Wakulla Spring LSTM feature selection data and code repository

Repository for all code and data used in Compare and Ye, \"Using process-based understanding and feature selection to inform a LSTM neural network model for simulating stage of an eogenetic karst spring\".

This repository contains folders for Code, Data, Plots, Models, Results and Additional Figures.

Some raw data was collected manually form the Northwest Florida Water Management District online portal (https://nwfwmd.aquaticinformatics.net/) and the Florida Climate Center (https://climatecenter.fsu.edu/climate-data-access-tools/downloadable-data), and this has been noted in the code when appropriate. USGS data of Wakulla Spring was downloaded within the notebook. Both raw and processed data can be found in the data folder.

Code is organized in 4 Juptyer Notebooks

1. Data processing
2. Hyperparameter tuning for each of the neural networks
3. Training the neural networks under the tuned parameters and saving the network weights and simulation results
4. Generating figures used in the reports
(Due to the computational demands of hyperparameter tuning, it is recommended to not run this code again unless you have a good amount of storage and time.)

Model weights for each of the models are saved in the Models folder from Notebook 3, and loaded from here in Notebook 4.

Plots generated with code are saved in the Plots folder from Notebook 4.

Some additional figures for this study were generated in ArcGIS Pro and Adobe Illustrator, and these files can be found in the AdditionalFigures folder.

This is a clone of the GitHub repo found at https://github.com/kylecompare/FeatureSelection-KarstSpringLSTM from February 7, 2024.