its effectiveness in mapping wetlands across the Columbia, North Carolina landscape. The study employs a U-Net model and trains the model using UAV imagery, multispectral data from the National Agriculture Imagery Program, Sentinel-2 imagery, and two LiDAR-derived datasets to generate wetland maps at a 1-meter spatial resolution. The results demonstrate how freely available data and advanced deep learning techniques can achieve high-resolution mapping without manual feature engineering. Given the dynamic nature of wetlands and their vital ecosystem services, such high-resolution mapping can transform decision-making in development and restoration projects.Â
Matilda Anokye, NC A&T University
2:15 – 2:30 PM – Remote Sensing InSAR Technology Applications and Services for Texas and Beyond
In Texas, Interferometric Synthetic Aperture Radar (InSAR) technology has proven highly effective in tracking and analyzing various geological and environmental events. This presentation provides a comprehensive overview of InSAR technology and its diverse applications in the region, with a specific focus on its role in natural hazard assessment, oil and gas operations, land subsidence, and coastal and urban infrastructure monitoring. InSAR’s ability to identify millimeter-scale ground deformations is particularly valuable for monitoring subterranean operations associated with oil and gas extraction. This facilitates the identification of potential environmental impacts and supports resource management. It contributes to sustainable practices, infrastructure resilience, and proactive responses to environmental changes. In Texas’s dynamic and varied geography, InSAR proves to be a vital tool due to its capacity to deliver precise and timely information on ground deformations. Awareness and knowledge of this technology is becoming increasingly important as the newest radar satellite deploys soon and will provide game-changing capabilities that will help us understand land deformation in the State of Texas like never before.
Dr. Danielle Smilovsky, Conrad Blucher Institute
2:30 – 2:45 PM – Evaluating the Accuracy of Multi-Return Lidar Sensors in UAS Mapping
The advent of Unmanned Aerial Systems (UAS) equipped with multi-return lidar sensors has revolutionized surveying and mapping by enabling efficient data collection, especially in challenging environments like densely wooded forests. Multi-return lidar R sensors, such as the DJI L2, provide significant advantages over single-return sensors by capturing multiple layers of data, allowing for the extraction of information below canopy coverage or other obstructions. These sensors are particularly effective in capturing detailed terrain information as well as vegetation vertical structures, crucial for surveying densely wooded areas. This paper explores the critical role of multi-return L2 lidar sensors in modern surveying, comparing the first through fifth returns from a DJI L2 sensor to ground truth data obtained with a terrestrial laser scanner (TLS). The experiment involves using the DJI L2 lidar sensor mounted on a Matrice 350 RTK to collect data from a dense forest, with terrain information subsequently extracted using software like DJI Terra. The TLS is set up to take measurements beneath trees, providing ground truth data to create a DEM. This DEM serves as a benchmark to validate the accuracy of the terrain measured by the multiple returns from the L2 lidar sensor. Statistical analysis involves sampling XY points and comparing their Z elevations to the corresponding points measured by the TLS. This comparison helps in assessing the accuracy and reliability of the multi-return L2 lidar data. By systematically comparing L2 lidar data to the TLS measurements, the study demonstrates the reliability of these advanced sensors in capturing detailed and accurate data in complex environments.
Rami Tamimi, The Ohio State University
2:45 – 3:00 PM – Evaluating the Impacts of Hurricane Irma on Georgia Heirs Property Owners Using NASA Earth Observations
In September 2017, Hurricane Irma hit the southeastern coast of the United Sates, making landfall in Georgia. Hurricane Irma caused flooding and widespread destruction in the region, resulting in heirs’ property owners being denied access to federal relief due to the legal status of their property titles. To understand the impact of hurricanes to this community, NASA DEVELOP partnered with the Georgia Heirs’ Property Law Center, a nonprofit law firm that works with heirs’ property owners. We used Landsat 7 Enhanced Thematic Mapper Plus (ETM+), Landsat 8 Operational Land Imager (OLI), Sentinel-2 Multispectral Instrument (MSI), Sentinel-1 C-band Synthetic Aperture Radar (C-SAR), and generated flood extent maps by consolidating NASA SERVIR’s Hydrologic Remote Sensing Analysis for Floods tool in Google Earth Engine. We also used Computer Assisted Mass Appraisal data to identify the heirs’ property owners and overlayed the data with the flood extent map to recognize owners in need of assistance. We validated the flood extent maps with USGS Hurricane Irma High Water Mark in situ data, taken the same day Irma moved across the Georgia border, showing that backscatter data were more reliable than surface reflectance data in identifying coastal flooding. To evaluate the affected we correlated the Federal Emergency Management Agency’s compensation denials based on property title issue to flood and heirs’ property likelihood maps. These end products will contribute to the partners decision making surrounding for legal support in fund allocation disaster assistance, community outreach, and asset education on home, to maintain the ownership for the future generations.
Nancee Uniyal, University of Georgia
3:00 – 3:15 PM – Establishing Knowledge Gaps, Best Practices, and Standard Tools for Invasive Plant Monitoring via UAS
Unmanned aerial vehicles (UAS, i.e., drones) are becoming increasingly common tools for invasive plant monitoring and management. The use of UAS allows invasive aquatic plant (IAP) managers and technicians to survey larger areas than can generally be managed on foot or by boat alone. UAS are also able to collect data for areas that would otherwise be difficult to access, including backwater areas and areas with thick stands of emergent vegetation that may be impassable. A number of peer-reviewed papers have shared UAS-based approaches for monitoring individual IAP species, but rarely in enough detail for the identification and mapping methodology to enable reproducibility, especially for natural resource managers with limited experience in aerial imagery collection and analysis. This USFWS-funded project in collaboration with the Great Lakes Commission (GLC) has focused on establishing guidance for collecting and analyzing UAS surveillance images of invasive aquatic plants. This presentation will focus on the methods used this project, including: 1) the collection of UAS imagery in study sites in the Great Lakes Region, 2) the classification tools used for the imagery analysis with a focus on aquatic invasive species classification, and 3) building a workflow that can be easily replicated by the project’s end users who are comfortable in an ESRI ArcGIS environment.
Dana Redhuis, Michigan Technological University
3:15 – 3:30 PM – Advancing Individual Tree Mapping in Deciduous Forests: A Deep Learning Approach for Stem Detection Using Terrestrial and Drone Lidar Data
The mapping of individual trees using remote sensing data holds immense potential for enhancing forest inventory efficiency, advancing precision forestry and ecological research, optimizing agricultural and urban forestry management, promoting biodiversity conservation, enabling technological automation and scalability, and addressing climate change and sustainable development. However, generating individual tree products at scale remains a significant challenge, particularly in natural deciduous forests. Recent advancements in lidar sensor technology and deep learning techniques offer promising opportunities to address this challenge. In this study, we leveraged globally distributed, publicly available terrestrial lidar data to train a deep learning-based stem detection model. We then applied this model to drone-collected lidar point clouds, demonstrating its strong transferability in detecting tree stems and generating individual tree point clouds based on the identified stems. Furthermore, we compared our model with existing algorithms relying on the widely used DBSCAN approach, revealing substantial improvements in performance achieved by our method. This work demonstrated the potential of integrating deep learning with lidar data to advance large-scale individual tree mapping in complex forest ecosystems.
Tao Liu, Michigan Technological University
