Browse

You are looking at 1 - 10 of 121,525 items for

  • Refine by Access: All Content x
Clear All
Ryan Lagerquist
and
Imme Ebert-Uphoff

Abstract

In the last decade, much work in atmospheric science has focused on spatial verification (SV) methods for gridded prediction, which overcome serious disadvantages of pixelwise verification. However, neural networks (NN) in atmospheric science are almost always trained to optimize pixelwise loss functions, even when ultimately assessed with SV methods. This establishes a disconnect between model verification during versus after training. To address this issue, we develop spatially enhanced loss functions (SELF) and demonstrate their use for a real-world problem: predicting the occurrence of thunderstorms (henceforth, “convection”) with NNs. In each SELF we use either a neighborhood filter, which highlights convection at scales larger than a threshold, or a spectral filter (employing Fourier or wavelet decomposition), which is more flexible and highlights convection at scales between two thresholds. We use these filters to spatially enhance common verification scores, such as the Brier score. We train each NN with a different SELF and compare their performance at many scales of convection, from discrete storm cells to tropical cyclones. Among our many findings are that (i) for a low or high risk threshold, the ideal SELF focuses on small or large scales, respectively; (ii) models trained with a pixelwise loss function perform surprisingly well; and (iii) nevertheless, models trained with a spectral filter produce much better-calibrated probabilities than a pixelwise model. We provide a general guide to using SELFs, including technical challenges and the final Python code, as well as demonstrating their use for the convection problem. To our knowledge this is the most in-depth guide to SELFs in the geosciences.

Significance Statement

Gridded predictions, in which a quantity is predicted at every pixel in space, should be verified with spatially aware methods rather than pixel by pixel. Neural networks (NN), which are often used for gridded prediction, are trained to minimize an error value called the loss function. NN loss functions in atmospheric science are almost always pixelwise, which causes the predictions to miss rare events and contain unrealistic spatial patterns. We use spatial filters to enhance NN loss functions, and we test our novel spatially enhanced loss functions (SELF) on thunderstorm prediction. We find that different SELFs work better for different scales (i.e., different-sized thunderstorm complexes) and that spectral filters, one of the two filter types, produce unexpectedly well calibrated thunderstorm probabilities.

Free access
Moritz Günther
,
Hauke Schmidt
,
Claudia Timmreck
, and
Matthew Toohey

Abstract

Volcanic aerosol forcing has previously been found to cause a weak global mean temperature response, as compared with CO2 radiative forcing of equal magnitude: its efficacy is supposedly low, but for reasons that are not fully understood. To investigate this, we perform idealized, time-invariant stratospheric sulfate aerosol forcing simulations with the MPI-ESM-1.2 and compare them with 0.5 × CO2 and 2 × CO2 runs. While the early decades of the aerosol forcing simulations are characterized by strong negative feedback (i.e., low efficacy), the feedback weakens on the decadal to centennial time scale. Although this effect is qualitatively also found in CO2-warming simulations, it is more pronounced for stratospheric aerosol forcing. The strong early and weak late cooling feedbacks compensate, leading to an equilibrium efficacy of approximately 1 in all simulations. The 0.5 × CO2 cooling simulations also exhibit strong feedback changes over time, albeit less than in the idealized aerosol forcing simulations. This suggests that the underlying cause for the feedback change is not exclusively specific to aerosol forcing. One critical region for the feedback differences between simulations with negative and positive radiative forcing is the tropical Indo-Pacific warm-pool region (30°S–30°N, 50°E–160°W). In the first decades of cooling, the temperature change in this region is stronger than the global average, whereas it is stronger outside it for 2 × CO2 warming. In cooling scenarios, this leads to an enhanced activation of the warm-pool region’s strongly negative lapse-rate feedback.

Significance Statement

Large volcanic eruptions can enhance the scattering aerosol layer in the stratosphere, which leads to a global cooling for a few years. Surprisingly, Earth has been found to cool less from radiative flux perturbations from stratospheric aerosol forcing, in comparison with how much it warms as a result of increases in CO2 concentration. We find that specific surface temperature change patterns after volcanic eruptions cause this effect. The temperature change in the tropical Indian and western Pacific Ocean determines how much global temperature change is needed to regain radiative equilibrium. Our findings contribute to understanding the climate response to volcanic eruptions and are relevant for understanding the mechanisms of climate change due to changes in CO2 concentration.

