Analog Forecast

Make forecasts from similar meteorological conditions (analogs) using ERA5.

Instructions

To run the basic analog forecast program from the command line:

1. Create the conda environment via conda env create -f environment.yml
2. Activate the conda environment (conda activate analog-forecast)
3. Store the path to the ERA5 daily data in the DATA_DIR environment variable: export DATA_DIR=/atlas_scratch/kmredilla/analog_forecast/
4. Store the path to the root directory of the repository in the PROJECT_DIR environment variable: e.g. export PROJECT_DIR=path/to/analog-forecast (or export PROJECT_DIR=$PWD if you are in the repo already).
5. Run the analog_forecast.py script with the desired options (see the script header for more info):

python analog_forecast.py -v sst -d 2021-12-01 -sd alaska -fd alaska

Spatial domains

The following four spatial domains are available for both analog search and forecast (bbox bounds in W,S,E,N):

"alaska": (-180, 44, -125, 76)
"northern_hs": (-180, 0, 180, 90)
"panarctic": (-180, 55, 180, 90)
"north_pacific": (120, 0, 240, 90)

You must use these names exactly as they are spelled here (case-sensitive).

Other options

Available variables are: sst, t2m, msl, and z. You may choose any reference date within 1959-2021 if you do not have a CDS API key, as this is what is currently available in the historical ERA5 archive. You may choose any date if you do have the CDS API key set up, as data for more recent reference dates will be downloaded automatically (if available). Dates must be in YYYY-mm-dd format.

Contents

analog_forecast.py: this is the main script and module for executing the analog forecast. It can be called from the command line or imported and run in a separate python environment, such as a jupyter notebook (done in run_forecast.ipynb. config.py: config file with some useful constants luts.py: like config but for lookup tables (dicts) qc.ipynb: notebook for quality checking different parts of the algorithm run_forecast.ipynb: notebook for running the analog forecast program. Includes an interactive visualization of forecast error.

scripts/

Scripts and modules for preparing the data. Use these scripts to download the necessary data. To actually run any of these scripts, set the PROJECT_DIR environment variable to the path of this repo on the local filesystem and add it to the PYTHONPATH variable. E.g. run the following from the directory containing this file:

export PROJECT_DIR=$PWD
export PYTHONPATH=$PYTHONPATH:$PROJECT_DIR

skill_profiling/

The contents of this subfolder are for evaluating the skill of the analog forecast method. The project directory needs to be in the PYTHONPATH variable as well for working in here.

Also, if working from Atlas or Chinook, you will need a script for initializing conda in the shell generated by the slurm scheduler. So do this:

export CONDA_INIT_SCRIPT=/path/to/conda_init.sh

Where the conda_init.sh looks something like this:

__conda_setup="$('/home/UA/kmredilla/miniconda3/bin/conda' 'shell.bash' 'hook' 2> /dev/null)"
if [ $? -eq 0 ]; then
    eval "$__conda_setup"
else
    if [ -f "/home/UA/kmredilla/miniconda3/etc/profile.d/conda.sh" ]; then
        . "/home/UA/kmredilla/miniconda3/etc/profile.d/conda.sh"
    else
        export PATH="/home/UA/kmredilla/miniconda3/bin:$PATH"
    fi
fi
unset __conda_setup