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.

The variability of the (rain)drop size distribution (DSD) in time and space is an intrinsic property of rainfall, which is of primary importance for various environmental fields such as remote sensing of precipitation, for example. DSD observations are usually collected using disdrometers deployed at the ground level. Like any other measurement of a physical process, disdrometer measurements are affected by noise and sampling effects. This uncertainty must be quantified and taken into account in further analyses. This paper addresses this issue for the Particle Size Velocity (PARSIVEL) optical disdrometer by using a large dataset corresponding to light and moderate rainfall and collected from two collocated PARSIVELs deployed during 15 months in Lausanne, Switzerland. The relative sampling uncertainty associated with quantities characterizing the DSD—namely the total concentration of drops Nt and the median-volume diameter D0—is quantified for different temporal resolutions. Similarly, the relative sampling uncertainty associated with the estimates of the most commonly used weighted moments of the DSD (i.e., the rain-rate R, the radar reflectivity at horizontal polarization Zh, and the differential reflectivity Zdr) is quantified as well for different weather radar frequencies. The relative sampling uncertainty associated with estimates of Nt is below 13% for time steps longer than 60 s. For D0, it is below 8% for D0 values smaller than 1 mm. The associated sampling uncertainty for estimates of R is on the order of 15% at a temporal resolution of 60 s. For Zh, the sampling uncertainty is below 9% for Zh values below 35 dBZ at a temporal resolution of 60 s. For Zdr values below 0.75 dB, the sampling uncertainty is below 36% for all temporal resolutions. These analyses provide relevant information for the accurate quantification of the variability of the DSD from disdrometer measurements.

Significance: Precipitation over Antarctica remains largely unknown, despite its crucial role in the surface mass balance of the Antarctic ice sheet. Using unprecedented observations covering an entire year, this work describes a previously unknown mechanism that leads to the sublimation of a large fraction of snowfall in the lower atmosphere, resulting from the interaction of precipitation and katabatic winds. Snowfall sublimation in the atmosphere, caused by katabatic winds, is in the order of 35% in the margins of East Antarctica. This process critically affects the interpretation of satellite-based remote sensing observations close to the ground and suggests that snowfall sublimation in a warming climate may counterbalance the expected increase of precipitation.

The raindrop size distribution (DSD) quantifies the microstructure of rainfall and is critical to studying precipitation processes. We present a method to improve the accuracy of DSD measurements from Parsivel (particle size and velocity) disdrometers, using a two-dimensional video disdrometer (2DVD) as a reference instrument. Parsivel disdrometers bin raindrops into velocity and equivolume diameter classes, but may misestimate the number of drops per class. In our correction method, drop velocities are corrected with reference to theoretical models of terminal drop velocity. We define a filter for raw disdrometer measurements to remove particles that are unlikely to be plausible raindrops. Drop concentrations are corrected such that on average the Parsivel concentrations match those recorded by a 2DVD. The correction can be trained on and applied to data from both generations of OTT Parsivel disdrometers, and indeed any disdrometer in general. The method was applied to data collected during field campaigns in Mediterranean France for a network of first- and second-generation Parsivel disdrometers, and on a first-generation Parsivel in Payerne, Switzerland. We compared the moments of the resulting DSDs to those of a collocated 2DVD, and the resulting DSD-derived rain rates to collocated rain gauges. The correction improved the accuracy of the moments of the Parsivel DSDs, and in the majority of cases the rain rate match with collocated rain gauges was improved. In addition, the correction was shown to be similar for two different climatologies, suggesting its general applicability.

Rapid mass movements (RMM) pose a substantial risk to people and infrastructure. Reliable and cost-efficient measures have to be taken to reduce this risk. One of these measures includes establishing and advancing the state of practice in the application of early warning systems (EWSs). EWSs have been developed during the past decades and are rapidly increasing. In this paper, we focus on the technical part of EWSs, i.e., the prediction and timely recognition of imminent hazards, as well as on monitoring slopes at risk and released mass movements. Recent innovations in assessing spatial precipitation, monitoring and precursors of the triggering and deformation of RMM offer new opportunities for next-generation EWSs. However, technical advancement can only be transferred into more reliable, operational EWSs with an adequate well-instructed dedicated staff. To this end, an intense dialog between scientists, engineers and those in charge of warning, as well as further experience with new comprehensive prototype systems jointly operated by scientists and practitioners, will be essential.