Echo State Network implementation and Datasets

Overview

ESN implementation and Datasets for 'Assessing the predictability of meteorological variables via spatial correlations using echo state networks'. This project implements time series data prediction using ESN.

Data

The time-series climate datasets used in our paper. The data is averaged over 1-degree grid of ``JRA-55 6-Hourly Model Resolution Land Surface Analysis Fields data''. The original data were obtained from NCAR's website https://rda.ucar.edu . The file name structure is following: data/[yyyy]/[city]/[kind]/[kind]_lat[uu]_lon[vv]_[yyyy].csv

• [yyyy]: year of data
• [city]: the name of target city
• [kind]: kind of climatic quantity to predict
• [uu]: latitude of data
• [vv]: longitude of data

Sample code

• prediction.ipynb is a demonstration of ESN prediction for temperature data. For three different observed data in the latitudinal direction. Actual data and model prediction are plotted and NRMSE are printed.
• esn_dts.py is the python code to use ESN model in prediction.ipynb.

Citation

If you use this code in your research, please cite our paper: arXiv preprint: https://arxiv.org/abs/2406.03061

Author

Shihori Koyama, Daisuke Inoue, Hiroaki Yoshida
Toyota Central R&D Labs., Inc.
shihori-koyama@mosk.tytlabs.co.jp
Kazuyuki Aihara
The University of Tokyo
Gouhei Tanaka
Nagoya Institute of Technology

License

This repository is licensed under LICENSE.md.