High-resolution Atmospheric Fields Research Articles

A Generative Deep Learning Approach to Stochastic Downscaling of Precipitation Forecasts.

Despite continuous improvements, precipitation forecasts are still not as accurate and reliable as those of other meteorological variables. A major contributing factor to this is that several key processes affecting precipitation distribution and intensity occur below the resolved scale of global weather models. Generative adversarial networks (GANs) have been demonstrated by the computer vision community to be successful at super-resolution problems, that is, learning to add fine-scale structure to coarse images. Leinonen etal. (2020, https://doi.org/10.1109/TGRS.2020.3032790) previously applied a GAN to produce ensembles of reconstructed high-resolution atmospheric fields, given coarsened input data. In this paper, we demonstrate this approach can be extended to the more challenging problem of increasing the accuracy and resolution of comparatively low-resolution input from a weather forecasting model, using high-resolution radar measurements as a "ground truth." The neural network must learn to add resolution and structure whilst accounting for non-negligible forecast error. We show that GANs and VAE-GANs can match the statistical properties of state-of-the-art pointwise post-processing methods whilst creating high-resolution, spatially coherent precipitation maps. Our model compares favorably to the best existing downscaling methods in both pixel-wise and pooled CRPS scores, power spectrum information and rank histograms (used to assess calibration). We test our models and show that they perform in a range of scenarios, including heavy rainfall.

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

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 datasets, 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 datasets. 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 dataset of the precipitation radar data. The source code to our GAN is available at https://github.com/jleinonen/downscaling-rnn-gan.

Oxygen cycling in the northern Benguela Upwelling System: Modelling oxygen sources and sinks

This paper elucidates the oxygen dynamics in the northern Benguela Upwelling System by means of process oriented, numerical modelling. Owing to the complex physical-biological interaction in this system, a coupled hydrodynamic-biogeochemical model is required to grasp the various aspects of the oxygen dynamics. We used high-resolution atmospheric fields derived from observations to force our model, available since 1999. The model results represent a 15years, consistent data set of realistic hydrographic and ecosystem variables, including oxygen distribution patterns. After a concise description of the main aspects of the model, we use the model data to analyse the components contributing to the oxygen dynamics, namely, the ocean circulation, the exchange between ocean and atmosphere as well as the local biogeochemical oxygen cycling in the system. We thoroughly validate the model with available field observations and remote sensing data. The strengths of coastal upwelling, which controls the nutrient supply to the euphotic zone, as well as the poleward undercurrent that carries oxygen and nutrients to the shelf in the northern Benguela Upwelling System are well reproduced in the model. Among the biological oxygen sinks, mineralisation in the sediment, respiration of zooplankton and nitrification in the water column are important. We also found that vertical migration of zooplankton in response to the oxygen conditions provides a regulating feedback, which may prevent a complete deoxygenation of suboxic waters. As long as oxygen or nitrate are available in the bottom waters, the activities of chemolithoautotrophic sulphur bacteria on the sediment surface keep the redoxcline within the sediment and prevent the release of hydrogen sulphide into the water column. By horizontal integration of the simulated ocean-atmosphere oxygen flux, it can be shown that the Kunene upwelling cell between 16°S and 18°S is a boundary between the equatorial ocean, characterise by weak oxygen release to the atmosphere, and the subtropical Benguela Upwelling System governed by an enhanced and seasonal varying flux. Furthermore, a comparison of oxygen fluxes controlled by physical transport versus biogeochemical processes shows that the physical processes dominate in the northern Benguela Upwelling System.

Metrology of ground-based satellite validation: co-location mismatch and smoothing issues of total ozone comparisons

Abstract. Comparisons with ground-based correlative measurements constitute a key component in the validation of satellite data on atmospheric composition. The error budget of these comparisons contains not only the measurement errors but also several terms related to differences in sampling and smoothing of the inhomogeneous and variable atmospheric field. A versatile system for Observing System Simulation Experiments (OSSEs), named OSSSMOSE, is used here to quantify these terms. Based on the application of pragmatic observation operators onto high-resolution atmospheric fields, it allows a simulation of each individual measurement, and consequently, also of the differences to be expected from spatial and temporal field variations between both measurements making up a comparison pair. As a topical case study, the system is used to evaluate the error budget of total ozone column (TOC) comparisons between GOME-type direct fitting (GODFITv3) satellite retrievals from GOME/ERS2, SCIAMACHY/Envisat, and GOME-2/MetOp-A, and ground-based direct-sun and zenith–sky reference measurements such as those from Dobsons, Brewers, and zenith-scattered light (ZSL-)DOAS instruments, respectively. In particular, the focus is placed on the GODFITv3 reprocessed GOME-2A data record vs. the ground-based instruments contributing to the Network for the Detection of Atmospheric Composition Change (NDACC). The simulations are found to reproduce the actual measurements almost to within the measurement uncertainties, confirming that the OSSE approach and its technical implementation are appropriate. This work reveals that many features of the comparison spread and median difference can be understood as due to metrological differences, even when using strict co-location criteria. In particular, sampling difference errors exceed measurement uncertainties regularly at most mid- and high-latitude stations, with values up to 10 % and more in extreme cases. Smoothing difference errors only play a role in the comparisons with ZSL-DOAS instruments at high latitudes, especially in the presence of a polar vortex due to the strong TOC gradient it induces. At tropical latitudes, where TOC variability is lower, both types of errors remain below about 1 % and consequently do not contribute significantly to the comparison error budget. The detailed analysis of the comparison results, including the metrological errors, suggests that the published random measurement uncertainties for GODFITv3 reprocessed satellite data are potentially overestimated, and adjustments are proposed here. This successful application of the OSSSMOSE system to close for the first time the error budget of TOC comparisons, bodes well for potential future applications, which are briefly touched upon.

