German Meteorological Service

Abstract A new Earth system reanalysis framework is being introduced consisting of the intermediate complexity coupled CESAM model together with its adjoint. The results presented are a first pilot demonstration of an Earth system reanalysis that shows the potential of such an approach to (a) improving model biases through parameter estimation and (b) accomplishing a fully coupled reanalysis over long time windows. Variational data assimilation applied to nonlinear chaotic atmospheric models is limited by predictability, and therefore restricted to short assimilation windows. Applications with ocean models, however, require much longer assimilation windows to propagate sparse data information through time and space efficiently. To overcome this contradiction we employ the adjoint method together with synchronization to atmospheric data. The first application presented requires nearly complete information on the atmospheric state for synchronization, for which it relies on European Centre for Medium‐Range Weather Forecasts Reanalysis v5 data, and a simple nudging technique to achieve synchronization. We demonstrate that, over 39 years, efficient assimilation of in‐situ and satellite ocean data and atmospheric reanalysis data is possible by adjusting the surface fluxes and internal model parameters. Given the coarse resolution of the model of , overturning and meridional transports of heat and fresh water are less realistic than previous higher resolution ocean syntheses based on the adjoint method. However, unsynchronized model runs with adjusted parameters show an improved climate and circulation consistent with the reanalysis, suggesting low initialization shocks if the reanalysis is used for initializing decadal predictions.

Abstract. Several sun photometer networks worldwide include instruments for aerosol optical depth (AOD) observations, such as Global Atmospheric Atmosphere Watch-Precision Filter Radiometer (GAW-PFR) and Aerosol Robotic Network (AERONET). AERONET provides additional aerosol properties such as the detailed volume size distribution and the single scattering albedo through inversion modelling of sky radiance measurements. However, the data availability for such properties is limited due to the limited number of daily almucantar sky radiance scans and cloudiness. AOD is measured significantly more frequently as there can be one measurement even every minute. Also, the AOD measurements are affected only by clouds being too close or covering the solar disk. The Generalized Retrieval of Atmosphere and Surface Properties (GRASP) is a flexible inversion model to retrieve aerosol properties from various observations. One of its capabilities is the retrieval of the volume concentration, the volume median radius and geometric standard deviation for each aerosol size distribution mode and the separation of AOD to each mode using only spectral AOD as an input parameter (known as the GRASP-AOD application). Such properties are important for various applications, as the size of aerosols affects their interaction with solar radiation, clouds and radiative forcing modelling. Particle size also shows significant differences depending on the aerosol type such as dust or biomass burning. In this study, we selected four common stations of GAW-PFR and AERONET, used GRASP to retrieve the bimodal size distribution parameters from AOD measured by GAW-PFR instruments (PFRs) and validated the results for different conditions using AERONET data as reference. One of those sites includes a multi-year parallel timeseries from two different BTS spectroradiometers that combined can provide direct spectral irradiance (and as a result AOD) in the 300–2150 nm range. Using this dataset, we were able to investigate the effect and potential benefits of the increased spectral range on GRASP-AOD retrievals. This is mostly focused on the retrieval of the coarse mode volume median radius, which is particularly challenging with the filter radiometers measuring up to 862 or 1020 nm. We also assessed the performance for certain dust and biomass burning cases. Our results showed good agreement between PFR AOD-based and AERONET sky radiance inversions for AOD modal separation and volume concentrations. Significant improvement of the PFR-AERONET intercomparison was also possible for the fine mode volume and effective radius when restricting the datasets to AOD at 500 nm > 0.1 and Ångström Exponent (AE) > 1. Also, the results showed consistency with previous study regarding the validation of such retrievals using AERONET AOD. Focusing on conditions with high proportion of dust particles, we found consistent results with the general cases Using AOD with a larger spectral range (from BTS spectroradiometer), we found that the wavelength selection may affect the results and that using longer wavelengths can increase the sensitivity of coarse mode volume median radius to AOD and improve the correlation of the GRASP BTS AOD-based and AERONET datasets. However, the available data were limited, so it is not clear under what conditions the inclusion of such wavelengths will result in more accurate retrievals or to what extent. Finally, we were able to reproduce with GRASP the aerosol size characteristics of unusual biomass burning cases from the Canadian wildfires during 2023, but the results showed systematically increased fine mode radius and concentration compared to the AERONET output.

