ALBERT

Description: Ice clouds play a key role in the atmospheric radiation budget, both by reflection of shortwave radiation and absorption-emission of longwave radiation. Through these radiative interactions, ice clouds can set atmospheric temperature gradients and thereby influence atmospheric circulation regimes. The radiative signature of ice clouds strongly depends on their macro- and microphysical characteristics, as well as their optical properties. While the new generation of storm-resolving models improves the representation of the vertical velocities that drive cloud formation, subgrid-scale differences, for example in ice optics and microphysics, generate large variability in the modeled atmospheric cloud-radiative heating (CRH) rates (Sullivan and Voigt, 2021). We propose both an idealized single-column and more realistic two-dimensional transect approach for investigating CRH, using the new ecRad radiative transfer module (Hogan and Bozzo, 2018). First, in a series of single-column calculations, we evaluate the impact of realistic perturbations in macro- and microphysical properties, such as cloud-top temperature and ice crystal effective radius, on CRH. For this approach, a heating sensitivity matrix visualization is presented as the response for the different levels of macro-micro properties perturbations. Secondly, we study the impact of using three different ice optical schemes (Fu, 1996; Fu et al., 1998; Yi et al., 2013; Baran et al., 2016) on CRH over three latitudinal transects located in the Eastern Pacific, Western Pacific, and passing over the Asian Monsoon Area. For each of these transects, ecRad is driven by realistic atmospheric conditions provided by ERA5.

Description: The increasing resolution of numerical weather prediction models makes 3D radiative effects more and more important. However, 3D radiative transfer solvers are still computationally expensive, largely preventing their use in operational weather forecasting. To address this issue, we present a new, “dynamic” 3D radiative transfer model that is based on the TenStream solver (Jakub and Mayer, 2015) and delivers a significant speed-up utilizing two main concepts. First, radiation in this model is not calculated from scratch every time the scheme is called, but uses a time-stepping scheme to update the radiative field based on the result from the previous radiation time step. Secondly, the model is based on incomplete solves, performing just the first few steps towards convergence every time it is called. Applied, these two concepts alone allow to produce radiative flux and heating rate fields close to the original TenStream results at dramatically increased speed. In addition, we use an optimized wavelength sampling that allows to noticeably reduce the number of spectral intervals to calculate integrated shortwave and longwave heating rates without a significant loss in precision. Together, these approaches allow to accelerate 3D radiative transfer towards the speed of currently employed 1D solvers. To demonstrate this, we apply the new solver to a precomputed shallow cumulus cloud time series and compare it to both a traditional 1D delta-Eddington solver, the original TenStream solver, as well as to a benchmark solution provided by the 3D Monte Carlo solver MYSTIC (Mayer, 2009).

Description: Interactions between radiation and clouds are a source of significant uncertainty in current numerical weather prediction (NWP) models. With increased resolution previously neglected effects like the horizontal propagation of radiation will become more important. Future operational models will have to incorporate more realistic description of physical processes and remain computationally efficient.Our approach for tackling these problems in the thermal spectral range is to combine a traditional twostream solver with treatment of subgrid-scale cloud overlap (Črnivec and Mayer, 2019) with the Neighbouring Column Approximation (NCA) model (Klinger and Mayer, 2019), which parametrizes horizontal photon transport between adjacent grid-cells. In addition to a generalized vertical cloud overlap the model introduces horizontal overlap between neighbouring clouds. Thereby the hybrid model includes for the first time both subgrid-scale and grid-scale 3D radiative effects at a reasonable additional computational cost.The performance of the model is evaluated using benchmark Monte-Carlo model MYSTIC (Mayer, 2009) calculations of realistic cloud scenes derived from LES simulations. In addition we assess the benefits of the hybrid model in comparison with classical one-dimensional solvers.

