Collection of bias correction procedures for single and multidimensional climate data
Project description
python-cmethods
This Python module serves as a collection of different scale- and distribution-based bias correction techniques for climate sciences.
The documentation is available at: https://python-cmethods.readthedocs.io/en/stable/
Please cite this project as described in https://zenodo.org/doi/10.5281/zenodo.7652755.
⚠️ For the application of bias corrections on large data sets it is recommended to also try the command-line tool BiasAdjustCXX.
Table of Contents
1. About
These programs and data structures are developed with the aim of reducing discrepancies between modeled and observed climate data. Historical data is utilized to calibrate variables from current and future time series to achieve distributional properties that closely resemble the possible actual values.
For instance, modeled data typically indicate values that are colder than the actual values. To address this issue, an adjustment procedure is employed. The figure below illustrates the observed, modeled, and adjusted values, revealing that the delta adjusted time series ($T^{*DM}{sim,p}$) are significantly more similar to the observed data ($T{obs,p}$) than the raw modeled data ($T{sim,p}$).
2. Available Methods
python-cmethods provides the following bias correction techniques:
- Linear Scaling
- Variance Scaling
- Delta Method
- Quantile Mapping
- Detrended Quantile Mapping
- Quantile Delta Mapping
Please refer to the official documentation for more information about these methods as well as sample scripts: https://python-cmethods.readthedocs.io/en/stable/
-
Except for the variance scaling, all methods can be applied on stochastic and non-stochastic climate variables. Variance scaling can only be applied on non-stochastic climate variables.
-
Non-stochastic climate variables are those that can be predicted with relative certainty based on factors such as location, elevation, and season. Examples of non-stochastic climate variables include air temperature, air pressure, and solar radiation.
-
Stochastic climate variables, on the other hand, are those that exhibit a high degree of variability and unpredictability, making them difficult to forecast accurately. Precipitation is an example of a stochastic climate variable because it can vary greatly in timing, intensity, and location due to complex atmospheric and meteorological processes.
-
-
Except for the detrended quantile mapping (DQM) technique, all methods can be applied to 1- and 3-dimensional data sets. The implementation of DQM to 3-dimensional data is still in progress.
-
Except for DQM, all methods can be applied using
cmethods.adjust
. Chunked data for computing e.g. in a dask cluster is possible as well. -
For any questions -- please open an issue at https://github.com/btschwertfeger/python-cmethods/issues
3. Installation
python3 -m pip install python-cmethods
4. Usage and Examples
import xarray as xr
from cmethods import adjust
obsh = xr.open_dataset("input_data/observations.nc")
simh = xr.open_dataset("input_data/control.nc")
simp = xr.open_dataset("input_data/scenario.nc")
# adjust only one grid cell
ls_result = adjust(
method="linear_scaling",
obs=obsh["tas"][:, 0, 0],
simh=simh["tas"][:, 0, 0],
simp=simp["tas"][:, 0, 0],
kind="+",
group="time.month",
)
# adjust all grid cells
qdm_result = adjust(
method="quantile_delta_mapping",
obs=obsh["tas"],
simh=simh["tas"],
simp=simp["tas"],
n_quantiles=1000,
kind="+",
)
# to calculate the relative rather than the absolute change,
# '*' can be used instead of '+' (this is preferred when adjusting
# stochastic variables like precipitation)
It is also possible to adjust chunked data sets. Feel free to have a look into
tests/test_zarr_dask_compatibility.py
to get a starting point.
Notes:
- For the multiplicative techniques a maximum scaling factor of 10 is defined.
This can be changed by passing the optional parameter
max_scaling_factor
. - Except for detrended quantile mapping, all implemented techniques can be
applied to single and multdimensional data sets by executing the
cmethods.adjust
function.
Examples (see repository on GitHub)
Notebook with different methods and plots: /examples/examples.ipynb
There is also an example script (/examples/biasadjust.py
) that can be used to apply the available bias correction methods
on 1- and 3-dimensional data sets (see /examples/input_data/*.nc
).
