More Like this

1. Understanding the Effects of Optimal Combination of Spectral Bands on Deep Learning Model Predictions: A Case Study Based on Permafrost Tundra Landform Mapping Using High Resolution Multispectral Satellite Imagery

   https://doi.org/10.3390/jimaging6090097  

   Bhuiyan, Md Abul; Witharana, Chandi; Liljedahl, Anna K.; Jones, Benjamin M.; Daanen, Ronald; Epstein, Howard E.; Kent, Kelcy; Griffin, Claire G.; Agnew, Amber (September 2020, Journal of Imaging)

   Deep learning (DL) convolutional neural networks (CNNs) have been rapidly adapted in very high spatial resolution (VHSR) satellite image analysis. DLCNN-based computer visions (CV) applications primarily aim for everyday object detection from standard red, green, blue (RGB) imagery, while earth science remote sensing applications focus on geo object detection and classification from multispectral (MS) imagery. MS imagery includes RGB and narrow spectral channels from near- and/or middle-infrared regions of reflectance spectra. The central objective of this exploratory study is to understand to what degree MS band statistics govern DLCNN model predictions. We scaffold our analysis on a case study that uses Arctic tundra permafrost landform features called ice-wedge polygons (IWPs) as candidate geo objects. We choose Mask RCNN as the DLCNN architecture to detect IWPs from eight-band Worldview-02 VHSR satellite imagery. A systematic experiment was designed to understand the impact on choosing the optimal three-band combination in model prediction. We tasked five cohorts of three-band combinations coupled with statistical measures to gauge the spectral variability of input MS bands. The candidate scenes produced high model detection accuracies for the F1 score, ranging between 0.89 to 0.95, for two different band combinations (coastal blue, blue, green (1,2,3) and green, yellow, red (3,4,5)). The mapping workflow discerned the IWPs by exhibiting low random and systematic error in the order of 0.17–0.19 and 0.20–0.21, respectively, for band combinations (1,2,3). Results suggest that the prediction accuracy of the Mask-RCNN model is significantly influenced by the input MS bands. Overall, our findings accentuate the importance of considering the image statistics of input MS bands and careful selection of optimal bands for DLCNN predictions when DLCNN architectures are restricted to three spectral channels. 

   more » « less
2. Transferability of the Deep Learning Mask R-CNN Model for Automated Mapping of Ice-Wedge Polygons in High-Resolution Satellite and UAV Images

   https://doi.org/10.3390/rs12071085  

   Zhang, Weixing; Liljedahl, Anna K.; Kanevskiy, Mikhail; Epstein, Howard E.; Jones, Benjamin M.; Jorgenson, M. Torre; Kent, Kelcy (April 2020, Remote Sensing)

   State-of-the-art deep learning technology has been successfully applied to relatively small selected areas of very high spatial resolution (0.15 and 0.25 m) optical aerial imagery acquired by a fixed-wing aircraft to automatically characterize ice-wedge polygons (IWPs) in the Arctic tundra. However, any mapping of IWPs at regional to continental scales requires images acquired on different sensor platforms (particularly satellite) and a refined understanding of the performance stability of the method across sensor platforms through reliable evaluation assessments. In this study, we examined the transferability of a deep learning Mask Region-Based Convolutional Neural Network (R-CNN) model for mapping IWPs in satellite remote sensing imagery (~0.5 m) covering 272 km2 and unmanned aerial vehicle (UAV) (0.02 m) imagery covering 0.32 km2. Multi-spectral images were obtained from the WorldView-2 satellite sensor and pan-sharpened to ~0.5 m, and a 20 mp CMOS sensor camera onboard a UAV, respectively. The training dataset included 25,489 and 6022 manually delineated IWPs from satellite and fixed-wing aircraft aerial imagery near the Arctic Coastal Plain, northern Alaska. Quantitative assessments showed that individual IWPs were correctly detected at up to 72% and 70%, and delineated at up to 73% and 68% F1 score accuracy levels for satellite and UAV images, respectively. Expert-based qualitative assessments showed that IWPs were correctly detected at good (40–60%) and excellent (80–100%) accuracy levels for satellite and UAV images, respectively, and delineated at excellent (80–100%) level for both images. We found that (1) regardless of spatial resolution and spectral bands, the deep learning Mask R-CNN model effectively mapped IWPs in both remote sensing satellite and UAV images; (2) the model achieved a better accuracy in detection with finer image resolution, such as UAV imagery, yet a better accuracy in delineation with coarser image resolution, such as satellite imagery; (3) increasing the number of training data with different resolutions between the training and actual application imagery does not necessarily result in better performance of the Mask R-CNN in IWPs mapping; (4) and overall, the model underestimates the total number of IWPs particularly in terms of disjoint/incomplete IWPs. 

   more » « less
3. Derived coastlines and toppled permafrost blocks at Drew Point, Beaufort Sea Coast, Alaska, 2018 and 2019

   https://doi.org/10.18739/a2n873213  

   Ward_Jones, Melissa; Jones, Benjamin (January 2025, NSF Arctic Data Center)