Description: Information on future changes in the climate system and their uncertainties is crucial for decision-makers. It is typically provided by combining output from multiple models to account for structural model errors. At the same time there is growing awareness that not all models provide equally plausible and independent simulations. Methods have, therefore, been developed to constrain model distributions based on observations and to account for model dependence, typically based on model output on climatological timescales to minimize the effect of internal variability. However, the recent advent of global km-scale models calls for novel evaluation methods that harness information from short simulations and do not rely on decadal simulations, which are not (yet) available due to computational constraints. Here we show that CMIP6 models can be distinguished from each other even based on daily data and that model interdependencies emerge already at daily timescales. We train a convolutional neural network (CNN) to distinguish 43 CMIP6 models based on daily surface temperature fields in the period 1981-2001. We find that the trained CNN is able to correctly identify up to 80% of test samples from the periods 2005-2014 and 2091-2100. Most misclassifications occur between closely related models. Notably, differences in model resolution and related differences in parametrizations emerge as a clear distinguishing factor. Our research shows that robust conclusions about similarities between models (and potentially models and observations) can already be made based on daily data. This will allow new models or model versions to be evaluated on much shorter time scales than previously possible.

Description: Cloud radiative heating (CRH) can affect the dynamics and predictability of extratropical cyclones. For an idealized cyclone, we have shown that the impact can be understood from the modulation of latent heating by CRH and subsequently through changes in the large-scale flow. However, CRH is uncertain in models, suggesting that the cloud radiative modulation of latent heating may vary with different representations of CRH and may affect model predictions of extratropical cyclones. Therefore, we quantify CRH uncertainties for an idealized cyclone by performing large eddy simulations and offline radiative transfer calculations. Several factors contributing to CRH uncertainty, such as cloud sub-grid variability, 3-D cloud radiative effects, and ice-optical parameterization, are quantified in different regions of the cyclone. Our results indicate that ice-optical parameterization and unresolved horizontal cloud inhomogeneity are the two factors contributing most to the CRH differences. On the other hand, 3-D radiative effects are comparatively smaller, especially for deep stratiform clouds within the cyclone. Our analysis shows that for a cyclone simulated at a horizontal resolution of 2.5 km, the CRH uncertainty due to unresolved horizontal cloud inhomogeneity is in the same range as the CRH uncertainty due to ice-optical parameterizations. This suggests that improving the ice-optical parameterization may be more important and efficient than increasing the horizontal resolution of the model. Future work should address to what extent changes in CRH due to different ice-optical parameterizations alter the magnitude of latent heating within the warm conveyor belt of the cyclone and ultimately the cyclone dynamics.

Description: "All models are wrong. Some are wrong in useful ways.” (adapted by the authors from Box, 1979) In this contribution, we take advantage of an error in the ICON atmosphere model to elucidate how cloud-radiative heating affects the intensity of idealized midlatitude cyclones. We present baroclinic life-cycle simulations with two versions of the global atmosphere model ICON. The simulations use either no radiative heating or the cloud-radiative heating inside the atmosphere, but no clear-sky radiative heating. This allows for a clean investigation of the dynamical impact of cloud-radiative heating. Both ICON versions simulate the same cyclone when run without radiative heating, but disagree when cloud-radiative heating is taken into account. Cloud-radiative heating weakens the cyclone in ICON2.1 but strengthens it in ICON2.6. The two model versions simulate very different low-level clouds, with many more low-level clouds and much more negative cloud-radiative heating in the boundary layer in ICON2.1. Negative cloud-radiative heating from low-level cloud tops weakens the cyclone by increasing static stability, while negative cloud-radiative heating from high-level cloud tops strengthens the cyclone by decreasing stability. Our results indicate that clouds and the vertical distribution of their radiative heating influence the dynamics of midlatitude cyclones. Our results call for future work that investigates the impact of cloud-radiative heating on “real” cyclones in the midlatitude storm track regions of Earth’s atmosphere. Such work should also address the interactions between cloud-radiative heating on the one hand and latent heating and cloud microphysics on the other hand.