The relative contributions of soil moisture heterogeneities, a stochastic boundary‐layer perturbation scheme and varied aerosol concentrations representing microphysical uncertainties on the diurnal cycle of convective precipitation and its spatial variability are examined conditional on the prevailing weather regime. To achieve this, separate perturbed‐parameter ensemble simulations are performed with the Consortium for Small‐scale Modeling (COSMO) model at convection‐permitting horizontal grid spacing for 10 days during a high‐impact weather episode in 2016 in Central Europe. We consider hourly precipitation amounts and their spatial distribution, focus on ensemble mean and spread aggregated over strong and weak forcing conditions, and employ spatial evaluation techniques. The convective adjustment time‐scale diagnostic is used to distinguish the different precipitation regimes. While the total amount of daily precipitation is hardly changed by the different perturbation approaches (less than 5%), the spatial variability of precipitation exhibits clear differences. Soil moisture heterogeneity primarily introduces variability during convection initiation causing a steeper increase in normalized rainfall spread prior to the onset of afternoon precipitation. The stochastic boundary‐layer perturbations lead to the largest spatial variability impacting precipitation from initial time onwards with an amplitude comparable to the operational ensemble spread. Similarly, perturbed aerosol concentrations impact spatial precipitation variability from the model start onwards, but to a smaller degree. Soil moisture heterogeneity shows the strongest weather regime dependence, with the greatest impact on convection during weak synoptic forcing. All types of perturbation increase dispersion of precipitation while maintaining the domain‐averaged precipitation rates.

The sign of soil moisture–precipitation coupling and its scale dependence are controversial issues when examining soil–atmosphere interactions. Using the operational convection‐permitting COSMO model with 2.8 km horizontal grid spacing, the interplay of soil‐moisture heterogeneities on different scales and the role of the midtropospheric background wind are addressed by performing a series of real‐case simulations for 11 synoptically weakly and moderately forced summer precipitation events across Central Europe. The cumulative effect of soil‐moisture bias and heterogeneities on different scales is investigated, with consideration of the prevailing synoptic forcing, by explicitly perturbing soil‐moisture conditions. A soil‐moisture bias of ±25% is combined with different soil‐moisture heterogeneity length‐scales ranging from 30 to 140 km introduced by chessboard patterns. For synoptically moderately forced cases, our experiments only show a small impact of soil‐moisture perturbations on convective afternoon precipitation. Averaged over all seven synoptically weakly forced cases, however, we find (a) a positive coupling between the overall soil‐moisture bias and the domain‐averaged precipitation and (b) a negative local soil moisture–precipitation coupling. Increased precipitation over drier soils is related to an interaction between thermally induced vertical circulations and the background wind, causing a persistent updraft region at the downstream flank of the dry patches. Convective precipitation is triggered preferentially near these soil‐moisture gradients. The circulation cells are most dominant for soil‐moisture perturbations at scales between 40 and 80 km. The spatial locking of convection at soil‐moisture boundaries at these scales results in an earlier triggering of convection and an overall reduction of the day‐to‐day variability of area‐averaged precipitation.

Precipitation is affected by soil moisture spatial variability. However, this variability is not well represented in atmospheric models that do not consider soil moisture transport as a three-dimensional process. This study investigates the sensitivity of precipitation to the uncertainty in the representation of terrestrial water flow. The tools used for this investigation are the Weather Research and Forecasting (WRF) Model and its hydrologically enhanced version, WRF-Hydro, applied over central Europe during April–October 2008. The model grid is convection permitting, with a horizontal spacing of 2.8 km. The WRF-Hydro subgrid employs a 280-m resolution to resolve lateral terrestrial water flow. A WRF/WRF-Hydro ensemble is constructed by modifying the parameter controlling the partitioning between surface runoff and infiltration and by varying the planetary boundary layer (PBL) scheme. This ensemble represents terrestrial water flow uncertainty originating from the consideration of resolved lateral flow, terrestrial water flow uncertainty in the vertical direction, and turbulence parameterization uncertainty. The uncertainty of terrestrial water flow noticeably increases the normalized ensemble spread of daily precipitation where topography is moderate, surface flux spatial variability is high, and the weather regime is dominated by local processes. The adjusted continuous ranked probability score shows that the PBL uncertainty improves the skill of an ensemble subset in reproducing daily precipitation from the E-OBS observational product by 16%–20%. In comparison to WRF, WRF-Hydro improves this skill by 0.4%–0.7%. The reproduction of observed daily discharge with Nash–Sutcliffe model efficiency coefficients generally above 0.3 demonstrates the potential of WRF-Hydro in hydrological science.

