A 1.2 Billion Pixel Human-Labeled Dataset for Data-Driven Classification of Coastal Environments

Abstract

The world’s coastlines are spatially highly variable, coupled-human-natural systems that comprise a nested hierarchy of component landforms, ecosystems, and human interventions, each interacting over a range of space and time scales. Understanding and predicting coastline dynamics necessitates frequent observation from imaging sensors on remote sensing platforms. Machine Learning models that carry out supervised (i.e., human-guided) pixel-based classification, or image segmentation, have transformative applications in spatio-temporal mapping of dynamic environments, including transient coastal landforms, sediments, habitats, waterbodies, and water flows. However, these models require large and well-documented training and testing datasets consisting of labeled imagery. We describe “Coast Train,” a multi-labeler dataset of orthomosaic and satellite images of coastal environments and corresponding labels. These data include imagery that are diverse in space and time, and contain 1.2 billion labeled pixels, representing over 3.6 million hectares. We use a human-in-the-loop tool especially designed for rapid and reproducible Earth surface image segmentation. Our approach permits image annotation by multiple labelers, in turn enabling quantification of pixel-level agreement over individual and collections of images.

A Reproducible and Reusable Pipeline for Segmentation of Geoscientific Imagery

“…The labeled imagery we use in the present contribution are one of the 10 data records that comprise the Coast Train data set (Buscombe et al., 2022; Wernette et al., 2022), specifically 419 Landsat‐8 (top‐of‐atmosphere) images and associated labels consisting of time‐series from seven coastal locations around the United States (Figure 3a). The data set consist of visible‐band (RGB) imagery, and 2D integer label masks (Figure 3b).…”

“…In addition to presenting the design of Segmentation Gym, we demonstrate its use with an example using a data set consisting of 419 image‐label pairs. The labeled imagery consist of Landsat‐8 scenes of coastal environments from a large collection of labeled images (Buscombe et al., 2022; Wernette et al., 2022). We examine the sensitivity of model outputs to hyperparameter choices that govern model architecture and training strategies.…”