Deep Learning for Downscaling Tropical Cyclone Rainfall to Hazard‐Relevant Spatial Scales

Vosper, Emily; Watson, P.A.; Harris, Lucy; McRae, Andrew T. T.; Santos-Rodríguez, Raúl; Aitchison, Laurence; Mitchell, Dann

doi:10.1029/2022jd038163

Cited by 9 publications

(15 citation statements)

References 37 publications

(95 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It is widely used in many super-resolution and downscaling studies (i.e. Harris et al, 2022;Leinonen et al, 2021;Vosper et al, 2023). Note we use the MSE loss function as opposed to the MAE loss as it is more sensitive to errors in extreme events (not shown).…”

Section: Generator Lossmentioning

confidence: 99%

“…Note we use the MSE loss function as opposed to the MAE loss as it is more sensitive to errors in extreme events (not shown). It is important to note that training with an 𝜆 𝑎𝑑𝑣 too large is often unstable (Isola et al, 2018;Vosper et al, 2023), and the majority of existing studies generally use values of 𝜆 𝑎𝑑𝑣 less than 0.005 (Harris et al, 2022;Izumi et al, 2022;Leinonen et al, 2021;Vosper et al, 2023;X. Wang et al, 2018).…”

Section: Generator Lossmentioning

confidence: 99%

“…While regression-based approaches (including deep learning) are skillful in capturing the "mean-state" in instantaneous predictions (i.e. they regress to the mean), they tend to underestimate extreme events and struggle to resolve fine scale details (Harris et al, 2022;Mardani et al, 2023;Rampal, 2024;Reddy et al, 2023;Vosper et al, 2023;J. Wang et al, 2021).…”

Section: Introductionmentioning

confidence: 99%

“…Wang et al, 2018), and have recently been adapted from the computer vision sub-field of super-resolution (which focuses on enhancing image resolution) to climate downscaling. GANs, also often described as conditional GANs (c-GANs) in this context, have significantly improved regression-based computer-vision algorithms in predicting local-scale extreme events and resolving highresolution spatial structure in the downscaled predictions (Annau et al, 2023;Brochet et al, 2023;Izumi et al, 2022;Leinonen et al, 2021;Miralles et al, 2022;Oyama et al, 2023;Price & Rasp, 2022;Ravuri et al, 2021a;Saha & Ravela, 2022;Vosper et al, 2023;J. Wang et al, 2021).…”

Section: Introductionmentioning

confidence: 99%

“…Existing studies in downscaling applications have only conducted their research using a specific value of 𝜆 adv (i.e. Annau et al, 2023;Harris et al, 2022;Leinonen et al, 2021;Vosper et al, 2023), with limited exploration of how the strength of the adversarial loss affects climate downscaling.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

A Robust Generative Adversarial Network Approach for Climate Downscaling and Weather Generation

Rampal,

Gibson,

Sherwood

et al. 2024

Preprint

View full text Add to dashboard Cite

Anticipating climate impacts and risks in present or future climates requires predicting the statistics of high-impact weather events at fine-scales. Direct numerical simulations of fine-scale weather are computationally too expensive for many uses. While regression-based (deep-learning or statistical) downscaling of low-resolution climate simulations is several orders of magnitude faster than direct numerical simulations, it suffers from several limitations. These limitations include the tendency to regress to the mean, which produces excessively smooth predictions and underestimates the magnitude of extreme events. Additionally, they also fail to preserve statistical measures that are key for climate research. We use a conditional GAN (c-GAN) architecture to downscale daily precipitation as a Regional Climate Model (RCM) emulator. The c-GAN generates plausible residuals on top of the predictable expectation state produced by a regression-based DL algorithm. The skill of c-GANs is highly sensitive to a hyperparameter known as the weight of the adversarial loss (λadv), and the value of λadv required for accurate results varies with season and performance metric, casting doubt on the robustness of c-GANs as usually implemented. But, by applying a simple intensity constraint to the loss function, it is possible to obtain robust performance results across λadv spanning two orders of magnitude. C-GANs are considerably more skillful in capturing climatological statistics including the distribution and spatial characteristics of extreme events. We expect c-GANs with this modification to be readily transferable to other problems and time periods, making them a useful weather generator for representing extreme event statistics in present and future climates.

