GPUs-RRTMG_LW: high-efficient and scalable computing for a longwave radiative transfer model on multiple GPUs

Optimizing makespan and resource utilization for multi-DNN training in GPU cluster

The paper analyzes the energetics involved in the two numerical experiments described in Part I, concerning a longwave radiative transfer model and a simple Newtonian cooling (or heating) model. A three-dimensional quasi-geostrophic model is developed and used over the entire annual cycle to test the climatic responses of the stratospheric circulation to the cited two different longwave radiative heating models. It is shown that the eddy energy parameters (eddy kinetic energy, eddy available potential energy, generation of the latter, and vertical propagation of eddy geopotential energy) undergo annual variations with maximum values in winter and minimum values in summer. The zonal energy parameters exhibit semiannual variations with major maximum values in winter, minor maxima in summer and minima in spring and fall. Analysis of the energy cycle reveals that the vertical propagation of eddy geopotential energy and the generation of eddy available potential energy are the energy sources for energetics in the upper stratosphere. Ways in which the longwave radiative processes may affect the stratospheric energetics are identified.

Altitude and latitude distributions of the populations of the bending and symmetric stretching levels of CO2 and of the cooling rates induced by the bands originating from them in the middle atmosphere (40‐120km), have been calculated for the CIRA 1986 reference atmosphere using a non‐LTE radiative transfer model. A sensitivity study of the more uncertain parameters that strongly affect these quantities has been undertaken. These factors are the carbon dioxide and atomic oxygen volume mixing ratios, and the rate coefficient for the deactivation of the bending mode of CO2 by atomic oxygen. the model makes use of the most recent values of these parameters: the abundances of CO2 in the upper mesosphere and lower thermosphere as measured by, and derived from, rocket and satellite instruments; the global distribution of atomic oxygen density calculated by Rees and Fuller‐Rowell (1988) and the measurements of Thomas (1990); and the fast rate for the deactivation of CO2(0, v2, 0) by O(3P), recently derived by Sharma and Wintersteiner (1990). Among the most important results are the following: (i) the CO2 15 μm fundamental band is close to LTE (Tv‐ Tk ≤ 6K) up to about 100km altitude; (ii) the first and second hot bands exhibit high vibrational temperatures, especially in the daytime summer mesopause region; (iii) the cooling rates by the CO2 15μm bands in the lower thermosphere can be very large, depending on the CO2 abundance. Finally, our results have been compared with other recent non‐LTE models in order to quantify the effects of different radiative transfer schemes on the calculations of vibrational temperatures and cooling rates.

Efficient breadth first search on multi-GPU systems

Heterogeneous acceleration algorithms for shallow cumulus convection scheme over GPU clusters

In this paper the effects of radiation on a meso-scale precipitating system is investigated during a severe storm in South China on June 8, 1998. This was done by using the Pennsylvania State University (PSU)/National Center for Atmospheric Research (NCAR)/Meso-scale Model 5 Version 3 (MM5_V3) after the introduction of radiative transfer schemes that are able to treat water clouds, ice crystals, snow, and groupel. The results suggest that the rainfall patterns do not differ too much for the various radiation schemes used in the numerical calculations, but rather influence the rainfall intensity in the central areas. The radiative effects on precipitation are more significant during daytime, as compared to nighttime. The diurnal variation of rainfall is enhanced by the radiative processes. Computed precipitation intensities, and radiative cooling/heating rates, are dependent on the specific radiative transfer scheme used. The results suggest that model improvement of daytime cloud radiative processes is crucial for a better representation of such effects on meso-scale precipitating system.

The Regional Climate Model Version 4 (RegCM4) with the conventional plane‐parallel radiative transfer scheme severely overestimates the summer precipitation over the Tibetan Plateau (TP) due to the excessive surface heat source, which results from the poor representation of the sub‐grid terrain‐related radiation processes. To realistically describe the surface sub‐grid radiation process in the RegCM4, a 3‐dimensional sub‐grid terrain solar radiative effect (3DSTSRE) parameterization scheme is implemented into the RegCM4 to improve the original plane‐parallel radiative transfer scheme. Results show that adopting the 3DSTSRE scheme in RegCM4 can significantly reduce the summer (June–August) wet bias over the TP produced by the model with the plane‐parallel radiative transfer scheme. Mechanism analysis indicates that the 3DSTSRE scheme largely improves the description of the TP surface energy balance in the RegCM4 by reducing the positive bias of downward surface solar radiation (DSSR). The reduced DSSR leads to the weakened surface heat source and cooler near‐surface air over the TP. Consequently, the local atmospheric circulation adapts to the temperature field as the low‐level anti‐cyclonic (high‐level cyclonic) anomaly over the TP. The adjustment of the temperature and wind field attenuates the water vapor transport, enhances the low‐level atmospheric stability, inhibits the updraft motion, and eventually reduces the rainfall over TP. Although the 3DSTSRE improves the DSSR simulation only during daytime, the precipitation simulation is also improved at nighttime, which is fundamentally attributed to the maintenance of the cooled atmosphere throughout the daytime and nighttime.

