This repository provides an implementation of the Deep-Pathfinder algorithm, extracting mixing layer height information from Lufft CHM15k ceilometer data using computer vision techniques.
Note that this script is designed for real-time application, so please replace the NetCDF file with your latest observation data at 12 s temporal and 10 m vertical resolution (at least 45 minutes). For real-time inference, the value of the nighttime indicator is set based on the current time when the script is run. If you are interested in analysing historical data, this should be changed.
For full details of the methodology and citation, please refer to: Wijnands, J.S., Apituley, A., Alves Gouveia, D., Noteboom, J.W. (2024). Deep-Pathfinder: a boundary layer height detection algorithm based on image segmentation. Atmospheric Measurement Techniques, doi: https://doi.org/10.5194/amt-2023-80
This script has several dependencies. For example, the following shows how to set up a suitable Anaconda environment:
conda create --name Deep-Pathfinder python=3.10 -y
conda activate Deep-Pathfinder
conda install -y netCDF4
conda install -y -c conda-forge xarray dask bottleneck
python -m pip install "tensorflow==2.10"
pip install matplotlib
pip install opencv-python
pip install "suntime==1.2.5"
To apply the Deep-Pathfinder algorithm using the last 45 minutes of ceilometer data in the supplied NetCDF file, run the command below.
python Deep-Pathfinder_inference.py
The following results will be written to the data folder:
- The latest model input to check what information has been used for model inference
- A plot to visualise results, with the black line representing Deep-Pathfinder estimates
- Mixing layer height values added as a Deep-Pathfinder variable to a copy of the NetCDF input file