show abstract

Section: Generator Lossmentioning

confidence: 99%

Section: Generator Lossmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

A Robust Generative Adversarial Network Approach for Climate Downscaling and Weather Generation

Rampal,

Gibson,

Sherwood

et al. 2024

Preprint

View full text Add to dashboard Cite

show abstract

Artificial intelligence for climate prediction of extremes: State of the art, challenges, and future perspectives

Materia,

García,

van Straaten

et al. 2024

WIREs Climate Change

View full text Add to dashboard Cite

Extreme events such as heat waves and cold spells, droughts, heavy rain, and storms are particularly challenging to predict accurately due to their rarity and chaotic nature, and because of model limitations. However, recent studies have shown that there might be systemic predictability that is not being leveraged, whose exploitation could meet the need for reliable predictions of aggregated extreme weather measures on timescales from weeks to decades ahead. Recently, numerous studies have been devoted to the use of artificial intelligence (AI) to study predictability and make climate predictions. AI techniques have shown great potential to improve the prediction of extreme events and uncover their links to large‐scale and local drivers. Machine and deep learning have been explored to enhance prediction, while causal discovery and explainable AI have been tested to improve our understanding of the processes underlying predictability. Hybrid predictions combining AI, which can reveal unknown spatiotemporal connections from data, with climate models that provide the theoretical foundation and interpretability of the physical world, have shown that improving prediction skills of extremes on climate‐relevant timescales is possible. However, numerous challenges persist in various aspects, including data curation, model uncertainty, generalizability, reproducibility of methods, and workflows. This review aims at overviewing achievements and challenges in the use of AI techniques to improve the prediction of extremes at the subseasonal to decadal timescale. A few best practices are identified to increase trust in these novel techniques, and future perspectives are envisaged for further scientific development.This article is categorized under: Climate Models and Modeling > Knowledge Generation with Models The Social Status of Climate Change Knowledge > Climate Science and Decision Making

show abstract

On the suitability of a convolutional neural network based RCM-emulator for fine spatio-temporal precipitation

Doury,

Somot,

Gadat

2024

Clim Dyn

View full text Add to dashboard Cite

High resolution regional climate models (RCM) are necessary to capture local precipitation but are too expensive to fully explore the uncertainties associated with future projections. To resolve the large cost of RCMs, Doury et al. (2023) proposed a neural network based RCM-emulator for the near-surface temperature, at a daily and 12 km-resolution. It uses existing RCM simulations to learn the relationship between low-resolution predictors and high resolution surface variables. When trained the emulator can be applied to any low resolution simulation to produce ensembles of high resolution emulated simulations. This study assesses the suitability of applying the RCM-emulator for precipitation thanks to a novel asymmetric loss function to reproduce the entire precipitation distribution over any grid point. Under a perfect conditions framework, the resulting emulator shows striking ability to reproduce the RCM original series with an excellent spatio-temporal correlation. In particular, a very good behaviour is obtained for the two tails of the distribution, measured by the number of dry days and the 99th quantile. Moreover, it creates consistent precipitation objects even if the highest frequency details are missed. The emulator quality holds for all simulations of the same RCM, with any driving GCM, ensuring transferability of the tool to GCMs never downscaled by the RCM. A first showcase of downscaling GCM simulations showed that the RCM-emulator brings significant added-value with respect to the GCM as it produces the correct high resolution spatial structure and heavy precipitation intensity. Nevertheless, further work is needed to establish a relevant evaluation framework for GCM applications.

show abstract

Deep Learning for Downscaling Tropical Cyclone Rainfall to Hazard‐Relevant Spatial Scales

Cited by 9 publications

References 37 publications

A Robust Generative Adversarial Network Approach for Climate Downscaling and Weather Generation

A Robust Generative Adversarial Network Approach for Climate Downscaling and Weather Generation

Artificial intelligence for climate prediction of extremes: State of the art, challenges, and future perspectives

On the suitability of a convolutional neural network based RCM-emulator for fine spatio-temporal precipitation

Contact Info

Product

Resources

About