Pacific Walrus Coastal Haulout Occurrences Interpreted from Satellite Imagery

For the 2023 haulout season: we examined and interpreted all available satellite imagery over the study area through the autumn study period, before, during and after dates when walruses had been previously reported at the haulout site. We retrieved all available Sentinel-2 and Planet Scope optical images and retained those in which we could see a clear view of the beach at the study area. We examined all SAR images collected by Sentinel-1, both in the 10 m resolution VH cross-polarization and 40 m resolution HV cross-polarization. We tasked collection of Umbra (HH) Spotlight images. We tasked collection of TerraSAR-X Staring Spotlight (approx. 7.5 km x 4.6 km) images with approx. 1.1 m resolution in a single polarization (HH) and had these processed with the "science" orbitography data option to optimize ground registration, and in the radiometrically enhanced processing mode. We tasked collection of ICEYE imagery primarily after walruses had departed the haulouts, Capella Space was not available for tasking during the 2023 haulout season, and we did not task Radarsat-2 because the Umbra constellation offered sufficiently comprehensive temporal coverage.

To recognize walrus groups resting on shore, we first examined imagery types in chronological order before walruses arrived, to orient ourselves with how the study area's coastal landscape appeared in the imagery, including how to recognize features of the barrier island such as variations in beach width, banks at the edge of the storm high-tide line, and upland vegetation, as well as remnants of the historic settlements and military infrastructure. Walruses disturb beaches and upland vegetation when they gather in large numbers, churning up the soil and defecating. In optical imagery, occupied haulout areas are discolored by these behaviors. Because this discoloration persists when walruses are no longer present, and may be mistaken for actual walrus herds, we examined images in chronological order to recognize these disturbed areas so that they may be distinguished from walrus herds. We performed all visual interpretation of imagery in a desktop GIS. We visualized the Sentinel-2 multi-spectral image bands as standard true-color red-green-blue (RGB) images when views were unobstructed by thin clouds and as false-color composites using the near infra-red, red, and green bands for the RGB channels respectively when thin clouds obstructed the view. For each visualization we stretched each band's color ramp from the 2nd to the 98th percentile of pixel values found within the study area and applied brightness and contrast settings to optimize visual contrast of landscape features. If thin patchy clouds were present over the study area, we applied the image enhancement methods separately to areas with and without clouds.

We obtained Sentinel-1 cross-polarization SAR imagery as a sigma-nought (sigma0) data product. Sigma0 is normalized with respect to the local incidence angle (i.e., terrain) so it represents the backscatter returned to the antenna from a unit area on the ground, expressed in decibels (dB) with a calibration level of 0 dB. We visualized sigma0 by stretching values from -25 to 0 and displaying them over a viridis magma color ramp to enhance geographic patterns. We recognized walrus groups on the ocean-facing side of the barrier island as large contiguous clusters of pixels with very high sigma0 that often contrasted with the surrounding coastal landscape.

We visualized the TerraSar-X ellipsoid corrected single look images by stretching the single band images from the 2nd to the 98th percentile of the radar backscatter values within the study area, displayed over a viridis magma color ramp. We recognized walrus groups on the ocean-facing side of the barrier island as large contiguous clusters of pixels with very high radar backscatter that often contrasted with the surrounding coastal landscape.

We visualized the Umbra Space ellipsoid corrected two-look images by stretching the single band images from 0 to 255 of the scaled radar backscatter values within the study area, displayed over a viridis magma color ramp. We recognized walrus groups on the ocean-facing side of the barrier island as large contiguous clusters of pixels with very high radar backscatter that often contrasted with the surrounding coastal landscape.

For satellite imagery interpretation of walrus groups, we identified walrus groups as contiguous clusters of pixels that met criteria for distinguishing walrus groups from the study site landscape, whereby pixel scale varied among the satellite imagery sources. Criteria for distinguishing walrus herds varied among the image types. For multi-spectral optical images, we recognized walrus aggregations in the true-color imagery by a distinctive red-brown coloration along the ocean-facing beach that contrasted with the adjacent brown beach, green upland vegetation, and dark brown trampled upland vegetation (or sometimes white fresh snow). Walruses in the false-color image composites were recognizable by similar patterns but with different contrasting colorations. We recognized walrus aggregations in the near infra-red band by distinctive mid-level intensity values that contrasted with darker levels of the adjacent beach, variable levels of the adjacent uplands, and brighter levels in the nearshore surf zone. For synthetic aperture radar images, we recognized walrus aggregations as large contiguous clusters of pixels with high radar backscatter on the ocean-facing side of the barrier island that contrasted distinctly with the surrounding coastal landscape.

To determine haulout area, we projected satellite imagery to the local UTM coordinate reference system (indicated in the projection metadata for each shape file) and digitized haulout outlines. A single experienced interpreter interpreted the satellite imagery. The interpreter digitized the periphery of apparent walrus groups that were comprised of at least 6 contiguous image pixels and that contrasted with the surrounding coastal landscape. The interpreter digitized walrus groups by hand with the use of a mouse and a desktop GIS at scale of: (a) 1:10,000 for imagery types with pixel sizes greater than 10 m, (b) 1:5,000 for imagery with pixel sizes between 3 m and 10 m, and (c) 1:2,000 scale for imagery with pixel sizes less than 3 m. These outlines were reviewed by a second interpreter. If concerns were raised by the second interpreter, the original interpreter revised the outlines such that the final published digitized outline represents a consensus outline from the two experienced interpreters.