PLS, data, and plotting for: Quantifying the radiative response to surface temperature variability: A critical comparison of current methods

The files in this folder contribute to the quantitative analysis, data, and plotting for the publication "Quantifying the radiative response to surface temperature variability: A critical comparison of current methods" by Leif Fredericks, Maria Rugenstein, David W.J. Thompson, Senne Van Loon, Fabrizio Falasca, Rory Basinski-Ferris, Paulo Ceppi, Quran Wu, Jonah Bloch-Johnson, Marc Alessi, and Sarah M. Kang. 

Scripts:

PLS_fp.ipynb: This notebook demonstrates the implementation of the Partial Least Squares method following the test and train protocol followed for all methods.

patches_fp.ipynb: This notebook reads in CNN-generated and ECHAM-generated Green's function results for the full-globe and reduced size Pacific patches, which generate the subplots for Fig. 3.

test_figure_fp.ipynb: This notebook generates Fig. 2 from the internal variability test results.

Data:

GF_CNN_ridge_full_patches.nc: This dataset includes the CNN-generated Green's Functions for the full globe with standard patches, Fig. 3c.

GF_CNN_ridge_Pacific_patches.nc: This dataset includes the CNN-generated and ECHAM6-simulated Green's Functions for the Equatorial Pacific using smaller patches, FIgs. 3f and 3g, respectively.

training_ds.nc: This dataset includes the portion of MPI-ESM1.2 pre-industrial control from the LongRunMIP project that was processed and distributed for training the various methods. It does not include the section of piControl reserved for the internal variabilty test. Annual and monthly means are detrended anomalies from climatology. The fields are near-surface air tempearture and global net radiaitve imbalance.

pred_4x.nc: This dataset includes the near-surface air temperature anomalies as simulated in a 4xCO2 step forcing simulation. These are what methods use to predict radiation for the 4xCO2 test.

testing_pred.nc: This dataset includes near-surface air temperature from the section of piControl internal variability withheld from the training data in training_ds.nc. Methods attempt to predict the corresponding radiative response in the internal variabilty test.

truth_4x.nc: These are the true radiative response values for the 4xCO2 test.

truth_test.nc: These are the true radiative response values for the internal variabilty test. 

The following outputs are the method-specific outputs that make up Figs. 1&2. File names have standard identifiers as follows:

MCA: Maximum Covariance Analysis

OLS: Ordinary Least Squares

ridge: Ridge Regression

LASSO: LASSO Regression

EN: Elastic Net Regression

PC: Principal Component Regression

PLS: Partial Least Squares

FDR_coupled: Fluctuation Dissipation Relation, Coupled

FDR_atmos: Fluctuation Dissipation Relation, Atmospheric

CNN: Convolutional Neural Network

GF: Green's Function

AMIP: AMIP Ridge Regression

Sensitivity maps: dR_dT_*.nc

Predictions for the internal variability test: pred_test_*.npy

Predictions for the 4xCO2 test: pred_4x_*.npy