Bias Correcting Climate Model Simulations Using Unpaired Image-to-Image Translation Networks

Fulton, D. James; Clarke, Ben; Hegerl, Gabriele C.

doi:10.1175/aies-d-22-0031.1

Cited by 7 publications

(11 citation statements)

References 36 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It is worth some discussion of why multiple studies found GANs unable to correct the PDF of precipitation on their own (François et al., 2021; Fulton et al., 2023; Pan et al., 2021) unlike here and in Hess et al. (2022).…”

Section: Discussionmentioning

confidence: 70%

“…Many methodological differences might explain why we were able to better simulate the probability of extreme precipitation events without QM. We correct only precipitation, without using other model output fields as dynamical constraints (Pan et al., 2021) or additional fields to be corrected (François et al., 2021; Fulton et al., 2023). In addition, our model is trained on more data than the previous studies with PDF biases.…”

Section: Discussionmentioning

confidence: 99%

“…In the past, cycle‐generative networks have been effectively used for offline bias correction. Several studies found it necessary to augment the cycle‐generative network with quantile mapping (QM) to achieve accurate probability distributions of precipitation (François et al., 2021; Fulton et al., 2023; Pan et al., 2021), with one more recent study achieving success without quantile mapping (Hess et al., 2022). These works focused on correcting daily‐mean precipitation toward an observational analysis over a historical period.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Global Precipitation Correction Across a Range of Climates Using CycleGAN

McGibbon,

Clark,

Henn

et al. 2024

Geophysical Research Letters

View full text Add to dashboard Cite

Accurate precipitation simulations for various climate scenarios are critical for understanding and predicting the impacts of climate change. This study employs a Cycle‐generative adversarial network (CycleGAN) to improve global 3‐hr‐average precipitation fields predicted by a coarse grid (200 km) atmospheric model across a range of climates, morphing them to match their statistical properties with those of reference fine‐grid (25 km) simulations. We evaluate its performance on both the target climates and an independent ramped‐SST simulation. The translated precipitation fields remove most of the biases simulated by the coarse‐grid model in the mean precipitation climatology, the cumulative distribution function of 3‐hourly precipitation, and the diurnal cycle of precipitation over land. These results highlight the potential of CycleGAN as a powerful tool for bias correction in climate change simulations, paving the way for more reliable predictions of precipitation patterns across a wide range of climates.

show abstract

Section: Discussionmentioning

confidence: 70%

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Global Precipitation Correction Across a Range of Climates Using CycleGAN

McGibbon,

Clark,

Henn

et al. 2024

Geophysical Research Letters

View full text Add to dashboard Cite

show abstract

“…Alternatively, strategies such as generative adversarial networks (J. J. McGibbon et al, 2023) and unsupervised image-to-image networks (UNIT) (Fulton et al, 2023) have been used to correct biases in average precipitation rates-an integral quantity which is less affected by stochastic variation. While ML correction operators using a purely statistical loss function can indeed generate trajectories with plausible statistics, this property alone does not guarantee the resulted spatio-temporal dynamics are always physically realistic.…”

Section: Introductionmentioning

confidence: 99%

A Non‐Intrusive Machine Learning Framework for Debiasing Long‐Time Coarse Resolution Climate Simulations and Quantifying Rare Events Statistics

