Code implementing the methods described in the paper ATTRICI 1.1 - counterfactual climate for impact attribution in Geoscientific Model Development. The code is archived at ZENODO.
- All general settings are defined in settings.py.
- The code can be run with run_estimation.py or run_single_cell.py.
- The probability model for different climate variables is specified in models.py
- The choice of the probability model for a variable is specified in estimator.py
Please do
conda config --add channels conda-forge
conda create -n attrici pymc3==3.7 python==3.7
conda activate attrici
conda install netCDF4 pytables matplotlib arviz
pip install func_timeout
You may optionally
cp config/theanorc ~/.theanorc
The parallelization part to run large datasets is currently taylored to the supercomputer at the Potsdam Institute for Climate Impact Research using the slurm scheduler. Generalizing it into a package is ongoing work. We use the GNU compiler as the many parallel compile jobs through jobarrays and JIT compilation conflict with the few Intel licenses.
module purge
module load compiler/gnu/7.3.0
conda activate yourenv
In the root package directory.
pip install -e .
Override the conda setting with: export CXX=g++
Load input data from https://data.isimip.org
The input for one variable is a single netcdf file containing all time steps.
Create smoothed gmt time series as predictor using preprocessing/calc_gmt_by_ssa.py with adjusted file-paths.
Get auxiliary tasskew and tasrange time series using preprocessing/create_tasrange_tasskew.py with adjusted file-paths.
Adjust settings.py
For estimating parameter distributions (above step 1) and smaller datasets
python run_estimation.py
For larger datasets, produce a submit.sh file via
python create_submit.py
Then submit to the slurm scheduler
sbatch submit.sh
For merging the single timeseries files to netcdf datasets
python merge_cfact.py
Copy the settings.py, run_estimation.py, merge_cfact.py and submit.sh to a separate directory,
for example myrunscripts. Adjust settings.py and submit.sh, in particular the output directoy, and continue as in Usage.
Example for GSWP3-W5E5 dataset, which is first priority in ISIMIP3a.
cd preprocess
Download decadal data and merge it into one file per variable.
Adjust output paths and
python merge_data.py
Approximately 1 hour.
Produce GMT from gridded surface air temperature and use SSA to smooth it.
Use a separate conda env to cover SSA package dependencies.
Adjust output paths and
python calc_gmt_by_ssa.py
Approximately less than an hour.
Create tasrange and tasskew from tas variables.
Adjust output paths and
python create_tasmin_tasmax.py
Approximately an hour.
For testing smaller dataset, use
python subset_data.py
Add sub01 to filenames, if no subsetting is used.
Land-sea file creation We use the ISIMIP2b land-sea mask to select only land cells for processing. Smaller datasets through subsetting were created using CDO.
For tasmin and tasmax, we do not estimate counterfactual time series individually to avoid large relative errors in the daily temperature range as pointed out by (Piani et al. 2010). Following (Piani et al. 2010), we estimate counterfactuals of the daily temperature range tasrange = tasmax - tasmin and the skewness of the daily temperature tasskew = (tas - tasmin) / tasrange. Use create_tasmin_tasmax.py with adjusted paths for the tas, tasrange and tasskew counterfactuals.
A counterfactual huss is derived from the counterfacual tas, ps and hurs using the equations of Buck (1981) as described in Weedon et al. (2010). Use derive_huss.sh with adjusted file names and the required time range.
We rely on the pymc3 package for probabilistic programming (Salvatier et al. 2016).
An early version of the code on Bayesian estimation of parameters in timeseries with periodicity in PyMC3 was inspired by Ritchie Vink's post on Bayesian timeseries analysis with additive models.
This code is licensed under GPLv3, see the LICENSE.txt. See commit history for authors.