Tools
Tools
Mahesh, Ankur, O'Brien, Travis A., Loring, Burlen, Elbashandy, Abdelrahman, Boos, William and Collins, William D. 
ORCID: https://orcid.org/0000-0002-4463-9848
(2024)
Identifying atmospheric rivers and their poleward latent heat transport with generalizable neural networks:ARCNNv1.
Geoscientific Model Development, 17 (8).
pp. 3533-3557.
ISSN 1991-959X
Preview |
PDF (gmd-17-3533-2024)
- Published Version
Available under License Creative Commons Attribution. Download (10MB) | Preview |
Abstract
Atmospheric rivers (ARs) are extreme weather events that can alleviate drought or cause billions of US dollars in flood damage. By transporting significant amounts of latent energy towards the poles, they are crucial to maintaining the climate system's energy balance. Since there is no first-principle definition of an AR grounded in geophysical fluid mechanics, AR identification is currently performed by a multitude of expert-defined, threshold-based algorithms. The variety of AR detection algorithms has introduced uncertainty into the study of ARs, and the thresholds of the algorithms may not generalize to new climate datasets and resolutions. We train convolutional neural networks (CNNs) to detect ARs while representing this uncertainty; we name these models ARCNNs. To detect ARs without requiring new labeled data and labor-intensive AR detection campaigns, we present a semi-supervised learning framework based on image style transfer. This framework generalizes ARCNNs across climate datasets and input fields. Using idealized and realistic numerical models, together with observations, we assess the performance of the ARCNNs. We test the ARCNNs in an idealized simulation of a shallow-water fluid in which nearly all the tracer transport can be attributed to AR-like filamentary structures. In reanalysis and a high-resolution climate model, we use ARCNNs to calculate the contribution of ARs to meridional latent heat transport, and we demonstrate that this quantity varies considerably due to AR detection uncertainty.
| Item Type: | Article |
|---|---|
| Additional Information: | Code and data availability: The file directory listing can be viewed at https://portal.nersc.gov/archive/home/a/amahesh/www/GMD_ARCNNs/ (Mahesh, 2024). Please note that the files are stored on tape, so there may be delays associated with retrieving the files from the tape filesystem. We have made nine datasets open-source: the ARCI labels on MERRA-2 IWV at https://portal.nersc.gov/archive/home/a/amahesh/www/GMD_ARCNNs/artmip_probabilistic_labels_all_vars.tar (last access: 28 April 2024); the ARCI labels on MERRA-2 IVT (same link as above); The ARCI labels on GridSat at https://portal.nersc.gov/archive/home/a/amahesh/www/GMD_ARCNNs/GRIDSAT.tar (last access: 28 April 2024); the ARCI labels on ERA-I IVT at https://portal.nersc.gov/archive/home/a/amahesh/www/GMD_ARCNNs/era_probabilistic_labels.tar (last access: 28 April 2024); the ARCI labels on ERA 20th Century Reanalysis IVT at https://portal.nersc.gov/archive/home/a/amahesh/www/GMD_ARCNNs/era_20cr.tar (last access: 28 April 2024); ARCNN detections in MERRA-2 IVT at https://portal.nersc.gov/archive/home/a/amahesh/www/GMD_ARCNNs/merra2_nn_preds.tar (last access: 28 April 2024), with these predictions using the ARCNN Experiment 1 in Table 1; ECMWF-IFS-HR IVT at https://portal.nersc.gov/archive/home/a/amahesh/www/GMD_ARCNNs/HighResMIP-ECMWF-IFS-HR_IVT.tar (last access: 28 April 2024); ARCNN detections in ECMWF-IFS-HR IVT at https://portal.nersc.gov/archive/home/a/amahesh/www/GMD_ARCNNs/ecmwf-ifs-hr_nn_preds.tar (last access: 28 April 2024). These predictions use the ARCNN Experiment 6 in Table 1; and the idealized single-layer simulation at https://portal.nersc.gov/archive/home/a/amahesh/www/GMD_ARCNNs/idealized_ar.tar (last access: 28 April 2024). At the Zenodo DOI (https://doi.org/10.5281/zenodo.7814401, Mahesh et al., 2023), we also provide open-source code for the following four tasks: our code to run the Atmospheric Rivers textbook threshold algorithm, our code to train the ARCNNs, our code to run the idealized climate simulation, and a tutorial of how to use our loss function and models. We recommend the tutorial as a starting point for users to understand how to use the loss function to train their own CNN. We also made the learned parameters open-source for the following six trained ARCNNs: one for MERRA-2 (using either IWV or IVT as input); one for GridSat (using the brightness temperatures in the infrared window); one for ERA-I (using IVT as input); one for ERA 20th Century Reanalysis (using IVT as input); one for C20C+ CAM5 (using IVT as input); and one for ECMWF-IFS-HR (using IVT as input). In the above tutorial, we include instructions on how to load the model (using PyTorch) and generate AR detections with it. These are available at the above Zenodo repository under the ar_segmentation_tutorial/trained_models folder. |
| Uncontrolled Keywords: | modelling and simulation,general earth and planetary sciences,sdg 13 - climate action ,/dk/atira/pure/subjectarea/asjc/2600/2611 |
| Faculty \ School: | Faculty of Science > School of Environmental Sciences |
| Related URLs: | |
| Depositing User: | LivePure Connector |
| Date Deposited: | 21 Jul 2026 15:02 |
| Last Modified: | 21 Jul 2026 15:02 |
| URI: | https://ueaeprints.uea.ac.uk/id/eprint/103938 |
| DOI: | 10.5194/gmd-17-3533-2024 |
Downloads
Downloads per month over past year
Actions (login required)
 |
View Item |