Restricted access
Xiaofeng Li
,
Dongliang Shen
,
Gang Zheng
,
Lizhang Zhou
, and
Aiyue Liu

Abstract

A winter storm triggered a significant sea smoke with the northwesterly wind to the Yellow Sea, China, on 7 January 2021. The ocean responses to this event lasted about 3 days. Satellite observations show that the sea surface temperature dropped from 5.7° to 4.7°C on the following day and then recovered to the previous level; the chlorophyll-a, a bio-growth indicator, increased from 3.6 to 3.9 mg m−3 due to cooling-induced coastal upwelling between 7 and 9 January. Two buoys measurements showed that the air temperature dropped to −13.3°C and high relative humidity with a maximum value of 89.0% above the sea surface, creating favorable conditions for sea smoke generation. A Regional Ocean Modeling System (ROMS) and Weather Research and Forecasting (WRF) Model coupled model with the Goddard Chemistry Aerosol Radiation and Transport (GOCART) chemical module was implemented to reproduce this sea smoke phenomenon and analyze the air–sea interaction. The 20°C temperature difference between extreme cold air (−13.3°C) and the relatively warm stable sea surface (4.7°–5.7°C) enhanced the seawater evaporation. In addition, we suppose the concentration of sea salt, a kind of condensation nucleus, with a particle diameter of 0.5–1.5 μm above the sea surface increased quickly on 7 January. The boiling-water-like sea surface was imaged on a synthetic aperture image. We developed an image analysis method to describe the cell-shaped texture characteristics imaged by Synthetic Aperture Radar (SAR). We also found that the sea surface imprints of sea smoke are governed by the thermal, not the dynamical instability.

Significance Statement

On 7 January 2021, a significant sea smoke event happened in the Yellow Sea. The ocean response to the event lasted 3 days. First, on a synoptic scale, this study presents the comprehensive satellite observations of the sea surface temperature drop and chlorophyll-a increase associated with the sea smoke. Second, a coupled air–sea interaction model with the Goddard Chemistry Aerosol Radiation and Transport (GOCART) chemical module was implemented to reproduce this sea smoke phenomenon and identify which condensation nucleus induced such heavy sea smoke. Third, we developed an image analysis method to analyze high-resolution synthetic aperture radar images and found that the sea surface imprints of sea smoke are governed by the thermal, not the dynamical instability.

Restricted access
Tianjiao Ma
,
Wen Chen
,
Shangfeng Chen
,
Chaim I. Garfinkel
,
Shuoyi Ding
,
Lei Song
,
Zhibo Li
,
Yulian Tang
,
Jingliang Huangfu
,
Hainan Gong
, and
Wei Zhao

Abstract

This study aims to better understand the ENSO impacts on climate anomalies over East Asia in early winter (November–December) and late winter (January–February). In particular, the possible mechanisms during early winter are investigated. The results show that ENSO is associated with a Rossby wave train emanating from the tropical Indian Ocean toward East Asia (denoted as tIO-EA) in early winter. This tIO-EA wave train in El Niño (La Niña) is closely related to a weakening (strengthening) of the East Asian trough, and thereby a weakened (strengthened) East Asian winter monsoon and warm (cold) temperature anomalies over northeastern China and Japan. By using partial regression analysis and numerical experiments, we identify that the formation of tIO-EA wave train is closely related to precipitation anomalies in the tropical eastern Indian Ocean and western Pacific (denoted as eIO/wP). In addition, the ENSO-induced North Atlantic anomalies may also contribute to formation of the tIO-EA wave train in conjunction with the eIO/wP precipitation. The response of eIO/wP precipitation to ENSO is stronger in early winter than in late winter. This can be attributed to the stronger anomalous Walker circulation over the Indian Ocean, which in turn is caused by higher climatological SST and stronger mean precipitation state in the Indian Ocean during early winter.

Restricted access
Ophélia Miralles
,
Daniel Steinfield
,
Olivia Martius
, and
Anthony C. Davison

Abstract

Near-surface wind is difficult to estimate using global numerical weather and climate models, because airflow is strongly modified by underlying topography, especially that of a country such as Switzerland. In this article, we use a statistical approach based on deep learning and a high-resolution digital elevation model to spatially downscale hourly near-surface wind fields at coarse resolution from ERA5 reanalysis from their original 25-km grid to a 1.1-km grid. A 1.1-km-resolution wind dataset for 2016–20 from the operational numerical weather prediction model COSMO-1 of the national weather service MeteoSwiss is used to train and validate our model, a generative adversarial network (GAN) with gradient penalized Wasserstein loss aided by transfer learning. The results are realistic-looking high-resolution historical maps of gridded hourly wind fields over Switzerland and very good and robust predictions of the aggregated wind speed distribution. Regionally averaged image-specific metrics show a clear improvement in prediction relative to ERA5, with skill measures generally better for locations over the flatter Swiss Plateau than for Alpine regions. The downscaled wind fields demonstrate higher-resolution, physically plausible orographic effects, such as ridge acceleration and sheltering, that are not resolved in the original ERA5 fields.

