This repo uses Python and Keras frameworks to build, train and test a neural network based on the paper.
Nimal C. Jayasundara, et al. Artificial Neural Network for Sacramento–San JoaquinDelta Flow–Salinity Relationship for CalSim 3.0
To try it on the cloud (mybinder.org) simply use 
To setup a local enviornment, first download miniconda3.
For preprocessing, create an environment based on preprocess_environment.yml file
conda env create -f preprocess_environment.yml
For ANN training, create an environment based on environment.yml file
conda env create -f environment.yml
Next, follow the instructions below to train ANN for 30cm Sea Level Rise Scenario.
Download the training dataset for 30cm Sea Level Rise Scenario here:
| Study Scenario | Model | File |
|---|---|---|
| Existing | CS3 | 1, 2, 3 |
| Existing | DSM2 | 1, 2 |
| SMSCG | CS3 | 1, 2, 3 |
| SMSCG | DSM2 | 1, 2 |
| PA6K | CS3 | 1, 2, 3 |
| PA6K | DSM2 | 1, 2 |
The repo contains jupyter notebooks and python code in two files. The starting point for input preprocessing are DSS files from one or mor runs of CALSIM based DSM2 studies.
- Preprocessing. The preprocessing notebook for 30cm SLR takes the .dss files and creates input and output csv files
- Training and Testing. The ANN training notebook uses the csv files and builds, trains, saves and tests the neural network