A High-Resolution Airborne Color-Infrared Camera Water Mask for the NASA ABoVE Campaign

Abstract

The airborne AirSWOT instrument suite, consisting of an interferometric Ka-band synthetic aperture radar and color-infrared (CIR) camera, was deployed to northern North America in July and August 2017 as part of the NASA Arctic-Boreal Vulnerability Experiment (ABoVE). We present validated, open (i.e., vegetation-free) surface water masks produced from high-resolution (1 m), co-registered AirSWOT CIR imagery using a semi-automated, object-based water classification. The imagery and resulting high-resolution water masks are available as open-access datasets and support interpretation of AirSWOT radar and other coincident ABoVE image products, including LVIS, UAVSAR, AIRMOSS, AVIRIS-NG, and CFIS. These synergies offer promising potential for multi-sensor analysis of Arctic-Boreal surface water bodies. In total, 3167 km 2 of open surface water were mapped from 23,380 km 2 of flight lines spanning 23 degrees of latitude and broad environmental gradients. Detected water body sizes range from 0.00004 km 2 (40 m 2 ) to 15 km 2 . Power-law extrapolations are commonly used to estimate the abundance of small lakes from coarser resolution imagery, and our mapped water bodies followed power-law distributions, but only for water bodies greater than 0.34 (±0.13) km 2 in area. For water bodies exceeding this size threshold, the coefficients of power-law fits vary for different Arctic-Boreal physiographic terrains (wetland, prairie pothole, lowland river valley, thermokarst, and Canadian Shield). Thus, direct mapping using high-resolution imagery remains the most accurate way to estimate the abundance of small surface water bodies. We conclude that empirical scaling relationships, useful for estimating total trace gas exchange and aquatic habitats on Arctic-Boreal landscapes, are uniquely enabled by high-resolution AirSWOT-like mappings and automated detection methods such as those developed here.

Semi-Automated Semantic Segmentation of Arctic Shorelines Using Very High-Resolution Airborne Imagery, Spectral Indices and Weakly Supervised Machine Learning Approaches

“…Similar work to create land/water masks from 1 m resolution, airborne, color-infrared imagery using NDWI threshold and object-based classification methods in Arctic-Boreal regions report recall, precision and IOU of 0.94, 0.87 and 0.83 respectively for the water class. Additionally, they report recall, precision and IOU of 0.98, 0.99 and 0.98, respectively, for the land class [95]. While we expect the performance between different models to be relatively similar for other regions for this task, we may see an improvement in performance in using DL methods with more number of land cover classes [96].…”

High-Throughput Estimation of Crop Traits: A Review of Ground and Aerial Phenotyping Platforms

“…CIR cameras can be used to obtain high-spatial-resolution imagery with a limited number of wide spectral bands at a relatively low cost. However, CIR cameras may be unstable, and they have a limited application, as they are not designed for precise radiometric measurements [263].…”

Advancing Field-Based GNSS Surveying for Validation of Remotely Sensed Water Surface Elevation Products

“…The open-water classification uses on an object-based unsupervised classification routine, with manual quality control. A threshold was applied to a normalized difference water index (Mcfeeters, 1996) in order to identify open-water seed regions, and a custom region-growing procedure was developed to determine a final open-water boundary (Kyzivat et al, 2019a; Figure 5). Here, we use the open-water mask created from August 17, 2017, CIR data over the North Saskatchewan River to systematically remove nonopen-water pixels from both July and August radar data.…”