Significance Statement

Statistical downscaling, which increases the resolution of atmospheric fields, is widely used to refine the outputs of global reanalysis and climate models, most commonly for temperature and precipitation. Near-surface winds are strongly modified by the underlying topography, generating local flow conditions that can be very difficult to estimate. This study develops a deep learning model that uses local topographic information to spatially downscale hourly near-surface winds from their original 25-km resolution to a 1.1-km grid over Switzerland. Our model produces realistic high-resolution gridded wind fields with expected orographic effects but performs better in flatter regions than in mountains. These downscaled fields are useful for impact assessment and decision-making in regions where global reanalysis data at coarse resolution may be the only products available.

Free access
Boyin Huang
,
Xungang Yin
,
Matthew J. Menne
,
Russell Vose
, and
Huai-Min Zhang

Abstract

NOAA global surface temperature (NOAAGlobalTemp) is NOAA’s operational global surface temperature product, which has been widely used in Earth’s climate assessment and monitoring. To improve the spatial interpolation of monthly land surface air temperatures (LSATs) in NOAAGlobalTemp from 1850 to 2020, a three-layer artificial neural network (ANN) system was designed. The ANN system was trained by repeatedly randomly selecting 90% of the LSATs from ERA5 (1950–2019) and validating with the remaining 10%. Validations show clear improvements of ANN over the original empirical orthogonal teleconnection (EOT) method: the global spatial correlation coefficient (SCC) increases from 65% to 80%, and the global root-mean-square difference (RMSD) decreases from 0.99° to 0.57°C during 1850–2020. The improvements of SCCs and RMSDs are larger in the Southern Hemisphere than in the Northern Hemisphere and are larger before the 1950s and where observations are sparse. The ANN system was finally fed in observed LSATs, and its output over the global land surface was compared with those from the EOT method. Comparisons demonstrate similar improvements by ANN over the EOT method: The global SCC increased from 78% to 89%, the global RMSD decreased from 0.93° to 0.68°C, and the LSAT variability quantified by the monthly standard deviation (STD) increases from 1.16° to 1.41°C during 1850–2020. While the SCC, RMSD, and STD at the monthly time scale have been improved, long-term trends remain largely unchanged because the low-frequency component of LSAT in ANN is identical to that in the EOT approach.

Significance Statement

The spatial interpolation method of an artificial neural network has greatly improved the accuracy of land surface air temperature reconstruction, which reduces root-mean-square error and increases spatial coherence and variabilities over the global land surface from 1850 to 2020.

Free access
Jussi Leinonen
,
Ulrich Hamann
, and
Urs Germann

Abstract

A deep learning model is presented to nowcast the occurrence of lightning at a 5-min time resolution 60 min into the future. The model is based on a recurrent-convolutional architecture that allows it to recognize and predict the spatiotemporal development of convection, including the motion, growth and decay of thunderstorm cells. The predictions are performed on a stationary grid, without the use of storm object detection and tracking. The input data, collected from an area in and surrounding Switzerland, comprise ground-based radar data, visible/infrared satellite data and derived cloud products, lightning detection, numerical weather prediction, and digital elevation model data. We analyze different alternative loss functions, class weighting strategies and model features, providing guidelines for future studies to select loss functions optimally and to properly calibrate the probabilistic predictions of their model. On the basis of these analyses, we use focal loss in this study but conclude that it only provides a small benefit over cross entropy, which is a viable option if recalibration of the model is not practical. The model achieves a pixelwise critical success index (CSI) of 0.45 to predict lightning occurrence within 8 km over the 60-min nowcast period, ranging from a CSI of 0.75 at a 5-min lead time to a CSI of 0.32 at a 60-min lead time.

Significance Statement

We have developed a method based on artificial intelligence to forecast the occurrence of lightning at 5-min intervals within the next hour from the forecast time. The method utilizes a neural network that learns to predict lightning from a set of training images containing lightning detection data, weather radar observations, satellite imagery, weather forecasts, and elevation data. We find that the network is able to predict the motion, growth, and decay of lightning-producing thunderstorms and that, when properly tuned, it can accurately determine the probability of lightning occurring. This is expected to permit more informed decisions to be made about short-term lightning risks in fields such as civil protection, electricity-grid management, and aviation.

