Many global climate models underestimate the cloud cover and overestimate the cloud albedo, especially for low-level clouds. We determine how a correct representation of the vertical structure of clouds can fix part of this bias. We use the 1D McICA framework and focus on low-level clouds. Using LES results as reference, we propose a method based on exponential-random overlap (ERO) that represents the cloud overlap between layers and the subgrid cloud properties over several vertical scales, with a single value of the overlap parameter. Starting from a coarse vertical grid, representative of atmospheric models, this algorithm is used to generate the vertical profile of the cloud fraction with a finer vertical resolution, or to generate it on the coarse grid but with subgrid heterogeneity and cloud overlap that ensures a correct cloud cover. Doing so we find decorrelation lengths are dependent on the vertical resolution, except if the vertical subgrid heterogeneity and interlayer overlap are taken into account coherently. We confirm that the frequently used maximum-random overlap leads to a significant error by underestimating the low-level cloud cover with a relative error of about 50%, that can lead to an error of SW cloud albedo as big as 70%. Not taking into account the subgrid vertical heterogeneity of clouds can cause an additional relative error of 20% in brightness, assuming the cloud cover is correct.

Numerical experiments have revealed the importance of horizontal transport of light in the presence of clouds (“3D effects’), with consequences for climate, weather and solar resource availability predictions. Yet, atmospheric models continue to use 1D radiative transfer schemes. In this paper, we provide observational evidence for 3D effects. Observed direct and diffuse surface fluxes are compared to outputs from the ecRad radiative transfer scheme run on retrieved cloud profiles. The direct-beam calculation takes careful account of the field-of-view of the pyrheliometer to ensure consistency between observed and modeled direct fluxes. Only the solver that accounts for 3D effects is able to reproduce the observed mean direct–diffuse partition of surface fluxes as a function of solar zenith angle and cloud cover. This validates the physically-based model for 3D radiative processes included in ecRad. These results are in favor of a shift of paradigm from 1D to 3D radiation in large-scale models.

We demonstrate a new approach for climate model tuning in a realistic situation. Our approach, described in detail in Part I, systematically uses a single-column configuration of a global atmospheric model on a series of test cases for which reference large-eddy-simulations are available. The space of free parameters is sampled running the single-column model from which metrics are estimated in the full parameter space using emulators. The parameter space is then reduced by retaining only the values that are consistent with the metrics computed on large eddy simulations within a given tolerance to error. The approach is applied to the recently designed 6A version of the LMDZ model, itself the result of a long investment in the development of physics parameterizations and by-hand tuning. The boundary layer is revisited by increasing the vertical resolution and varying parameters that were kept fixed so far. The approach allows us to automatically reach a tuning as good as that of the 6A version, after some improvements are done at process scale. This approach helps accelerate the introduction of new parameterizations, by avoiding a tedious manual tuning process and preventing some of the error compensations that could occur if calibration was carried out directly with the full atmospheric model. This way of using machine learning techniques allows us to maintain the physical foundations of the model and to ensure that the improvement of global metrics is obtained for a reasonable behavior at process level. That is, we get things right for the right reasons.

Process-scale development, evaluation and calibration of physically-based parameterizations are key to improve weather and climate models. Cloud–radiation interactions are a central issue because of their major role in global energy balance and climate sensitivity. In a series of papers, we propose papers a strategy for process-based calibration of climate models that uses machine learning techniques. It relies on systematic comparisons of single-column versions of climate models with explicit simulations of boundary-layer clouds (LES). Parts I and II apply this framework to the calibration of boundary layer parameters targeting first boundary layer characteristics and then global radiation balance at the top of the atmosphere. This third part focuses on the calibration of cloud geometry parameters that appear in the parameterization of radiation. The solar component of a radiative transfer scheme (ecRad) is run in offline single-column mode on input cloud profiles synthesized from an ensemble of LES outputs. A recent version of ecRad that includes explicit representation of the effects of cloud geometry and horizontal transport is evaluated and calibrated by comparing radiative metrics to reference values provided by Monte Carlo 3D radiative transfer computations. Errors on TOA, surface and absorbed fluxes estimated by ecRad are computed for an ensemble of cumulus fields. The average root-mean-square error can be less than 5 Wm$^{-2}$ provided that 3D effects are represented and that cloud geometry parameters are well calibrated. A key result is that configurations using calibrated parameters yield better predictions than those using parameter values diagnosed in the LES fields.

The development of parameterizations is a major task in the development of weather and climate models. Model improvement has been slow in the past decades, due to the difficulty of encompassing key physical processes into parameterizations, but also of calibrating or â\euro˜tuningâ\euro™ the many free parameters involved in their formulation. Machine learning techniques have been recently used for speeding up the development process. While some studies propose to replace parameterizations by data-driven neural networks, we rather advocate that keeping physical parameterizations is key for the reliability of climate projections. In this paper we propose to harness machine learning to improve physical parameterizations. In particular we use Gaussian process-based methods from uncertainty quantification to calibrate the model free parameters at a process level. To achieve this, we focus on the comparison of single-column simulations and reference large-eddy simulations over multiple boundary-layer cases. Our method returns all values of the free parameters consistent with the references and any structural uncertainties, allowing a reduced domain of acceptable values to be considered when tuning the 3D global model. This tool allows to disentangle deficiencies due to poor parameter calibration from intrinsic limits rooted in the parameterization formulations. This paper describes the tool and the philosophy of tuning in single-column mode. Part 2 shows how the results from our process-based tuning can help in the 3D global model tuning.