Peat depth is an important control on peatland carbon storage and ecosystem function, but field observations are unevenly distributed and remain difficult to extrapolate across large areas. Recent satellite embedding products provide compact representations of Earth observation data, but it is not yet clear how much information they contain about peat depth or how well relationships learned in one region transfer to another. This exploratory project will assess the usefulness of satellite embeddings for predicting peat depth from existing georeferenced field observations. The observations will be harmonised and linked to annual satellite embeddings and a limited set of environmental variables prepared in Google Earth Engine. Several regression approaches will be compared using spatially separated training and evaluation data, reducing the risk that nearby observations produce overly optimistic accuracy estimates. The work will begin with relatively simple baseline models and will then examine whether GPU-supported models provide a meaningful improvement. The expected outputs are a documented modelling dataset, an evaluation of predictive performance and geographical transferability, and initial peat-depth predictions for selected study areas. The project is intended as a feasibility study that will help determine where satellite embeddings are useful and where additional environmental information or field data may be required.