PyEarthTools: Accelerating the use of machine learning for reproducible environmental science

Machine learning opens up many new opportunities in environmental science, through discovering new insights from data to reduced costs of simulation. However a lot of environmental data does not work “out the box” with standard machine learning tools. As a result researchers need to spend a lot of effort into custom data pipelines for each problem. As the field has matured common usage patterns have emerged. PyEarthTools aims to provide common functionality for the use of ML in environmental science research that enables researchers to quickly build shareable, reproducible ML pipelines for training and inference that can run on many different platforms. A key focus is on enabling common patterns of data access and transformation required for ML. We will demonstrate how a domain specific tool that implements common domain patterns in data and functionality can greatly accelerate research and make it easier to follow FAIR principles. In encapsulating expert knowledge into components and pipelines, this can also make it easier for data scientists and experts from other domains to incorporate environmental science data into their work.