Radiative_Response_with_Radiative_Kernel

Use the radiative kernel (Soden et al. 2008) to diagnose the TOA radiative response $dR_i$ due to change in climate variables i: temperature (ta and ts), water vapor (wv), albedo and cloud.

Then we can compute the climate feedbacks:

$$\lambda_i = \frac{dR_i}{dts_{gm}}$$

where ${dts_{gm}}$ is the global mean surface temp. change.

Usage

Create the Python environment.

use r3k_env.yml file

conda env create -f r3k_env.yml
conda activate r3k_env
python -m ipykernel install --user --name r3k_env --display-name "r3k_env" # Register the kernel for jupyter

or

conda create -y -n r3k_env python=3.10
conda activate r3k_env
conda install  -y  -c conda-forge numpy numba scikit-learn matplotlib
conda install  -y  -c conda-forge xarray netCDF4 ipykernel 
python -m ipykernel install --user --name r3k_env --display-name "r3k_env" # Register the kernel for jupyter

Download the kernel file and the example data (4G)

wget https://tigress-web.princeton.edu/~cw55/share_data/r3k_example_data.tar
wget https://raw.githubusercontent.com/ChenggongWang/Radiative_Response_with_Radiative_Kernel/main/Radiative_Repsonse_with_Raditive_kernel.py -O Radiative_Repsonse_with_Raditive_kernel.py

Run example notebook (requires the example data below).

Example

The example notebook (r3k_example.ipynb) shows how to compute the climate feedback of GFDL-CM4 model (using the abrupt-4xCO2 and piControl experiments). The raw data is available at CMIP6 data nodes LLNL. But we have to regrid the original data to the same resolution as the kernel file (or the other way around). The regridded data can be downloaded from google drive or princeton.edu compressed file.

Decription

decompose_dR_rk_toa_core(var_pert, var_cont,f_RK, forced=True) is the core function to call. The option forced=True will estimate the all-sky forcing using clear-sky forcing. Set forced=False for experiments without TOA radiative forcing (e.g. fixed SST experiment). It will return a xarray dataset that contains variables as follows (also its global-mean, variable name with _gm):

dR_wv_lw : $\frac{\partial R_{all-sky\ lw}}{\partial wv}\Delta wv ,\qquad lw\ R_{toa}$ change due to water vapor(wv) change

dR_wv_sw : $\frac{\partial R_{all-sky\ sw}}{\partial wv}\Delta wv $

dR_wvcs_lw : $\frac{\partial R_{clr-sky\ lw}}{\partial wv}\Delta wv $

dR_wvcs_sw : $\frac{\partial R_{clr-sky\ sw}}{\partial wv}\Delta wv $

dR_ta : $\frac{\partial R_{all-sky\ lw}}{\partial ta}\Delta ta ,\qquad lw\ R_{toa}$ change due to air temp.(ta) change (including lapse rate change)

dR_tacs : $\frac{\partial R_{clr-sky\ lw}}{\partial ta}\Delta ta $

dR_lr : $\frac{\partial R_{all-sky\ lw}}{\partial lr}\Delta lr ,\qquad lw\ R_{toa}$ change due to lapse rate(lr) change (vertial structure differs from dts)

dR_lrcs : $\frac{\partial R_{clr-sky\ lw}}{\partial lr}\Delta lr $

dR_ts : $\frac{\partial R_{all-sky\ lw}}{\partial ts}\Delta ts ,\qquad lw\ R_{toa}$ change due to surface temp.(ts) change (including lapse rate change)

dR_tscs : $\frac{\partial R_{clr-sky\ lw}}{\partial ts}\Delta ts $

dR_alb : $\frac{\partial R_{all-sky\ sw}}{\partial albedo}\Delta albedo ,\qquad sw\ R_{toa}$ change due to surface albedo(alb) change

dR_albcs : $\frac{\partial R_{clr-sky\ sw}}{\partial albedo}\Delta albedo$

dR_cloud_lw : $\frac{\partial R_{all-sky\ sw}}{\partial cloud}\Delta cloud ,\qquad lw\ R_{toa}$ change due to cloud change (computed as residual)

Dcs_lw : $dF_{clr-sky\ lw}$ estimated clear-sky lw forcing

Dcs_sw : $dF_{clr-sky\ sw}$ estimated clear-sky sw forcing

dR_sw : $dR_{all-sky\ sw}$ all-sky TOA sw net flux change

dR_lw : $dR_{all-sky\ lw}$ all-sky TOA lw net flux change

dRcs_sw : $dR_{clr-sky\ sw}$ clear-sky TOA sw net flux change

dRcs_lw : $dR_{clr-sky\ lw}$ clear-sky TOA lw net flux change

ts : $ts$ surface temperature

dts : $dts$ 'surface temperature change'

Decomposition base on RH:

dR_rh_lw : $\frac{\partial R_{all-sky\ lw}}{\partial rh}\Delta rh ,\qquad lw\ R_{toa}$ change due to relative humidity(rh) change

dR_rh_sw : $\frac{\partial R_{all-sky\ sw}}{\partial rh}\Delta rh $

dR_rhcs_lw : $\frac{\partial R_{clr-sky\ lw}}{\partial rh}\Delta rh $

dR_rhcs_sw : $\frac{\partial R_{clr-sky\ sw}}{\partial rh}\Delta rh $

dR_Ta_rh_lw : $\frac{\partial R_{all-sky\ lw}}{\partial ta}\mid_{RH} \Delta ta ,\qquad lw\ R_{toa}$ change due to air temp.(ta) change (including lapse rate change) when relative humidity(rh) is fixed

dR_Ta_rh_sw : $\frac{\partial R_{all-sky\ sw}}{\partial ta}\mid_{RH} \Delta ta $

dR_Tacs_rh_lw : $\frac{\partial R_{clr-sky\ lw}}{\partial ta}\mid_{RH} \Delta ta $

dR_Tacs_rh_sw : $\frac{\partial R_{clr-sky\ sw}}{\partial ta}\mid_{RH} \Delta ta $

Note

Everything is common (xarray to load/create netcdf data, numpy, matplotlib to show results) except Numba.

Numba is the core package and is used to parallel/accelerate the computation (recommended for large datasets or many datasets/experiments).

For jobs that are not time sensitive, you should be able to use numpy or xarray version of functions without Numba.

A version of functions that use only numpy/xarray and is easier to understand is also provided for understanding and modification (see the benchmark code).

The time for 150 years 2x2.5 [latxlon] data (~4GB) is ~ 10 seconds on princeton jupyterhub (expect similar time for CPUs>=4).

Numpy version takes ~ 30 seconds (single CPU).

Xarray version takes 1~2 mins (single CPU).

To do plan

Add regrid option for convenience