Machine learning dataset construction and model training

With preprocessed data in hand, the focus shifts to the machine learning core of the pipeline. Everything up to this point has concerned analysis-ready data, whereas from here the concern is ML-ready data, in the sense fixed in the “Processing levels, analysis-ready, ML-ready” definition (see Overview of the EO data infrastructure landscape). This section covers two stages: constructing an ML-ready dataset (see Constructing a machine-learning-ready dataset) and designing and training the model (see Model design and training). Each stage raises challenges that are specific to the EO context, and the decisions made at these stages directly determine the quality of the pixel-level predictions.