Help:
╰─ python3 biasadjust.py --help
(1.) Example - Quantile Mapping bias correction on the provided example data:
╰─ python3 biasadjust.py \
--ref input_data/observations.nc \
--contr input_data/control.nc \
--scen input_data/scenario.nc \
--kind "+" \
--variable "tas" \
--quantiles 10 \
--method quantile_mapping
(2.) Example - Linear Scaling bias correction on the provided example data:
╰─ python3 biasadjust.py \
--ref input_data/observations.nc \
--contr input_data/control.nc \
--scen input_data/scenario.nc \
--kind "+" \
--variable "tas" \
--group "time.month" \
--method linear_scaling
Notes:
- Data sets must have the same spatial resolutions.
- This script is far away from perfect - so please see it, as a starting point. (:
5. Notes
- Computation in Python takes some time, so this is only for demonstration. When adjusting large datasets, you should either use chunked data using for example a dask cluster or to apply the command-line tool BiasAdjustCXX.
- Formulas and references can be found in the implementations of the corresponding functions, on the bottom of the README.md and in the documentation.
Space for improvements
- Since the scaling methods implemented so far scale by default over the mean
values of the respective months, unrealistic long-term mean values may occur
at the month transitions. This can be prevented either by selecting
group='time.dayofyear'
. Alternatively, it is possible not to scale using long-term mean values, but using a 31-day interval, which takes the 31 surrounding values over all years as the basis for calculating the mean values. This is not yet implemented, because even the computation for this takes so much time, that it is not worth implementing it in python - but this is available in BiasAdjustCXX.
6. 🆕 Contributions
… are welcome but:
- First check if there is an existing issue or PR that addresses your problem/solution. If not - create one first - before creating a PR.
- Typo fixes, project configuration, CI, documentation or style/formatting PRs will be rejected. Please create an issue for that.
- PRs must provide a reasonable, easy to understand and maintain solution for an existing problem. You may want to propose a solution when creating the issue to discuss the approach before creating a PR.
7. References
- Schwertfeger, Benjamin Thomas and Lohmann, Gerrit and Lipskoch, Henrik (2023) "Introduction of the BiasAdjustCXX command-line tool for the application of fast and efficient bias corrections in climatic research", SoftwareX, Volume 22, 101379, ISSN 2352-7110, (https://doi.org/10.1016/j.softx.2023.101379)
- Schwertfeger, Benjamin Thomas (2022) "The influence of bias corrections on variability, distribution, and correlation of temperatures in comparison to observed and modeled climate data in Europe" (https://epic.awi.de/id/eprint/56689/)
- Linear Scaling and Variance Scaling based on: Teutschbein, Claudia and Seibert, Jan (2012) "Bias correction of regional climate model simulations for hydrological climate-change impact studies: Review and evaluation of different methods" (https://doi.org/10.1016/j.jhydrol.2012.05.052)
- Delta Method based on: Beyer, R. and Krapp, M. and Manica, A.: "An empirical evaluation of bias correction methods for palaeoclimate simulations" (https://doi.org/10.5194/cp-16-1493-2020)
- Quantile and Detrended Quantile Mapping based on: Alex J. Cannon and Stephen R. Sobie and Trevor Q. Murdock "Bias Correction of GCM Precipitation by Quantile Mapping: How Well Do Methods Preserve Changes in Quantiles and Extremes?" (https://doi.org/10.1175/JCLI-D-14-00754.1)
- Quantile Delta Mapping based on: Tong, Y., Gao, X., Han, Z. et al. "Bias correction of temperature and precipitation over China for RCM simulations using the QM and QDM methods". Clim Dyn 57, 1425–1443 (2021). (https://doi.org/10.1007/s00382-020-05447-4)
- I'd like to express my gratitude to @riley-brady for initiating and contributing to the discussion on https://github.com/btschwertfeger/python-cmethods/issues/47. I appreciate all the valuable suggestions provided throughout the implementation of the subsequent changes.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distributions
Built Distribution
Hashes for python_cmethods-2.0.1-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | fd87397a363c1d57975c057dd49b951fb16009d3744c46ff1faf22acb9ba81b5 |
|
MD5 | a129a0e1124543a6fcb8b9fdaefe0ac1 |
|
BLAKE2b-256 | ebd982c00318f3635352bc2cf34e57c5d860f992eb8a331648abb68720e796f6 |