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

Abstract

Generative adversarial networks (GANs) have been recently adopted for super-resolution, an application closely related to what is referred to as "downscaling" in the atmospheric sciences: improving the spatial resolution of low-resolution images. The ability of conditional GANs to generate an ensemble of solutions for a given input lends itself naturally to stochastic downscaling, but the stochastic nature of GANs is not usually considered in super-resolution applications. Here, we introduce a recurrent, stochastic super-resolution GAN that can generate ensembles of time-evolving high-resolution atmospheric fields for an input consisting of a low-resolution sequence of images of the same field. We test the GAN using two data sets: one consisting of radar-measured precipitation from Switzerland; the other of cloud optical thickness derived from the Geostationary Earth Observing Satellite 16 (GOES-16). We find that the GAN can generate realistic, temporally consistent super-resolution sequences for both data sets. The statistical properties of the generated ensemble are analyzed using rank statistics, a method adapted from ensemble weather forecasting; these analyses indicate that the GAN produces close to the correct amount of variability in its outputs. As the GAN generator is fully convolutional, it can be applied after training to input images larger than the images used to train it. It is also able to generate time series much longer than the training sequences, as demonstrated by applying the generator to a three-month data set of the precipitation radar data. The source code to our GAN is available at https://github.com/jleinonen/downscaling-rnn-gan.

Reconstruction of the mass and geometry of snowfall particles from multi angle snowflake camera (MASC) images

“…Its measurement principle is detailed in Garrett et al (2012). Several works exploited MASC data to investigate geometry and fall speed characteristics of hydrometeors (Garrett and Yuter, 2014;Garrett et al, 2015;Jiang et al, 2019), and others were devoted to hydrometeor classification techniques as Praz et al (2017); Hicks and Notaros (2019); Leinonen and Berne (2020).…”

“…More recently, accurate and high-resolution depictions of snowflakes could be obtained with imagers like the Snow Video Imager/Particle Image Probe (Newman et al, 2009) or with the Multi-Angle Snowflake Camera (MASC Garrett et al, 2012). The availability of actual images has promoted the development and rapid improvement of several automatic hydrometeor classification techniques (Grazioli et al, 2014;Gavrilov et al, 2015;Praz et al, 2017;Leinonen and Berne, 2020) adapted to the data of these sensors. While the accuracy of the measurements of fall velocity provided by those instruments is often hampered by wind and turbulence (Nešpor et al, 2000;Garrett and Yuter, 2014;Fitch et al, 2021), the added value in terms of microphysical characterization is significant.…”

“…GANs are nowadays finding application in the field of environmental and atmospheric sciences (e.g. Leinonen and Berne, 2020;Leinonen et al, 2020), thanks to their versatility and their ability to perform 3D reconstruction of images has been already explored, for example, in the medical field (Yang et al, 2017). The GAN presented here is trained on a set of simulated snowflakes (generated using the technique of Leinonen et al, 2013;Leinonen and Moisseev, 2015;Leinonen and Szyrmer, 2015;Karrer et al, 2020) and evaluated on 3D-printed 1 : 1 scale snowflake replicas repeatedly dropped into the MASC sampling area.…”

Modeling Noise: Paths toward AI-Enabled Stochastic Earth System Models and Parameterizations

“…A third approach is a Linear Inverse Modeling framework (Newman et al, 2003;Martinez-Villalobos et al, 2017), where the predictive modes are represented as covariance functions in a reduced space (e.g., functions of PCAs). We can also model the system with reduced complexity and represent higher complexity processes as AI-driven stochastic processes (Chattopadhyay et al, 2020;Crommelin and Edeling, 2020;Alcala and Timofeyev, 2020;Leinonen et al, 2020). To characterize noise relevant for predicting high-frequency signals, convection-resolving simulations such as the DYAMOND ensemble (Stevens et al, 2019) provide comprehensive data coverage to characterize variability in small-scale processes (Christensen, 2020).…”

Fast and accurate learned multiresolution dynamical downscaling for precipitation

“…Other studies have used a recurrent neural network structure (Leinonen et al, 2020) to permit generated outputs to evolve in time in a consistent manner, so that the GAN generator can model the time evolution of fields and the discriminator can evaluate the plausibility of image sequences rather than single images. The 3-hourly data that we use in this study are potentially too coarse to consider time dependencies: short-duration events may disappear between timesteps, and even for long-duration events, 3 hours may be too long to capture a smooth transition from one timestep to the next, as preferred by the learning process.…”