Role of Resolved and Parameterized Eddies in the Labrador Sea Balance of Heat and Buoyancy

Abstract Deep convection in the Labrador Sea is an important component of the global ocean ventilation. The associated loss of heat to the atmosphere from the interior of the sea is thought to be mostly supplied by mesoscale eddies, generated either remotely or as a result of convection itself—processes that are not resolved by low-resolution ocean climate models. The authors first employ a high-resolution (°) ocean model forced with high-resolution (33 km, 3 h) atmospheric fields to further elaborate on the role of mesoscale eddies in maintaining the balance of heat and buoyancy in the Labrador Sea. In general agreement with previous studies, it is found that eddies remove heat along the coast and supply it to the interior. Some of the eddies that are generated because of the barotropic instability off the west coast of Greenland are recaptured by the boundary current. In the region of deep convection, the convergence of heat and buoyancy by eddies significantly increases with the deepening of the winter mixed layer. In addition, the vertical eddy flux plays an important part in the heat budget of the upper Labrador Sea, accounting for up to half of the heat loss to the atmosphere north of 60°N. A low-resolution (1°) model with parameterized eddies is then applied to show that it does capture, qualitatively, the general structure of eddy buoyancy advection along the Labrador Current. However, the 1° model is deficient in this regard in the most eddy active region off the west coast of Greenland, although some improvements can be made by forcing it with the high-resolution atmospheric fields.

Reconstruction of high resolution atmospheric fields for Northern Europe using analog-upscaling

Abstract. The analog method (AM) has found application to reconstruct gridded climate fields from the information provided by proxy data and climate model simulations. Here, we test the skill of different setups of the AM, in a controlled but realistic situation, by analysing several statistical properties of reconstructed daily high-resolution atmospheric fields for Northern Europe for a 50-yr period. In this application, station observations of sea-level pressure and air temperature are combined with atmospheric fields from a 50-yr high-resolution regional climate simulation. This reconstruction aims at providing homogeneous and physically consistent atmospheric fields with daily resolution suitable to drive high resolution ocean and ecosystem models. Different settings of the AM are evaluated in this study for the period 1958–2007 to estimate the robustness of the reconstruction and its ability to replicate high and low-frequency variability, realistic probability distributions and extremes of different meteorological variables. It is shown that the AM can realistically reconstruct variables with a strong physical link to daily sea-level pressure on both a daily and monthly scale. However, to reconstruct low-frequency decadal and longer temperature variations, additional monthly mean station temperature as predictor is required. Our results suggest that the AM is a suitable upscaling tool to predict daily fields taken from regional climate simulations based on sparse historical station data.

Generation of the Cape Ghir upwelling filament: A numerical study

Filaments are narrow, shallow structures of cool water originating from the coast. They are typical features of the four main eastern boundary upwelling systems (EBUS). In spite of their significant biological and chemical roles, through the offshore exportation of nutrient-rich waters, the physical processes that generate them are still not completely understood. This paper is a process-oriented study of filament generation mechanisms. Our goal is twofold: firstly, to obtain a numerical solution able to correctly represent the characteristics of the filament off Cape Ghir (30°38′N, northwest Africa) in the Canary EBUS and secondly, to explain its formation by a simple mechanism based on the balance of potential vorticity. The first goal is achieved by the use of the ROMS model (Regional Ocean Modeling System) with embedded domains around Cape Ghir, with a horizontal resolution going up to 1.5 km for the finest domain. The latter gets its initial and boundary conditions from a parent solution and is forced by climatological, high-resolution atmospheric fields. The modeled filaments display spatial, temporal and physical characteristics in agreement with the available in situ and satellite observations. This model solution is used as a reference to compare the results with a set of process-oriented experiments. These experiments allow us to reach the second objective. The solutions serve to highlight the contributions of various processes on the filament generation. Since the study is focused on general processes present under climatological forcing conditions, inter-annual forcing is not necessary. The underlying idea for the filament generation is the balance of potential vorticity in the Canary EBUS: the upwelling jet is characterized by negative relative vorticity and flows southward along a narrow band of uniform potential vorticity. In the vicinity of the cape, an injection of relative vorticity induced by the wind breaks the existing vorticity balance. The upwelling jet is prevented from continuing its way southward and has to turn offshore to follow lines of equal potential vorticity. The model results highlight the essential role of wind, associated with the particular topography (coastline and bottom) around the cape. The mechanism presented here is general and thus can be applied to other EBUS.

Characterization of uncertainty in outdoor sound propagation predictions

Predictive skill for outdoor sound propagation is assessed using high-resolution atmospheric fields from large-eddy simulations (LES). Propagation calculations through the full LES fields are compared to calculations through subsets of the LES fields that have been processed in typical ways, such as mean vertical profiles and instantaneous vertical profiles synchronized to the sound propagation. It is found that mean sound pressure levels can be predicted with low errors from the mean profiles, except in refractive shadow regions. Prediction of sound pressure levels for short-duration events is much less accurate, with errors of 8 -10 dB for near-ground propagation being typical.