KALS - Dynamical System Neural Network code

Author

Derek Karssenberg, Utrecht University

Introduction

This repository provides all the code required to create data, tables, and figures for the DSNN model for catchment hydrology. For explanation the reader is referred to the associated publication. This is not intended as a 'ready-to-use' model. To use the code some explanation is needed and possible users are requested to reach out to the author.

The main Python scripts are:

• kals_model.py, the DSNN model itself
• postprocessing.py, script to postprocess the DSSN model outputs

All other files are for running in batch or pdf post processing.

Structure of kals_model.py

The script contains the following components (from top to bottom).

1. Main configurations. Main configurations and explanation of how the model can be run in batch.
2. Data preparation. Functions to convert the input data (timeseries of meteo data) into a format that can be used in the model.
3. Dynamical System Neural Network model. This is the model itself. The first part of the code defines the neural network components. The second part defines the three model components, eva_f_calculate, sno_f_calculate, sub_f_calculate, each time with three options of code for synthetic data generation, process-based expert model, or neural network. The third part defines the iteration over time, calling the model components.
4. Create artificial observations. Code to create the synthetic data set. It calls the model with the configuration such that synthetic data model components are used.
5. Training. Code for training the model (function: training). It loops over epochs. Each epoch the model is run twice. The first run is for training, on the training fold. This same run is used for validation (testing), on the remaining data. The second run is for testing (independent validation), on the independent validation (testing) data. The second part of this code creates intermediate figures and writes data to disk.
6. Data set creation. Calls the function from Data preparation to create the input data for the model.
7. Fitting scenarios. Defines fitting scenarios.