On the Extrapolation of Generative Adversarial Networks for downscaling precipitation extremes in warmer climates

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Rampal, Neelesh, Gibson, Peter B., Sherwood, Steven, Abramowitz, Gab
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913599384977408
author Rampal, Neelesh
Gibson, Peter B.
Sherwood, Steven
Abramowitz, Gab
author_facet Rampal, Neelesh
Gibson, Peter B.
Sherwood, Steven
Abramowitz, Gab
contents While deep-learning downscaling algorithms can generate fine-scale climate projections cost-effectively, it is still unclear how well they will extrapolate to unobserved climates. We assess the extrapolation capabilities of a deterministic Convolutional Neural Network baseline and a Generative Adversarial Network (GAN) built with this baseline, trained to predict daily precipitation simulated by a Regional Climate Model (RCM). Both approaches emulate future changes in annual mean precipitation well, even when trained on historical data, though training on a future climate improves performance. For extreme precipitation (99.5th percentile), RCM simulations predict a robust end-of-century increase with future warming (~5.8%/°C on average from five simulations). When trained on a future climate, GANs capture 97% of the warming-driven increase in extreme precipitation compared to 65% in a deterministic baseline. Even GANs trained historically capture 77% of this increase. Overall, GANs offer better generalization for downscaling extremes, which is important in applications relying on historical data.
format Preprint
id arxiv_https___arxiv_org_abs_2409_13934
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle On the Extrapolation of Generative Adversarial Networks for downscaling precipitation extremes in warmer climates
Rampal, Neelesh
Gibson, Peter B.
Sherwood, Steven
Abramowitz, Gab
Atmospheric and Oceanic Physics
While deep-learning downscaling algorithms can generate fine-scale climate projections cost-effectively, it is still unclear how well they will extrapolate to unobserved climates. We assess the extrapolation capabilities of a deterministic Convolutional Neural Network baseline and a Generative Adversarial Network (GAN) built with this baseline, trained to predict daily precipitation simulated by a Regional Climate Model (RCM). Both approaches emulate future changes in annual mean precipitation well, even when trained on historical data, though training on a future climate improves performance. For extreme precipitation (99.5th percentile), RCM simulations predict a robust end-of-century increase with future warming (~5.8%/°C on average from five simulations). When trained on a future climate, GANs capture 97% of the warming-driven increase in extreme precipitation compared to 65% in a deterministic baseline. Even GANs trained historically capture 77% of this increase. Overall, GANs offer better generalization for downscaling extremes, which is important in applications relying on historical data.
title On the Extrapolation of Generative Adversarial Networks for downscaling precipitation extremes in warmer climates
topic Atmospheric and Oceanic Physics
url https://arxiv.org/abs/2409.13934