To study the combined impact of soil moisture and microphysical perturbations on convective clouds and precipitation over Central Europe, an ensemble of five dozen real-world weather prediction forecasts was conducted with the COnsortium for Small-scale MOdeling (COSMO) model at convection-permitting resolution for a case with weak large-scale forcing (6 June 2016). We find a large sensitivity of precipitation, ranging from +10% to −23% in 12-hr precipitation totals. While the homogeneous soil-moisture bias of ±25% primarily controls the timing of convection initiation and the amount of surface rainfall, the number of cloud condensation nuclei and width of the cloud droplet size distribution mainly control the number, size, and lifetime of convective clouds. In moisture-limited conditions, mainly positive couplings are acting.Drier soils, cleaner air, and a broader cloud droplet size distribution result in less rainfall. Wetter soils and more polluted conditions lead to fewer, but larger, cloud clusters. Since microphysical process rates depend systematically on the sign of the perturbations, but rainfall does not, there are compensating effects at work that buffer microphysical perturbations directly and impact the cloud condensate amount and the rainfall at the ground.

<p>Satellite images in the solar spectrum provide high-resolution cloud and aerosol information and present promising observations for data assimilation and model evaluation. While visible channels contain information on the cloud distribution and cloud optical thickness, near-infrared channels are in addition more sensitive to cloud microphysical properties and can be used to distinguish between water and ice clouds. The assimilation of these channels is therefore expected to improve the vertical cloud structure and correct errors in effective droplet and ice particle radii.</p><p>The direct assimilation of satellite radiance in operational systems has so far been restricted to infrared and microwave observations. This is because sufficiently fast and accurate forward operators for visible and near-infrared radiances were not yet available, which is related to the fact that multiple scattering makes radiative transfer at solar wavelengths complicated and standard radiative transfer solvers computationally expensive. MFASIS, a 1D radiative transfer method based on compressed look-up tables, is sufficiently accurate and orders of magnitude faster, but limited to visible channels and clouds.</p><p>After discussing the limitations in the current version of MFASIS that prevent it from simulating near-infrared channels accurately, we present an alternative approach that increase the accuracy significantly for near-infrared channels. In this novel approach, the look-up tables are replaced by a neural network reducing the computational costs and allowing for additional input parameters. Those parameters enable us to describe the vertical distribution of cloud parameters, in particular the effective radius profiles, more accurately. We will demonstrate that the errors are reduced considerably, compared to the original MFASIS method. The new approach is tested for the SEVIRI 1.6mu channel using model output from IFS and the convective-scale data assimilation system KENDA, which is based on the ICON-D2 model.</p>

<p>Solar satellite channels of instruments onboard geostationary or polar orbiting satellites provide high resolution information on clouds and aerosols that is valuable for numerical weather prediction. The solar channels are sensitive to the microphysical properties of cloud and aerosol particles and contain better information on water content than the thermal channels. The direct assimilation of solar satellite images or their application for the evaluation of numerical weather prediction (NWP) models requires sufficiently fast and accurate forward operators, which solve radiative transfer (RT) problems to compute synthetic images from the NWP model output. As multiple scattering complicates the solution of radiative transfer problems in the solar spectral range, standard RT methods are too slow for this purpose. Faster methods have been developed for cloud-affected visible channels, but are limited to non-absorbing channels and do not take aerosols into account. Machine learning methods provide a promising way to accelerate the complex radiative transfer computations in satellite forward operators and to overcome the limitations of previous approaches. Here we report on experiments based on deep feed forward neural network. It is demonstrated that using neural networks the amount of training data that has to be computed with standard radiative transfer methods can be reduced by several orders of magnitude, compared to previous approaches, while increasing the speed by an order of magnitude and improving accuracy. Moreover, tangent linear and adjoint versions required for variational data assimilation can easily be implemented and do not have to be adapted when network structure or training data are changed. We discuss optimizations to reduce the computational effort and provide examples for applications that require more input parameters than cloud-affected visible channels and have only become feasible with the new approach.</p>

Abstract. This study presents an extension to the method for fast satellite image synthesis (MFASIS) to allow simulating reflectances for the 1.6 μm near-infrared channel based on a computationally efficient neural network with improved accuracy. Such a fast forward operator enables using 1.6 μm channels from different satellite instruments in applications like model evaluation or operational data assimilation. It thus paves the way for the exploitation of additional information at this frequency, e.g. on cloud phase and particle sizes, which is complementary to the visible and thermal infrared range. To achieve similar accuracy for 1.6 μm NIR as for the visible channels 0.4–0.8 μm, it is important to represent vertical gradients of effective cloud particle radii, as well as mixed-phase clouds and molecular absorption. A comprehensive dataset sampled from IFS forecasts is used to develop the method. A new approach for describing the complex vertical cloud structure with a two layer model of water, ice and mixed-phase clouds optimized to obtain small reflectance errors is described and the relative importance of the different input parameters describing the idealized profiles is analyzed. Additionally, a different parameterization of the effective water and ice particle radii was used for testing. Further evaluation uses a month of ICON-D2 hindcasts with effective radii directly determined by the two-moment microphysics scheme of the model. The fast neural network approach itself does not add any significant additional error compared to the profile simplifications. In all cases, the mean absolute reflectance error achieved is about 0.01 or smaller, which is an order of magnitude smaller than typical differences between reflectance observations and corresponding model values.