Barthel Sorensen,

Charalampopoulos,

Zhang

et al. 2024

J Adv Model Earth Syst

View full text Add to dashboard Cite

Due to the rapidly changing climate, the frequency and severity of extreme weather is expected to increase over the coming decades. As fully‐resolved climate simulations remain computationally intractable, policy makers must rely on coarse‐models to quantify risk for extremes. However, coarse models suffer from inherent bias due to the ignored “sub‐grid” scales. We propose a framework to non‐intrusively debias coarse‐resolution climate predictions using neural‐network (NN) correction operators. Previous efforts have attempted to train such operators using loss functions that match statistics. However, this approach falls short with events that have longer return period than that of the training data, since the reference statistics have not converged. Here, the scope is to formulate a learning method that allows for correction of dynamics and quantification of extreme events with longer return period than the training data. The key obstacle is the chaotic nature of the underlying dynamics. To overcome this challenge, we introduce a dynamical systems approach where the correction operator is trained using reference data and a coarse model simulation nudged toward that reference. The method is demonstrated on debiasing an under‐resolved quasi‐geostrophic model and the Energy Exascale Earth System Model (E3SM). For the former, our method enables the quantification of events that have return period two orders longer than the training data. For the latter, when trained on 8 years of ERA5 data, our approach is able to correct the coarse E3SM output to closely reflect the 36‐year ERA5 statistics for all prognostic variables and significantly reduce their spatial biases.

show abstract

“…Mahto and Mishra (2019) determined that ERA5 outperforms alternative reanalysis products, establishing its suitability for hydrological assessments over India. Recent studies have used ERA5 as a reference data to bias‐correct the climate model simulation over South Asia (Fulton et al, 2023). Due to the coarse resolution (1° × 1°) of IMD gridded temperature dataset, the ERA5 SAT reanalysis data (0.1° × 0.1°) were used as the reference data in this study to carry out bias correction for the CMIP6 historical and future projections.…”

Section: Introductionmentioning

confidence: 99%

Application of deep learning algorithms to correct bias in CMIP6 simulations of surface air temperature over the Indian monsoon core region

Sabarinath,

Naga Rajesh,

Gunthe

et al. 2023

Intl Journal of Climatology

View full text Add to dashboard Cite

Indian subcontinent witnessed a rise in surface air temperature (SAT) in recent decades, during the summer months of March, April and May. The monsoon core region (MCR) of India experiences a hot and humid climate, with temperatures typically highest in May and June before the onset of the monsoon. Global climate model (GCM) simulations of SAT are very much essential to understand the future climate of Indian MCR. Biases in GCMs simulations are due to insufficient knowledge of parameterizations and various assumptions that are made to simulate the complex interactions between land, ocean and atmosphere. The objective of this study is to correct the bias in the Coupled Model Intercomparison Project Phase 6 (CMIP6)–GCM simulations of SAT during March, April and May months over MCR for the historical period 1985–2014 and shared socio‐economic pathways (SSPs) SSP2‐4.5 and SSP5‐8.5 for the period 2015–2100. SAT dataset of fifth‐generation reanalysis (ERA5) of the European Centre for Medium‐Range Weather Forecasts (ECMWF) is used as reference dataset to perform bias correction for the historical period. Preliminary investigation of both SAT datasets has shown that there exists considerable warm bias (1.47°C) over the MCR. Bias correction is performed using a one‐dimensional convolutional neural network (CNN‐1D) and a convolutional long short‐term memory network (CNN‐LSTM) deep learning algorithm. The performance of these algorithms is evaluated with the statistical metrics such as root‐mean‐square error (RMSE), normalized root‐mean‐square error, Nash–Sutcliffe efficiency, mean absolute error, percent bias, correlation coefficient and dynamic time warping. RMSE and percent bias were decreased to 0.35°C and 0.8% with CNN‐LSTM algorithm. The CNN‐LSTM algorithm also preserves the year‐to‐year variability of SAT. Hence, CNN‐LSTM algorithm is found to be suitable for the bias correction of GCM simulations of SAT with encouraging results.

show abstract

Bias Correcting Climate Model Simulations Using Unpaired Image-to-Image Translation Networks

Cited by 7 publications

References 36 publications

Global Precipitation Correction Across a Range of Climates Using CycleGAN

Global Precipitation Correction Across a Range of Climates Using CycleGAN

A Non‐Intrusive Machine Learning Framework for Debiasing Long‐Time Coarse Resolution Climate Simulations and Quantifying Rare Events Statistics

Application of deep learning algorithms to correct bias in CMIP6 simulations of surface air temperature over the Indian monsoon core region

Contact Info

Product

Resources

About