Standardizing Machine Learning APIs for Earth Observation Data Cubes

The accessibility of satellite data has catalyzed the hosting of these data sets in cloud computing environments, leading to several Earth Observation (EO) cloud platforms with analytical capabilities. However, transitioning between these platforms remains a challenge. The openEO specification addresses this by unifying backend services to be accessed via REST protocol using R, Javascript, and Python clients, yet it lacks a stable machine learning (ML) specification. This paper proposes extending the openEO API by developing a ML Application Programming Interface (API) specification for EO data cubes to facilitate ML models' reproducibility, reusability, and interoperability in multiple backend services and federated cloud platforms. Our efforts focus on establishing protocols for data pre-processing, model training, model parameter tuning, model prediction, model saving, and loading of existing models, facilitating the use of classical ML and deep learning (DL) algorithms for critical applications such as environmental monitoring and disaster response. Future efforts will focus on implementing this ML specification across multiple EO backend services and cataloging spatio-temporal ML models.