LightNetPlus

Directory description

data_dir　　　　　# Validation and test data are stored.
　|-AWS　　　　　# Automatic weather station data.
　|-LIG　　　　　　# Lightning observation data.
　|-WRF　　　　　# WRF simulation data: micro-physical parameters and maximum vertical velocity.
　|-WRF_ncl　　　　# WRF simulation data: radar reflectivity.

test_dir　　　　　# Files related to testing are stored.
　|-curves　　　　# Performance curves in hours.
　|-results　　　　# Prediction results by every model.
　|-scores　　　　# Performance scores (POD, FAR, TS, ETS) for every model.
　|-visualization　　# Case visualization results for every test period.

train_dir　　　　　# Files related to training are stored.
　|-models　　　　# Trained models.
　|-records　　　　# Training log files.

data_generator.py　　　# Load data from data_dir and formats them into numpy arrays.

draw_each_hour_curve.py　　　# Draw performance curves in hours.

global_var.py　　　# Define global variables which will be used in the whole project.

model_def.py　　　# Define structure of all models.

score.py　　　# Calculate performance scores for prediction results.

test.py　　　# Test a trained model and calculate performance scores for the model.

test_periods.txt　　　# The periods used for test. 2017.08-09.

train.py　　　# Train a deep neural network model.

training_periods.txt　　　# The periods used for training. 2015.06-09 & 2016.05-09 & 2017.05-06

validation_periods.txt　　　# The periods used for validation. 2017.07

visualization_case.py　　　# Case visualization for every test period.

Runtime environment

Our code requires python 3.6 with packages: tensorflow 1.13.1, keras 2.2.4 and numpy. If you use Anaconda, the following commands will help you create a feasible runtime environment.

conda create -n py36_keras224 python=3.6

conda activate py36_keras224 

conda install tensorflow-gpu==1.13.1

pip install keras==2.2.4

Then, you may need to remove the following code

    inputs, initial_state, constants = _standardize_args(
         inputs, initial_state, constants, self._num_constants)

from "keras/layers/convolutional_recurrent.py", due to a bug in ConvLSTM2D of keras. cf. keras-team/keras#9761

Reference

@article{geng2021deep,
  title={A deep learning framework for lightning forecasting with multi-source spatiotemporal data},
  author={Geng, Yangli-ao and Li, Qingyong and Lin, Tianyang and Yao, Wen and Xu, Liangtao and Zheng, Dong and Zhou, Xinyuan and Zheng, Liming and Lyu, Weitao and Zhang, Yijun},
  journal={Quarterly Journal of the Royal Meteorological Society},
  volume={147},
  number={741},
  pages={4048--4062},
  year={2021},
  publisher={Wiley Online Library}
}