Abstract. This study investigates the sensitivity of modeled surface melt and subsurface heating on the Antarctic ice sheet to a new spectral snow albedo and radiative transfer scheme in the Regional Atmospheric Climate Model (RACMO), version 2.3p3 (Rp3). We tune Rp3 to observations by performing several sensitivity experiments and assess the impact on temperature and melt by incrementally changing one parameter at a time. When fully tuned, Rp3 compares well with in situ and remote sensing observations of surface mass and energy balance, melt, near-surface and (sub)surface temperature, albedo and snow grain specific surface area. Near-surface snow temperature is especially sensitive to the prescribed fresh snow specific surface area and fresh dry snow metamorphism. These processes, together with the refreezing water grain size and subsurface heating, are important for melt around the margins of the Antarctic ice sheet. Moreover, small changes in the albedo and the aforementioned processes can lead to an order of magnitude overestimation of melt, locally leading to runoff and a reduced surface mass balance.

A spectrally resolved discrete‐ordinates radiative transfer model is used to calculate the change in downwelling surface and top‐of‐the‐atmosphere (TOA) outgoing longwave (3.9–500 μm) radiative fluxes induced by tropospheric aerosols of the type observed over the Indian Ocean during the Indian Ocean Experiment (INDOEX). Both external and internal aerosol mixtures were considered. Throughout the longwave, the aerosol volume extinction depends more strongly on relative humidity than in most of the shortwave (0.28–3.9 μm), implying that particle growth factors and realistic relative humidity profiles must be taken into account when modeling the longwave radiative effects of aerosols. A typical boundary layer aerosol loading, with a 500‐nm optical depth of 0.3, will increase the downwelling longwave flux at the surface by 7.7 W m−2 over the clean air case while decreasing the outgoing longwave radiation by 1.3 W m−2. A more vertically extended aerosol loading, exhibiting a high opacity plume between 2 and 3 km above the surface and having a typical 500‐nm optical depth of 0.7, will increase the downwelling longwave flux at the surface by 11.2 W m−2 over the clean air case while decreasing the outgoing longwave radiation by 2.7 W m−2. For a vertically extended aerosol profile, approximately 30% of the TOA radiative forcing comes from sea salt and approximately 60% of the forcing comes from the combination of sea salt and dust. The remaining forcing is from anthropogenic constituents. These results are for the external mixture. For an internal mixture, TOA longwave forcings can be up to a factor of two larger. Therefore, to complete our understanding of this region's longwave aerosol radiative properties, more detailed information is needed about aerosol mixing states. These longwave radiative effects partially offset the large shortwave aerosol radiative forcing and should be included in regional and global climate modeling simulations.

The assimilation of satellite microwave measurements and the retrieval of geophysical parameters require a fast and accurate radiative transfer model. In this study, a scheme is developed to solve the vector radiative transfer equation using a polarimetric two-stream approximation. In the scheme, the integration of the phase matrix over azimuth angle is derived as an analytic form that can also be directly utilized for the general radiative transfer scheme. Each Stokes radiance component is expressed as an analytical function of atmospheric and surface optical parameters. The model is applicable for spherical and randomly oriented nonspherical scatters. The differences of brightness temperatures between the polarimetric two-stream model and the matrix operator method are less than 2 K for various frequencies.

