We are providing you with cloud-native seismic files in two different formats: OpenVDS+ and Real Simple Seismic (rss). Our team converted these from SEG-Y so you can have fast slice access for your machine learning pipeline! Please see our example notebooks on various ways to access the seismic data here and here. Cloud versions (for SageMaker) are here and here. Following seismic datasets are available. We may add more files (seismic attributes/features) during the competition.
We ingested publicly available wells into a pandas DataFrame and we serialized it to our AWS S3 bucket in JSON format. The well names are blanked out and two of the wells are being kept as testing data. You can see examples of the well data ingestion and some basic visualizations here and here.
The cloud well data, after deserialization, unfolds into a MultiIndex pandas DataFrame. Well IDs and two-way-time values make up the MultiIndex. We also provide inline and crossline positions, true vertical depths (relative to sea level), and elastic logs.
Beware! There are missing portions of some of the logs, just like real life, so you either throw away useful data or augment it.
We encourage competitors to predict missing well logs with empirical or predictive models! Available well info:
We interpreted a bunch of horizons and converted them to a pandas DataFrame. Serialized it to our AWS S3 bucket in JSON format, per horizon! Click here and here to see examples of getting the data. We interpreted the following surfaces (shallow to deep)