HIDRA2: Deep-Learning Ensemble Sea Level and Storm Tide Forecasting in the Presence of Seiches – the Case of Northern Adriatic

paper live demo BibTeX

HIDRA2 is a state-of-the-art deep neural model for sea-level prediction based on past sea level observations and future tidal and atmospheric forecasts. This repository contains the implementation of HIDRA2 and code to train and evaluate the model. We also include HIDRA2 weights pretrained on 2006 – 2018 and predictions for all 50 ensembles on Jun 2019 – Dec 2020. Training and test datasets are published at destinations listed below.

Demo

Live forecast of the operational HIDRA2 model for the Koper Mareographic Station (northern Adriatic Sea) are visualized at: lojzezust.github.io/hidra-visualization.

Setup

To install the required packages, run

pip3 install -r requirements.txt

Evaluation

Predictions of HIDRA2 on Jun 2019 – Dec 2020 are stored in the data folder. To evaluate them, run

cd src
python3 eval.py

To recalculate the predictions, download the test data from this link and run

cd src
python3 test.py

Training

Pretrained weights are available in the data folder. To retrain the model, download the training data from this link and run

cd src
python3 train.py

The code will train the model on 2006 – 2018 data and save the trained model to the data folder.

Structure of the Data

Inputs to the HIDRA2 model are atmospheric, SSH and tidal data, each structured as follows:

Name	Shape	Description
atmos	96 × 3 × 9 × 12	Hourly atmospheric forecasts from 24 h prior the prediction point to 72 h into the future. Second dimension contains pressure and two channels of wind. Last two dimensions are latitude and longitude.
ssh	72	Past SSH.
tide	144	Past and future tide.

The model outputs 72 points representing hourly sea level forecast. All inputs and output of HIDRA2 are normalized, the statistics are stored in data/normalization_parameters.yaml.

Publication

Rus, M., Fettich, A., Kristan, M., and Ličer, M.: HIDRA2: deep-learning ensemble sea level and storm tide forecasting in the presence of seiches – the case of the northern Adriatic, Geosci. Model Dev., 16, 271–288, https://doi.org/10.5194/gmd-16-271-2023, 2023.

Cite as:

@Article{gmd-16-271-2023,
    AUTHOR = {Rus, M. and Fettich, A. and Kristan, M. and Li\v{c}er, M.},
    TITLE = {HIDRA2: deep-learning ensemble sea level and storm tide forecasting in the presence of seiches -- the case of the northern Adriatic},
    JOURNAL = {Geoscientific Model Development},
    VOLUME = {16},
    YEAR = {2023},
    NUMBER = {1},
    PAGES = {271--288},
    URL = {https://gmd.copernicus.org/articles/16/271/2023/},
    DOI = {10.5194/gmd-16-271-2023}
}