Despite significant recent advances in cloud-based geospatial analysis platforms and free access to immense earth observation archives, global-scale monitoring of many coastal ecosystem types has remained unfeasible. This is largely due to the considerable resources required to develop large, analysis-ready, reference datasets suitable for training and testing remote sensing classification models.
We have developed an open-access reference library of coastal ecosystems suitable for training and verifying the latest generation of remote sensing classification models. coastTrain currently consists of >190,000 point occurrence records of 7 coastal ecosystem types. Absence data is also included (permanent water and other terrestrial data) to enable researchers, modellers and conservation organisations to quickly develop maps of the distribution of coastal ecosystem types.
coastTrain includes training data from several high profile coastal mapping projects, including the Global Tidal Flats project (http://intertidal.app), Global Mangrove Watch (https://www.globalmangrovewatch.org/), Global Tidal Wetlands Change (www.globalintertidalchange.org) and the Allen Coral Atlas (https://allencoralatlas.org/).
A full description of the methods, dataset, usage notes and data updates is provided in the published paper, Zenodo data archive and the coastTrain github repository.
The paper describing the coastTrain dataset is published in Remote Sensing. The paper includes important usage notes for the dataset and describes the genesis of coastTrain. Read the paper here.
Data and metadata
coastTrain is a version controlled dataset available in the open access data archive Zenodo (version 1.0). Metadata, live updates and any issues are available ia the Github coastTrain repository.
Version 1.0 of the coastTrain dataset made available via Zenodo data archive. https://doi.org/10.5281/zenodo.7080756
coastTrain Github repository is live. https://github.com/nick-murray/coastTrain
The coastTrain manuscript accepted at Remote Sensing.