Code release for the paper "TAASRAD19, a high-resolution weather radar reflectivity dataset for precipitation nowcasting"
Franch, G., Maggio, V., Coviello, L. et al. TAASRAD19, a high-resolution weather radar reflectivity dataset for precipitation nowcasting. Sci Data 7, 234 (2020).
https://doi.org/10.1038/s41597-020-0574-8
The code includes the scripts for sequence extraction, a deep learning model for precipitation nowcasting and an online visualization build on TAASRAD19 dataset. The dataset can be downloaded from the following repositories:
A NETCDF version of the radar sequences (not needed for the code in this repository) is available here:
All the code was developed and tested on Ubuntu 18.04 with python 3.6+.
Create a new virtualenv (for example with venv
):
python3 -m venv .venv
source .venv/bin/ctivate
Install all required packages in the virtualenv:
pip install \
opencv-python PyYAML pandas \
numba numpy scipy tqdm imageio \
Pillow jupyterlab h5py umap-learn \
joblib matplotlib
The nowcasting deep learning model was tested with mxnet 1.5.1 that can be installed with CPU support:
pip install mxnet==1.5.1.post0
or for CUDA 10.1 GPUs (see other versions at mxnet website) :
pip install mxnet-cu101==1.5.1.post0