Coastal ecosystems are critical ecosystems that have faced increasing pressure from land use change and climate change in recent years. Various threats faced by coastal ecosystems also means communities have become vulnerable from environmental change, which could lead to flooding or coastal erosion.
A new collaboration initiative, called coastTrain, helps to provide data so that scientists can better address challenges coastal regions face.
Modeling coastal communities
Modeling or formal assessments of coastal communities requires remote sensing collected to demonstrate how coastal areas have changed or been affected by environment and land use change in recent years. This often involves classification techniques that use satellite data and machine learning efforts to demonstrate change.
The new coastTrain initiative helps scientists put together up to 193,105 occurrence records of seven coastal ecosystem types to help evaluate these coastal ecosystems and make better assessments of their vulnerability. These include muddy shorelines, mangroves, coral reefs, coastal saltmarshes, seagrass meadows, rocky shoreline, and kelp forests.
Remote sensing data have now been classified for these ecosystems, which helps the next generation of scientists working in these ecosystems better evaluate remote sensing change using machine learning or deep learning models that depend on training data and data to evaluate how effective classification is.
Increasingly, mapping ecosystems in coastal regions is becoming critical to monitoring their long-term status and make more informed decisions to support policy or management interventions.
What is coastTrain?
The coastTrain initiative covers diverse geographies, with data from every continent except Antarctica and every major sea. In total, four major global coastal ecosystem mapping efforts were combined to make the larger database and more than 120 countries were covered.
The project states its intent is to see coastTrain enable rapid development, utilization and validation of global-scale maps showing coastal ecosystems and their change. The satellite-based classifications and data collected are critical because they help reduce setup and model evaluation time needed to assess coastal ecosystems.
The data include point-format occurrence records where data inform on the project source and acquisition method for imagery, including the scale used and reference period.
To combine records from the other sources used to create this larger database, a workflow was established that included operations to generate unique record identifiers, applied the new coastTrain classification codes, and add corresponded fields to the International Union for the Conservation of Nature Ecosystem’s typologies.
Training coastal ecosystem classification models
The coastTrain tool is version controlled via GitHub and delivered using Zenodo. Although the main purpose of coastTrain is to train classification models and be used for validating developed distribution models that classify given coastal ecosystems, there are biases in the dataset as not all regions are represented comparably and data points should be evaluated for suitability in classification.
Some regions are probably less suitable to use this tool as well, such as more remote regions in far northern latitudes.
Other similar initiatives
Overall, the initiative is similar to other large-scale, global efforts to combine data so that better efforts to monitoring ecosystem change is possible. For instance, the Global Forest Watch is comparable in providing a data portal that can also be used to develop or monitor forest change efforts.
The Harmonized World Soil Database is another comparable effort bringing together different data on soils around the world.
These initiatives all show the need to better combine data, particularly remote sensing data, so that long-term monitoring of ecosystems is possible and better tools and decisions can be made to aid in decision-making and policy around given ecosystem resources.
While all of the global-scale remote sensing and data efforts, along with coastTrain, demonstrate the increasing interest of bringing large, remote sensing-based data repositories together, more datasets may need to be created and developed to cover different ecosystems.
Challenges around our oceans, in particular, stand out as some of these ecosystems are remote or difficult to map at large scales using high resolution.
Ecosystems of all types are facing increased threats and developing large, global-scale efforts that create repositories for training machine learning models or enable time-based assessment will be crucial if we are to understand how we are changing our planet and how it is changing our communities and ecosystems.
The coastTrain initiative represents an important part in helping our long-term policy efforts to make more informed decisions and monitor the health of our coastal regions.
 For more on the article describing coastTrain, see here: Murray, N. J., Bunting, P., Canto, R. F., Hilarides, L., Kennedy, E. V., Lucas, R. M., Lyons, M. B., Navarro, A., Roelfsema, C. M., Rosenqvist, A., Spalding, M. D., Toor, M., & Worthington, T. A. (2022). coastTrain: A Global Reference Library for Coastal Ecosystems. Remote Sensing, 14(22), 5766. https://doi.org/10.3390/rs14225766