The AI-Ready Downhole Microseismic Benchmark Database (AMBER)

AMBER contains labelled waveforms from microseismic events recorded on deep downhole sensor arrays, generated to stimulate research and development in the use of AI tools for downhole microseismic processing tasks. Raw SEGY waveforms can be converted into Seisbench-compatible datasets (waveforms.hdf5 + metadata.csv) using the extraction pipeline extract.py. This repository also provides an event-centric PyTorch dataset with configurable downhole-specific augmentations for training deep-learning models on multi-station, multi-event waveforms. AMBER has been compiled from 10 datasets (or sub-datasets): - Cotton Valley Stage B - Aneth CCS - Clearfield mw4 monitoring well - Clearfield mw6 monitoring well - MSEEL stage 3H - MSEEL stage 5H - FORGE geothermal 2019 - FORGE geothermal 2022 - Preston New Road PNR-1 - Preston New Road PNR-2