Free access
Prasanth Prabhakaran
,
Subin Thomas
,
Will Cantrell
,
Raymond A. Shaw
, and
Fan Yang

Abstract

The role played by fluctuations of supersaturation in the growth of cloud droplets is examined in this study. The stochastic condensation framework and the three regimes of activation of cloud droplets— namely, mean dominant, fluctuation influenced, and fluctuation dominant—are used for analyzing the data from high-resolution large-eddy simulations of the Pi convection-cloud chamber. Based on a detailed budget analysis the significance of all the terms in the evolution of the droplet size distribution equation is evaluated in all three regimes. The analysis indicates that the mean-growth rate is a dominant process in shaping the droplet size distribution in all three regimes. Turbulence introduces two sources of stochasticity, turbulent transport and particle lifetime, and supersaturation fluctuations. The transport of cloud droplets plays an important role in all three regimes, whereas the direct effect of supersaturation fluctuations is primarily related to the activation and growth of the small droplets in the fluctuation-influenced and fluctuation-dominant regimes. We compare our results against the previous studies (experimental and theory) of the Pi chamber, and discuss the limitations of the existing models based on the stochastic condensation framework. Furthermore, we extend the discussion of our results to atmospheric clouds, and in particular focus on recent adiabatic turbulent cloud parcel simulations based on the stochastic condensation framework, and emphasize the importance of entrainment/mixing and turbulent transport in shaping the droplet size distribution.

Restricted access
Jiawei Bao
,
Vishal Dixit
, and
Steven C. Sherwood

Abstract

The horizontal temperature gradients in the tropical free troposphere are generally assumed to be weak. We show with ERA5 data that substantial zonal virtual temperature (Tυ ) gradients persist climatologically in the tropical free troposphere and investigate their causes. The gradients change seasonally: Tυ at 500 hPa over the equatorial western Pacific Ocean (EWP) is usually much warmer (up to 3 K) than that over the equatorial central Pacific Ocean (ECP) during December–February (DJF), while the temperature differences between EWP and ECP are much smaller during June–August (JJA). During DJF, Tυ gradients over the Pacific prevail throughout the entire free troposphere, especially in the upper troposphere near 300 hPa. We find that the associated hydrostatic pressure gradients are mainly balanced by the nonlinear terms in the momentum equation, in particular via zonal wind advection. Strong zonal winds occur near the equator in boreal winter, transporting zonal momentum so as to balance the pressure gradient force. The zonal winds are due to large-scale equatorial waves, excited by a heating pattern that is relatively symmetric about the equator. In boreal summer, the large-scale equatorial waves are less active in the Pacific region due to a more asymmetric temperature pattern, so the zonal momentum advection and Tυ gradients are both much weaker. The results point to an important role of the nonlinear terms in the tropical balanced dynamics, stressing the need for an improved theoretical understanding and modeling framework of the tropical atmosphere that includes these nonlinear terms, or their net effect.

Restricted access
Joaquín E. Blanco
,
Rodrigo Caballero
,
George Datseris
,
Bjorn Stevens
,
Sandrine Bony
,
Or Hadas
, and
Yohai Kaspi

Abstract

The Northern and Southern Hemispheres reflect on average almost equal amounts of sunlight due to compensating hemispheric asymmetries in clear-sky and cloud albedo. Recent work indicates that the cloud albedo asymmetry is largely due to clouds in extratropical oceanic regions. Here, we investigate the proximate causes of this extratropical cloud albedo asymmetry using a cloud-controlling factor (CCF) approach. We develop a simple index that measures the skill of CCFs, either individually or in combination, in predicting the asymmetry. The index captures the contribution to the asymmetry due to interhemispheric differences in the probability distribution function of daily CCF values. Cloud albedo is quantified using daily MODIS satellite retrievals, and is related to range of CCFs derived from the ERA5 reanalysis product. We find that sea-surface temperature is the CCF that individually explains the largest fraction of the asymmetry, followed by surface wind. The asymmetry is predominantly due to low clouds, and our results are consistent with prior local-scale modelling work showing that marine boundary-layer clouds become thicker and more extensive as surface wind increases and surface temperature cools. The asymmetry is consistent with large-scale control of storm track intensity and surface winds by meridional temperature gradients: persistently cold and windy conditions in the Southern Hemisphere keep cloud albedo high year-round. Our results have important implications for global-scale cloud feedbacks and contribute to efforts to develop a theory for planetary albedo and its symmetry.

Restricted access