English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Zendy Pro

Presents the zaia usage as premium feature

ZAIA

Presents AI Discovery search usage as a premium feature

AI Discovery

Presents AI Discovery explain and summary usage as a premium feature

Content Analysis

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Presents the pdf analysis as premium feature

PDF Analysis

1500 Top-up

3000 Top-up

Zendy Basic

2500 Top-up

Zendy Unlimited

Presents the access of premium content as premium feature

Premium Content

Generative adversarial networks (GANs) have been recently adopted forsuper-resolution, an application closely related to what is referred to as"downscaling" in the atmospheric sciences: improving the spatial resolution oflow-resolution images. The ability of conditional GANs to generate an ensembleof solutions for a given input lends itself naturally to stochasticdownscaling, but the stochastic nature of GANs is not usually considered insuper-resolution applications. Here, we introduce a recurrent, stochasticsuper-resolution GAN that can generate ensembles of time-evolvinghigh-resolution atmospheric fields for an input consisting of a low-resolutionsequence of images of the same field. We test the GAN using two datasets, oneconsisting of radar-measured precipitation from Switzerland, the other of cloudoptical thickness derived from the Geostationary Earth Observing Satellite 16(GOES-16). We find that the GAN can generate realistic, temporally consistentsuper-resolution sequences for both datasets. The statistical properties of thegenerated ensemble are analyzed using rank statistics, a method adapted fromensemble weather forecasting; these analyses indicate that the GAN producesclose to the correct amount of variability in its outputs. As the GAN generatoris fully convolutional, it can be applied after training to input images largerthan the images used to train it. It is also able to generate time series muchlonger than the training sequences, as demonstrated by applying the generatorto a three-month dataset of the precipitation radar data. The source code toour GAN is available at https://github.com/jleinonen/downscaling-rnn-gan.

Stochastic Super-Resolution for Downscaling Time-Evolving Atmospheric Fields With a Generative Adversarial Network