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.

Recently published studies of triple‐frequency radar observations of snowfall have demonstrated that naturally occurring snowflakes exhibit scattering signatures that are in some cases consistent with spheroidal particle models and in others can only be explained by complex aggregates. Until recently, no in situ observations have been available to investigate links between microphysical snowfall properties and their scattering properties. In this study, we investigate for the first time relations between collocated ground‐based triple‐frequency observations with in situ measurements of snowfall at the ground. The three analyzed snowfall cases obtained during a recent field campaign in Finland cover light to moderate snowfall rates with transitions from heavily rimed snow to open‐structured, low‐density snowflakes. The observed triple‐frequency signatures agree well with the previously published findings from airborne radar observations. A rich spatiotemporal structure of triple‐frequency observations throughout the cloud is observed during the three cases, which often seems to be related to riming and aggregation zones within the cloud. The comparison of triple‐frequency signatures from the lowest altitudes with the ground‐based in situ measurements reveals that in the presence of large (>5 mm) snow aggregates, a bending away in the triple‐frequency space from the curve of classical spheroid scattering models is always observed. Rimed particles appear along an almost horizontal line in the triple‐frequency space, which was not observed before. Overall, the three case studies indicate a close connection of triple‐frequency signatures and snow particle structure, bulk snowfall density, and characteristic size of the particle size distribution.

The PyTMatrix package was designed with the objective of providing a simple, extensible interface to T-Matrix electromagnetic scattering calculations performed using an extensively validated numerical core. The interface, implemented in the Python programming language, facilitates automation of the calculations and further analysis of the results through direct integration of both the inputs and the outputs of the calculations to numerical analysis software. This article describes the architecture and design of the package, illustrating how the concepts in the physics of electromagnetic scattering are mapped into data and code models in the computer software. The resulting capabilities and their consequences for the usability and performance of the package are explored.

The capability to detect the state of snowflake riming reliably from remote measurements would greatly expand the understanding of its global role in cloud‐precipitation processes. To investigate the ability of multifrequency radars to detect riming, a three‐dimensional model of snowflake growth was used to generate simulated aggregate and crystal snowflakes with various degrees of riming. Three different growth scenarios, representing different temporal relationships between aggregation and riming, were formulated. The discrete dipole approximation was then used to compute the radar backscattering properties of the snowflakes at frequencies of 9.7, 13.6, 35.6, and 94 GHz. In two of the three growth scenarios, the rimed snowflakes exhibit large differences between the backscattering cross sections of the detailed three‐dimensional models and the equivalent homogeneous spheroidal models, similarly to earlier results for unrimed snowflakes. When three frequencies are used simultaneously, riming appears to be detectable in a robust manner across all three scenarios. In spite of the differences in backscattering cross sections, the triple‐frequency signatures of heavily rimed particles resemble those of the homogeneous spheroids, thus explaining earlier observational results that were compatible with such spheroids.

The link between stratiform precipitation microphysics and multifrequency radar observables is thoroughly investigated by exploiting simultaneous airborne radar and in situ observations collected from two aircraft during the OLYMPEX/RADEX (Olympic Mountain Experiment/Radar Definition Experiment 2015) field campaign. Above the melting level, in situ images and triple‐frequency radar signatures both indicate the presence of moderately rimed aggregates. Various mass‐size relationships of ice particles and snow scattering databases are used to compute the radar reflectivity from the in situ particle size distribution. At Ku and Ka band, the best agreement with radar observations is found when using the self‐similar Rayleigh‐Gans approximation for moderately rimed aggregates. At W band, a direct comparison is challenging because of the non‐Rayleigh effects and of the probable attenuation due to ice aggregates and supercooled liquid water between the two aircraft. A variational method enables the retrieval of the full precipitation profile above and below the melting layer, by combining the observations from the three radars. Even with three radar frequencies, the retrieval of rain properties is challenging over land, where the integrated attenuation is not available. Otherwise, retrieved mean volume diameters and water contents of both solid and liquid precipitation are in agreement with in situ observations and indicate local changes of the degree of riming of ice aggregates, on the scale of 5 km. Finally, retrieval results are analyzed to explore the validity of using continuity constraints on the water mass flux and diameter within the melting layer in order to improve retrievals of ice properties.

The sensitivity of radar backscattering cross sections on different snowflake shapes is studied at C, Ku, Ka, and W bands. Snowflakes are simulated using two complex shape models, namely, fractal and aggregate, and a soft spheroid model. The models are tuned to emulate physical properties of real snowflakes, that is, the mass–size relation and aspect ratio. It is found that for particle sizes up to 5 mm and for frequencies from 5 to 35 GHz, there is a good agreement in the backscattering cross section for all models. For larger snowflakes at the Ka band, it is found that the spheroid model underestimates the backscattering cross sections by a factor of 10, and at W band by a factor of 50–100. Furthermore, there is a noticeable difference between spheroid and complex shape models in the linear depolarization ratios for all frequencies and particle sizes.