Data from: Convolutional neural networks trained on internal variability predict forced response of TOA radiation by learning the pattern effect
Data files
Feb 13, 2025 version files 630.04 KB
-
GRAD_MPI_IV_hpt7_s0.nc
85.19 KB
-
GRAD_MultiModelELU_s3.nc
42.18 KB
-
GRAD_SingleModel_CanESM5_linear_s2.nc
42.18 KB
-
GRAD_SingleModel_CanESM5_s2.nc
42.18 KB
-
GRAD_SingleModel_IPSL_linear_s3.nc
42.18 KB
-
GRAD_SingleModel_IPSL_s3.nc
42.18 KB
-
GRAD_SingleModel_MIROC6_linear_s3.nc
42.18 KB
-
GRAD_SingleModel_MIROC6_s3.nc
42.18 KB
-
GRAD_SingleModel_MPI_linear_s2.nc
42.18 KB
-
GRAD_SingleModel_MPI_s2.nc
42.18 KB
-
Predictions_1pctCO2.nc
15.22 KB
-
Predictions_hist_rcp85.nc
17.37 KB
-
Predictions_IV.nc
16.46 KB
-
Predictions_MultiModel_CanESM5.nc
17.73 KB
-
Predictions_MultiModel_IPSL.nc
17.73 KB
-
Predictions_MultiModel_IV.nc
42.21 KB
-
Predictions_MultiModel_MIROC6.nc
17.73 KB
-
Predictions_MultiModel_MPI.nc
17.73 KB
-
README.md
3.04 KB
Abstract
Predicting forced, long-term radiative feedbacks from internal climate variability has been a decades-long quest in climate science. We train a convolutional neural network (CNN) to predict annual- and global-mean top of the atmosphere radiation anomalies from time-varying maps of near-surface temperature in climate models. Trained on internal variability alone, the nonlinear CNN can predict radiation under strong climate change, outperforms a regularized linear regression approach, and works within and across different climate models. We show with explainable artificial intelligence methods that the CNN draws predictive skill from physically meaningful regions but at much smaller spatial scales than currently assumed.
Convolutional neural networks to predict global-mean radiation from surface temperature maps. Code available from https://github.com/SnnVL/CNN_PatternEffect
Tensorflow Code
This code was written in python 3.10.13, tensorflow 2.15.0, tensorflow-probability 0.15.0 and numpy 1.26.4.
The CNNs are trained by running _train_driver.py. All experiment settings are contained in experiment_settings.py. For plotting and interpreting the results, jupyter notebooks were created.
File description
All files are provided in the netCDF format .nc, and contain the data needed to recreate the figures in Ref. 1 (Rugenstein et al., 2025). Specifically:
predictions_*.ncfiles contain the predictions and truth of $R$ made from CNNs with different SST inputsprediction_IV.nccontains the data for Fig. 1a (CNN/linear network trained on MPI data, predicting internal variability in MPI)prediction_hist_rcp85.nccontains the data for Fig. 1b (CNN/linear network trained on MPI, predicting historical + RCP8.5 in MPI)prediction_1pctCO2.nccontains the data for Fig. 1c (CNN/linear network trained on MPI, predicting 1% CO_2 experiment in MPI)predictions_MultiModel_*.nccontains data for Fig. 3a-b. It contains the predictions of the CNN that was trained on 4 different climate models. The final part of the file name quantifies which SST data is used to make predictions (IV: Internal Variability;IPSL: historical + SSP2-4.5 from IPSL-CM6A-LR;CanESM5: historical + SSP2-4.5 from CanESM5;MIROC6: historical + SSP2-4.5 from MIROC6;MPI: historical + RCP8.5 from MPI-ESM)
GRAD_*.ncfiles contain the Gradients of the CNN/linear networksGRAD_MPI_IV_hpt7_s0.nccontains the gradient of the MPI-ESM CNN on the original grid (Fig. 2a)GRAD_SingleModel_*.nccontains the gradients for the single model CNNs; the second part of the file name denotes the model used. If the_linearflag is added, it contains the linear network gradient instead of the CNN gradient.GRAD_MultiModelELU_s3.nccontains the gradient of the multi-model CNN (Fig. 3c)- the
s*naming denotes the seed used to train the CNN or linear network.
Credits
This work is a collaborative effort between Dr. Maria Rugenstein, Dr. Senne Van Loon, and Dr. Elizabeth A. Barnes.
References
[1] Maria Rugenstein, Senne Van Loon, & Elizabeth A. Barnes (2025), Convolutional neural networks trained on internal variability predict forced response of TOA radiation by learning the pattern effect, Geophysical Research Letters (accepted).
[2] Senne Van Loon, Maria Rugenstein, & Elizabeth A. Barnes (2025), Observation-based estimate of Earth’s effective radiative forcing, EarthArXiv