   To assess coastal erosion dynamics during the entire 2018 and 2019 open water seasons at Drew Point, Beaufort Sea Coast, Alaska, we derived 16 coastlines position using satellite, airborne and unmanned aerial vehicle (UAV) sensors. Sensors with associated image dates are: Worldview 1 imagery ©Maxar (14 April 2019), Worldview 2 panchromatic imagery ©Maxar (5 April 2019, 26 September 2019, and 3 April 2020); Modular Aerial Camera System (MACS-Polar) during the Polar-6 airborne operations during the ThawTrend-Air campaign (13 July 2019, 23 July, 2019, and 30 July 2019) and DJI Phantom 4 UAV surveys (24 July 2018, 29 July 2018, 3 August 2018, 30 September 2018, 2 August 2019, 6 August 2019, 10 August 2019, 12 August 2019 and 15 August 2019). Pixel resolution for the satellite, airborne and UAV imagery was 50 cm (Worldview 1), 46 centimeter (cm) (Worldview 2), 10 cm and 4 cm, respectively. The satellite-image derived coastlines span the 9 kilometer (Km) segment described in Jones et al. (2018; DOI: 10.1088/1748-9326/aae471), while the other coastline spans a 1.5 Km sub-section of the study area that includes the coastline, part of inland coastal area (~125 meters (m)) and fallen toppled permafrost blocks in front of the bluff. Fallen toppled permafrost blocks were digitized using the airborne and UAV images. The satellite imagery was too coarse to digitize blocks. All datasets are in WGS84 UTM Zone 5N. 

   more » « less
4. Using Object-Oriented Classification for Coastal Management in the East Central Coast of Florida: A Quantitative Comparison between UAV, Satellite, and Aerial Data

   https://doi.org/10.3390/drones3030060  

   Yang, Bo; Hawthorne, Timothy L.; Torres, Hannah; Feinman, Michael (September 2019, Drones)

   High resolution mapping of coastal habitats is invaluable for resource inventory, change detection, and inventory of aquaculture applications. However, coastal areas, especially the interior of mangroves, are often difficult to access. An Unmanned Aerial Vehicle (UAV), equipped with a multispectral sensor, affords an opportunity to improve upon satellite imagery for coastal management because of the very high spatial resolution, multispectral capability, and opportunity to collect real-time observations. Despite the recent and rapid development of UAV mapping applications, few articles have quantitatively compared how much improvement there is of UAV multispectral mapping methods compared to more conventional remote sensing data such as satellite imagery. The objective of this paper is to quantitatively demonstrate the improvements of a multispectral UAV mapping technique for higher resolution images used for advanced mapping and assessing coastal land cover. We performed multispectral UAV mapping fieldwork trials over Indian River Lagoon along the central Atlantic coast of Florida. Ground Control Points (GCPs) were collected to generate a rigorous geo-referenced dataset of UAV imagery and support comparison to geo-referenced satellite and aerial imagery. Multi-spectral satellite imagery (Sentinel-2) was also acquired to map land cover for the same region. NDVI and object-oriented classification methods were used for comparison between UAV and satellite mapping capabilities. Compared with aerial images acquired from Florida Department of Environmental Protection, the UAV multi-spectral mapping method used in this study provided advanced information of the physical conditions of the study area, an improved land feature delineation, and a significantly better mapping product than satellite imagery with coarser resolution. The study demonstrates a replicable UAV multi-spectral mapping method useful for study sites that lack high quality data. 

   more » « less
5. Automated High-Resolution Time Series Mapping of Mangrove Forests Damaged by Hurricane Irma in Southwest Florida

   https://doi.org/10.3390/rs12111740  

   McCarthy, Matthew J.; Jessen, Brita; Barry, Michael J.; Figueroa, Marissa; McIntosh, Jessica; Murray, Tylar; Schmid, Jill; Muller-Karger, Frank E. (June 2020, Remote Sensing)

   null (Ed.)

   In September of 2017, Hurricane Irma made landfall within the Rookery Bay National Estuarine Research Reserve of southwest Florida (USA) as a category 3 storm with winds in excess of 200 km h−1. We mapped the extent of the hurricane’s impact on coastal land cover with a seasonal time series of satellite imagery. Very high-resolution (i.e., <5 m pixel) satellite imagery has proven effective to map wetland ecosystems, but challenges in data acquisition and storage, algorithm training, and image processing have prevented large-scale and time-series mapping of these data. We describe our approach to address these issues to evaluate Rookery Bay ecosystem damage and recovery using 91 WorldView-2 satellite images collected between 2010 and 2018 mapped using automated techniques and validated with a field campaign. Land cover was classified seasonally at 2 m resolution (i.e., healthy mangrove, degraded mangrove, upland, soil, and water) with an overall accuracy of 82%. Digital change detection methods show that hurricane-related degradation was 17% of mangrove forest (~5 km2). Approximately 35% (1.7 km2) of this loss recovered one year after Hurricane Irma. The approach completed the mapping approximately 200 times faster than existing methods, illustrating the ease with which regional high-resolution mapping may be accomplished efficiently. 

   more » « less