This repository contains jupyter notebooks to create basin-scale stratigraphic geometries from a conceptual geometric model, train a K-nearest neighbor classifier on the dataset, and perform a grid search to determine optimal parameters for the model.
Notebooks are organized as follows:
- 00_map_patterns.ipynb - Creates ambiguous map patterns from truncated and onlapping stratal geometries
- 01_training_data.ipynb - Creates training data from a geometric model for a varying number of adjacent wells. This dataset can be downloaded from here
- 02_grid_search_predictions.ipynb - Runs a grid search to optimize the number of adjacent wells and KNN classifier hyperparameters for the training datasets
- 03_predictions_tsne.ipynb - Investigates which feature group is the most important for classification, makes predictions on a subsurface dataset available from both here (
subsurface_data.csv)and here (Appendix 1, tab 4). The subsurface predictions from the model are saved to shapefiles (original shapefiles are in thepredictionsdirectory) - 04_spatial_prediction_viz.ipynb - Visualizes the spatial distribution of predictions and plots the predictions with isochore maps and outcrop exposures of the formations
Training data can be generated locally, or downloaded from OSF Subsurface data is also located at OSF or direct from the Wyoming State Geological Survey
To run this repository, you need to either clone or download it to your local machine. After cloning, you need to change directories to this directory. To ease in reproducibility we have included a Python virtual environment. Install the virtual environment from a terminal window or anaconda prompt with:
$ conda env create -f strata.yml
After the virtual environment is installed, activate it and start a jupyter notebook or jupyterlab session
$ conda activate strata
$ jupyter lab
or
$ jupyter notebook
When the jupyter server has loaded, you can walk through the notebooks in order to reproduce the results from this study.