<p indent="0mm">Typhoons, characterized by their high destructive potential, significantly impact coastal residents’ lives and property safety. To optimize numerical models’ typhoon simulation, carefully selecting appropriate physical parameterization schemes is crucial, offering robust support for disaster prevention and reduction efforts. This study focuses on Typhoon Mujigae, conducting a comparative analysis of different physical parameterization schemes (microphysics, cumulus parameterization, shortwave radiation, and longwave radiation) in WRF simulations. The key findings are as follows: cumulus and microphysics parameterization schemes notably influence the simulation of typhoon tracks and intensity, while the impact of longwave and shortwave radiation schemes is relatively minor. Typhoon intensity is more sensitive to the choice of parameterization schemes than track. Together, the Kain-Fritsch cumulus convection scheme, WRF Single Moment 5-class scheme, and Dudhia/RRTM radiation scheme yield the best intensity simulation results. Compared with the Betts-Miller-Janjić and Grell 3D scheme, the use of the Kain-Fritsch scheme results in a clearer, taller eyewall and more symmetric deep convection, enhancing precipitation and latent heat release, and consequently improving the simulated typhoon intensity. More complex microphysics schemes like Purdue Lin, WRF Single Moment 5-class, and WRF Double Moment 6-class perform better in simulations, while simpler schemes like Kessler and WSM3 exhibit significant deviations in typhoon simulations. Particularly, the large amount of supercooled water clouds simulated by the Kessler scheme is a major source of bias. Furthermore, a coupling effect exists between cumulus convection and microphysics parameterization schemes, and only a reasonable combination of both can achieve optimal simulation results.

A radiative transfer model for an idealized and non-scattering atmosphere and its application for ground-based remote sensing

A multi-perspective revisit to the optimization methods of Neural Architecture Search and Hyper-parameter optimization for non-federated and federated learning environments

Based on advanced spectroscopic databases, line-by-line and layer-by-layer radiative transfer codes numerically solve the radiative transfer equation with a very high accuracy. Taking advantage of its pre-calculated optical depth look-up table, the fast and accurate radiative transfer model Automatized Atmospheric Absorption Atlas OPerational (4A/OP) calculates the transmission and radiance spectra for a user defined layered atmospheric model. Here we present a module, called 4A-Flux, developed and implemented into 4A/OP in order to include the calculation of the clear-sky longwave radiative flux profiles and heating rate profiles at a very high spectral resolution. Calculations are performed under the assumption of local thermodynamic equilibrium, plane-parallel atmosphere and specular reflection on the surface. The computation takes advantage of pre-tabulated exponential integral functions that are used instead of a classic angular quadrature. Furthermore, the sublayer variation of the Planck function is implemented to better represent the emission of layers with a high optical depth. Thanks to the implementation of 4A-Flux, 4A/OP model have participated in the Radiative Forcing Model Intercomparison Project (RFMIP-IRF) along with other state-of-the-art radiative transfer models. 4A/OP hemispheric flux profiles are compared to other models over the 1800 representative atmospheric situations of RFMIP, yielding an Outgoing Longwave Radiation (OLR) mean difference between 4A/OP and other models of −0.148 W .m−2 and a mean standard deviation of 0.218 W .m−2, showing a good agreement between 4A/OP and other models. 4A/OP is applied to the Thermodynamic Initial Guess Retrieval (TIGR) atmospheric database to analyze the response of the OLR and vertical heating rate to several perturbations of temperature or gas concentration. This work shows that 4A/OP with 4A-Flux module can successfully be used to simulate accurate flux and heating rate profiles and provide useful sensitivity studies including sensitivities to minor trace gases such as HFC134a, HCFC22 and CFC113. We also highlight the interest for the modeling community to extend intercomparison between models to comparisons between spectroscopic databases and modelling to improve the confidence in model simulations.

The cloud longwave (LW) scattering effect has been ignored in most current climate models. To investigate its climate impact, we apply an eight-stream DIScrete Ordinates Radiative Transfer (DISORT) scheme to include the cloud LW scattering in the General circulation model version of the LongWave Rapid Radiative Transfer Model (RRTMG_LW) and the Community Atmospheric Model Version 5 (CAM5). Results from the standalone RRTMG_LW and from diagnostic runs of CAM5 (no climate feedback) show that the cloud LW scattering reduces the upward flux at the top of the atmosphere and leads to an extra warming effect in the atmosphere. In the interactive runs with climate feedback included in CAM5, the cloud LW scattering effect is amplified by the water vapor-temperature feedback in a warmer atmosphere and has substantial influences on cloud fraction and specific humidity. The thermodynamic feedbacks are more significant in the northern hemisphere and the resulting meridional temperature gradient is different between the two hemispheres, which strengthens the southern branch of Hadley circulation, and modulates the westerly jet near 50° S and the upper part of Walker circulation. Our study concludes that the cloud LW scattering effect could have complex impacts on the global energy budget and shall be properly treated in future climate models.