<strong class="journal-contentHeaderColor">Abstract.</strong> Accurately forecasting summertime convection remains a challenge for convection-permitting ensemble prediction systems, which often show insufficient spread in precipitation forecasts. This study examines the role of microphysical uncertainties using the ICOsahedral Non-hydrostatic (ICON) model for four representative convective cases over central Europe. A 108-member cloud microphysics ensemble (MPHYS-ENS) was generated by perturbing cloud condensation nuclei (CCN) and ice-nucleating particle (INP) concentrations, graupel sedimentation velocity, and the cloud droplet size distribution. Microphysical perturbations alone produced substantial variability in convective intensity and location, despite identical initial and boundary conditions. Precipitation totals were highly sensitive to CCN and graupel sedimentation, with deviations of 17&ndash;33 % across cases, while the timing of convection onset was only weakly affected. Rapid domain-wide error growth indicated strong thermodynamic impacts even in cloud-free regions. Process diagnostics showed that water&ndash;ice and vapor&ndash;liquid phase changes dominate mean hydrometeor mass rates, while the most frequent processes involved evaporation. Cold-rain pathways consistently governed precipitation; higher CCN and INP concentrations enhanced this dominance, whereas faster graupel sedimentation weakened it. The ratio of cold- to warm-rain processes emerged as a potential diagnostic for identifying regimes in which increased aerosol loading enhances, rather than suppresses, precipitation. Comparison with operational ensembles highlighted the importance of ensemble size. The 108-member MPHYS-ENS generated the largest spread, while bootstrapped 20-member subsets approached operational ensemble system levels. This study demonstrates that cloud microphysics is a major source of forecast uncertainty in summertime convection and should be explicitly represented in ensemble design.

Abstract. A reliable quantification of greenhouse gas emissions is important for climate change mitigation strategies. Inverse methods based on observations and atmospheric transport simulations can support emission quantification at the national scale, yet, they are often limited by the observing systems, transport model uncertainties, and inversion methodologies. This two-part study introduces a system for observation-based, regional methane flux estimation. In the present Part 2, we apply this system to estimate German methane emissions in 2021. The numerical weather prediction model ICON with its ART module for trace gases is used to simulate the atmospheric transport while estimating uncertainties using a transport ensemble. We use a priori fluxes from national reporting to facilitate the validation of reported fluxes. Posterior fluxes are estimated with a modified synthesis inversion method introduced in Part 1, relying on in-situ observations. Compared to the a priori, we find a significant increase in methane emissions in Germany and in the Benelux. We estimate German methane emissions (32±19) % higher than the anthropogenic emissions in the national inventory, and our inversion method attributes this difference mainly to the agricultural sector, although separation from Land Use, Land Use Change and Forestry (LULUCF) as well as natural fluxes requires further research. The combination of an ensemble-enhanced numerical weather prediction model for atmospheric transport and good observation coverage paves the way to sector-specific, observation-based national emission estimates.

Abstract. This paper addresses a major challenge in assimilating 3D radar reflectivity data with a localized ensemble transform Kalman filter (LETKF). In the case of observations with significant reflectivity and small or zero corresponding simulated reflectivities for all ensemble members, i.e., when the ensemble spread is vanishing, the filter ignores the observations based on its low-variance estimate for the background uncertainty. For such low-variance cases, the LETKF is insensitive to observations and their contribution to the analysis increment is effectively zero. Targeted covariance inflation (TCI) has been suggested to deal with the ensemble spread deficiency (Yokota et al., 2018; Dowell and Wicker, 2009; Vobig et al., 2021). To actually make TCI work in a fully cycled convective-scale data assimilation framework, here we will introduce a process-oriented approach to the TCI in combination with a conditional approach formulating criteria under which targeted covariance inflation is efficient. The process-oriented conditional TCI addresses the challenge of underrepresented reflectivity in the prior by constructing artificially simulated reflectivities for each ensemble member based on current observations and typical convective processes. Furthermore, certain conditions are used to restrict this spread inflation process to a carefully selected minimal set of eligible observations, reducing the noise introduced into the system. We will describe the theoretical basis of the new TCI approach. Furthermore, we will present numerical results of a case study in an operational framework, for which the TCI is applied to radar observations at each hourly assimilation step throughout a data assimilation cycle. We are able to demonstrate that the TCI is able to clearly improve the assimilation of radar reflectivities, making the system dynamically generate reflectivity that would otherwise be missing. Related to this, we are able to show that the fractional skill score of radar reflectivity forecasts over lead times of up to 6 h is significantly improved by up to 10 %. All of the results are based on the German radar network and the ICON-D2 model covering central Europe.

