Fig 1.
1A. Location of the Chukchi Sea. 1B. Location of the two parks and relevant coastal landmarks. Elevation data USGS 30 ARC-second Global Elevation Data, GTOPO30. Research Data Archive at the National Center for Atmospheric Research, Computational and Information Systems Laboratory. https://doi.org/10.5065/A1Z4-EE71. CC By 4.0.
Fig 2.
Examples of coastal geomorphology features.
A. Patterned salt marsh at the mouth of the Nugnugaluktuk Estuary, eastern BELA. B. Bedrock outcrop on the coast of Goodhope Bay, eastern BELA. C. Ice-wedge polygon tundra bluff along the beach north of Cape Krusenstern Lagoon, CAKR. D. Active and stabilized sand dunes on the barrier island of Cowpack Lagoon, northern BELA. Photo A NPS, public domain, B, C & D photos from ShoreZone [33], licensed CC By 3.0.
Fig 3.
Training points were (a) digitized via interpretation of the imagery, (b) intersected with the segmentation with spatial, spectral and topographic variables extracted, and (c) classified to land cover class via Random Forest.
Imagery: Manley, WF, Sanzone, DM, Lestak, LR, and Parrish, EG 2007. Index Layers for High-Resolution Orthorectified Imagery from 2003 for the Coastal Areas of Bering Land Bridge NP (BELA). https://irma.nps.gov/DataStore/Reference/Profile/1044493 Public domain.
Table 1.
Pearson correlation between site variables and the NMS ordination scores.
Table 2.
Summary of model design.
Fig 4.
Ordination of plots using NMS.
The point cloud was rotated to align the hydric index maximally with Axis 1. The 5 variables most correlated with Axes 1 and 2 are displayed (Table 3). Colored symbols refer to the vegetation classes, defined in Table 2. Classes were grouped by convex hulls, with Willow Scrub and unclassified plots excluded.
Table 3.
Segment variables and their imagery sources.
Table 4.
Map land cover classes.
Fig 5.
Mean Decrease in Accuracy (MDA) for the unified study area model, by land cover class.
The first, labeled column displays the MDA for the model overall.
Table 5.
The producer’s error for the independent test set of excluded field plots for all four sets of models and out-of-bag producer’s error from each RF model.
Fig 6.
OOB producer’s error for the four modeling approaches, subdivided by land cover class.
Land cover classes are ordered alphabetically. The n of training data is presented in parentheses.
Fig 7.
OOB producer’s error for the four modeling approaches, subdivided by satellite scene.
Scenes are ordered by amount of training data.
Table 6.
Error matrix for the field test plots.
Fig 8.
Map of coastal vegetation of Bering Land Bridge National Preserve.
Background imagery Landsat mosaic produced for the Arctic Network. https://irma.nps.gov/DataStore/Reference/Profile/2171608 Public domain.
Fig 9.
Map of the coastal vegetation of Cape Krusenstern National Monument.
Background imagery Landsat mosaic produced for the Arctic Network. https://irma.nps.gov/DataStore/Reference/Profile/2171608 Public domain.
Fig 10.
Highlights of the coastal vegetation map.
A. Inner beach ridge complex of Cape Espenberg, BELA. B. Outer barrier island of Ikpek Lagoon, BELA. C. Beach ridge complex of Cape Krusenstern, CAKR. D. Nugnugaluktuk estuary, BELA. Background imagery Landsat mosaic produced for the Arctic Network. https://irma.nps.gov/DataStore/Reference/Profile/2171608 Public domain.