Abstract. Atmospheric measurements of the short-lived radioactive noble gas radon-222 (222Rn) have many applications. Its concentrations are driven by atmospheric mixing as well as seasonal variations, which follow the seasonality of 222Rn exhalation from continental soils, with lower values during the wet winter and spring months compared to more dry summer conditions. Hence, it can be used as a tracer to distinguish marine from continental air masses or for transport model validation. The Heidelberg Radon Monitor (HRM) is a static filter detector measuring atmospheric 214polonium (214Po), which is a progeny of 222Rn. These measurements can be used to infer atmospheric 222Rn activity concentrations if the radioactive disequilibrium between 214Po and 222Rn at the measurement site is known. In this study, 214Po activity concentrations measured with the HRM at eight stations in the Integrated Carbon Observation System (ICOS) Germany network are presented, along with guidelines for evaluating these data to estimate atmospheric 222Rn activity concentrations. In addition to the established line loss and disequilibrium corrections applied when sampling through long tubing or from air intake heights close to the ground, respectively, an upper limit for relative humidity (RH) is suggested, where secular equilibrium can still be assumed. At higher RH, aerosol scavenging effects can cause disequilibrium between 214Po and 222Rn. Using comparison with the model this threshold is determined to be at about 98 % RH and was applied uniformly at all measurement sites. A clear diurnal cycle of 222Rn is observed at all German tower stations during the summer and autumn months as is a seasonal cycle with a maximum during summer and autumn months. Overall, our results demonstrate that the 214Po-based 222Rn measurements with the HRM are reliable if the equilibrium conditions between 214Po and 222Rn can be ensured, i.e., for air intake heights above 80–90 ma.g.l. during conditions with RH&lt;98 %. The corrected and ready-to-use dataset of 10-year radon activity concentrations from the eight ICOS stations is published alongside this paper (https://doi.org/10.18160/Q2M8-B1HJ, Fischer et al., 2024).

<strong class="journal-contentHeaderColor">Abstract.</strong> A reliable quantification of greenhouse gas emissions is important for climate change mitigation strategies. Inverse methods based on observations and atmospheric transport simulations can support emission quantification down to the national scale, yet, they are often limited by the observing systems, transport model uncertainties, and inversion methodologies. Here, we present a system for observation-based, regional methane flux estimation, which has the potential for long-term operational support of national emission reporting. We apply this to Central Europe in 2021 with focus on Germany, where we distinguish emissions from different anthropogenic sectors. The atmospheric transport is calculated with the numerical weather prediction model ICON&ndash;ART at 6.5 km resolution, sampling the meteorological uncertainty with a 12-member transport ensemble. We use a priori fluxes from national reporting to facilitate the validation of reported fluxes. Posterior fluxes are estimated with a modified synthesis inversion method, relying on observations from the Integrated Carbon Observation System (ICOS). Compared to the a priori, we find a significant increase in methane emissions in Germany and in the Benelux. We estimate German methane emissions (32 &plusmn; 19) % higher than the anthropogenic emissions in the national inventory, and attribute this difference mainly to the agricultural sector, although separation from Land Use, Land Use Change and Forestry (LULUCF) as well as natural fluxes requires further research. The combination of an ensemble-enhanced numerical weather prediction model for atmospheric transport and good observation coverage paves the way to sector-specific, observation-based national emission estimates.

Abstract This study leverages the Field Experiment on Sub‐mesoscale Spatio‐Temporal Variability in Lindenberg (FESSTVaL), including comprehensive observations of the surface, atmospheric boundary layer (ABL), and clouds, to compare the performance of a numerical weather prediction (NWP) model at sub‐km resolutions with traditional large‐eddy simulations (LES). This comparison is both relevant and timely: as the typical grid spacings of both techniques are converging, so should their results if the LES is assumed to be a good virtual laboratory for the ABL and if NWP is resolving the turbulence well. The representative of NWP is the Icosahedral Nonhydrostatic (ICON) model run at horizontal resolutions ranging from 2.5 km to 78 m. The LES model MicroHH is run at resolutions from 75 m to 38 m. ICON is set up in a limited area with realistic boundary conditions and heterogeneous land surface, whereas the setup of MicroHH is doubly periodic above a homogeneous surface and flat terrain. We focus our comparison at a horizontal grid resolution of about 78 m, where the two models overlap. ICON can represent the ABL processes with high fidelity. It approaches the performance of the MicroHH‐LES in representing surface turbulent and radiation fluxes due to better‐resolved ABL dynamics, clouds, and land‐surface properties at sub‐km resolutions. The modeled turbulence shows good agreement with observations, highlighting the equal importance of resolved and subgrid turbulent mixing at sub‐km resolutions. Our modeling setup also reproduces deep convective cold pools, although with only a qualitatively realistic onset and development. In addition to the FESSTVaL observations used here, extensive datasets are available for follow‐up studies focusing on specific processes. The complete set of ICON simulations covers the intensive observation period between June 5 and July 5, 2021, and can be used in process studies as well as providing a benchmark for model development.

<strong class="journal-contentHeaderColor">Abstract.</strong> We investigate how small-scale orography influences the evolution and structure of the exceptionally deep convective boundary layer (CBL) over the Tibetan Plateau (TiP). Using large-eddy simulations (LES) at 50 m resolution under semi-idealized dry conditions, we compare three experiments over an elevated plateau (4.2 km above sea level, ASL): FLAT (no orography), REAL (realistic terrain), and REALu10 (REAL plus an upper-level wind of 10 m s⁻&sup1;). All simulations produce very deep CBLs, reaching ~9 km ASL by late afternoon, consistent with record-high values observed over the TiP. Small-scale orography substantially accelerates early growth: by midday the CBL in REAL is ~80 m higher than in FLAT, and locally above the mountain it is ~500 m deeper. This terrain-induced advantage narrows later in the day as all cases attain extreme depths. In REALu10, the CBL grows even higher, reaching 9.4 km ASL by 19 LT. Added shear organizes convection into longitudinal roll vortices, contrasting with the isolated terrain-anchored plumes in the no-shear case, and enhances vertical mixing. These results demonstrate that unresolved small-scale orography can increase daytime CBL height by up to 15 % and expedite entrainment of free-tropospheric air. Under clear-sky conditions, the plateau&rsquo;s CBL can exceed 9 km within a single day given strong surface heating and weak stability aloft. Our findings highlight the importance of including fine-scale terrain and shear effects in models, as their omission may underestimate CBL growth and vertical exchange over high-altitude regions.

Abstract The focus of nowcasting development is transitioning from physically motivated advection methods to purely data-driven machine learning (ML) approaches. Nevertheless, recent work indicates that incorporating advection into the ML value chain has improved skill for radar-based precipitation nowcasts. However, the generality of this approach and the underlying causes remain unexplored. This study investigates the generality by probing the approach on satellite-based thunderstorm nowcasts for the first time. Resorting to a scale argument, we then put forth an explanation when and why skill improvements can be expected. In essence, advection guarantees that thunderstorm patterns relevant for nowcasting are contained in the receptive field at long forecast times. To test our hypotheses, we train residual U-Net (ResU-Net) solving segmentation tasks with lightning observations as ground truth. The input of the baseline neural network (BNN) is short time series of multispectral satellite imagery and lightning observations, whereas the advection-informed neural network (AINN) additionally receives the Lagrangian persistence nowcast of all input channels at the desired forecast time. Overall, we find only a minor skill improvement of the AINN over the BNN when considering fully averaged scores. However, assessing skill conditioned on forecast time and advection speed, we demonstrate that our scale argument correctly predicts the onset of skill improvement of the AINN over the BNN after 2-h forecast time. We confirm that, generally, advection becomes gradually more important with longer forecast times and higher advection speeds. Our work accentuates the importance of considering and incorporating the underlying physical scales when designing ML-based forecasting models. Significance Statement This research tests the effectiveness of combining physically motivated algorithms with artificial intelligence for short-term thunderstorm forecasts. Our neural network uses geostationary satellite imagery and lightning observations together with their advection forecasts as inputs for predicting the probability of lightning occurrence. Additionally, we propose a simple scale-based explanation for when and why incorporating physically motivated algorithms benefits convolutional neural networks, and we test this hypothesis. We find that our scale argument can roughly predict the onset of usefulness. The application of the scale argument is quite general and can aid in the development of improved forecasting models in multiple domains.