AS1.14

Radar observations play a crucial role in detecting and measuring precipitation (QPE, nowcasting) and are useful in various ways to improve numerical weather prediction (NWP) models. Accurate quantitative precipitation estimation remains a challenge as relationships between radar reflectivity and precipitation rate are inheritly ambiguous.Polarimetric observations have the potential to constrain hydrometeor microphysics (size, shape, orientation, etc.) better than conventional ones. Beside improving precipitation measurements, polarimetric observations can be used to evaluate, validate, and improve the representation of hydrometeors in NWP models.Calculating radar observables from prognostic NWP state variables, forward operators (FOs) are a crucial link in comparing radar measurements to NWP output. This requires that the FOs can accurately simulate corresponding observations and that they are consistent with the model(s), e.g. regarding hydrometeor microphysics. However, a wide range of parameters that affect FO output, are not constrained well by the NWP models. This includes, e.g. the melting state, the shape and microstructure, and the orientation of the hydrometeors. Characterization of the uncertainties of an FO, hence, is fundamental to allow its optimal exploitation. Here, we present the revised and polarimetry-extended version of EMVORADO (Efficient Modular VOlume RADar forward Operator) that is coupled to the ICON and COSMO NWP models and applied by DWD in operational weather forecast/data assimilation. Recent developments have focused on enabling polarimetric simulations with computational speed comparable to the Mie-based simulations so far applied in the operational data assimilation as well as to transferability of the code and intermediate calculation results (lookup tables, namely) between different computer architectures. The ability of the FO to reproduce observed polarimetric signatures is evaluated. Uncertainties resulting from weakly or unconstrained assumptions as well as effects of certain techniques and approximations to enhance efficiency are discussed, regarding their impact on analysis of observations and evaluation of NWP models.

How to cite: Mendrok, J., Carlin, J., Snyder, J., Trömel, S., Shrestha, P., and Blahak, U.: An efficient polarimetric radar forward operator for NWP model validation and data assimilation, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-11786, https://doi.org/10.5194/egusphere-egu22-11786, 2022.

Radar polarimetry and novel developed microphysical retrievals offer great potential for evaluating and improving the parameterization of numerical models. In this study, we intend to inform the subgrid parameterizations of the ICON general circulation model (ICON-GCM), more specifically, to assess and improve the spatial heterogeneity of ice water content at ICON subgrid scales. QVPs (Quasi-Vertical Profiles), generated by azimuthal averaging of various polarimetric radar variables from PPIs (Plan Position Indicators) acquired during standard conical scans at antenna elevation angles of 18°, successfully reduce statistical errors or uncertainties, especially in phase-based measurements such as K_DP (specific differential phase). The QVP method, with a horizontal resolution of about 50 km and a vertical resolution of 30-300 m, provides the ideal data basis for various robust polarimetric microphysical retrievals of ice water content (IWC), total number concentration (Nt), and mean volume diameter (Dm). Moreover, an attempt is made to improve the specific threshold used in ICON-GCM for the onset of aggregation (particle diameter < 0.1 mm for ice and > 0.1 mm for snow) by using estimated particle size distributions (PSD) assuming an exponential function. Although BoXPol (the polarimetric X-band radar in Bonn, Germany) is not sensitive to the smallest ice particles, this indirect method opens the possibility to determine and analyze the variabilities of ice and snow separately and finally evaluate and improve the parameterizations of ICON-GCM. However, the use of QVPs reduces information on sub-grid scale variability compared to the higher-resolved PPIs. A key question is therefore how much averaging is required for robust estimates of IWC, Nt, and Dm, and how we can separate spatial variability from noise. Statistical errors and spatial variabilities/azimuthal standard deviations of different IWC retrievals are analyzed using measurements from BoXPol. It is assumed that the real spatio-temporal variability equals the difference between the azimuthal standard deviation and the standard error of the mean computed over different ranges/heights. The standard error of the mean is calculated by Gaussian error propagation using Bienaymé's law, and it is investigated to what extent the azimuthal window size of 360° in the QVP methodology can be reduced while still guarantee acceptable statistical errors of the IWC, Dm, and Nt retrievals. Finally, based on a large BoXPol data set, statistics of the real IWC variability as a function of height are presented and compared to simulations of ICON-GCM. A new method is presented in which Shannon's information entropy is used to test the distribution of Zlin (linear reflectivity) for homogeneity within the PPIs.

How to cite: Scharbach, T. and Trömel, S.: Variabilities of ice water content, total number concentration and mean volume diameter for improved parametrizations using polarimetric retrievals, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-7667, https://doi.org/10.5194/egusphere-egu22-7667, 2022.

Polarimetric microphysical retrievals bear great potential for the evaluation of numerical models and data assimilation. However, a solid database is still lacking to evaluate their accuracy. In order to evaluate these retrievals and assess their accuracy, ground based polarimetric radar measurements are collocated spatially and temporally with airborne in-situ microphysical data collected during the OLYMPEX campaign (Olympic Mountain Experiment). Retrievals for ice water content, total number concentration, and mean volume diameter of ice particles are assessed exploiting both X-band Doppler on Wheels (DOW) measurements and an in-situ measurements obtained by the University of Dakota (UND) Citation aircraft. Vertical profiles of the microphysical retrievals are derived from sector-averaged RHI scans. The comparison of the retrievals with in-situ data above the freezing level reveals new insights into the strengths, weaknesses, and accuracies of the different retrievals, as well as the advantages using polarimetric retrievals rather than non-polarimetric ones. Results clearly demonstrate the superiority of the polarimetric retrievals. Furthermore, the recently introduced hybrid ice water content retrieval exploiting reflectivity ZH, differential reflectivity ZDR and specific differential phase KDP outperforms other retrievals based on either (ZH, ZDR) or (ZH, KDP) or non-polarimetric retrievals in terms of correlations with in-situ measurements and the root mean square error. ZH-based retrievals for the mean volume diameter partly exhibit significant deviations from airborne in-situ measurements, while polarimetric retrievals show a good agreement.

How to cite: Blanke, A., Heymsfield, A., Moser, M., and Trömel, S.: Evaluation of state-of-the-art polarimetric ice microphysical retrievals exploiting ground based radar and airborne in-situ measurements, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-7623, https://doi.org/10.5194/egusphere-egu22-7623, 2022.

In recent years the dendritic growth zone (DGZ, between -20 and -10°C) has gained a lot of attention. It plays a significant role in the production of precipitation since the ice particle size and number concentration increase significantly in the DGZ. Previous studies have found layers of intense aggregation in the DGZ. Polarimetric radar measurements have revealed that these layers of enhanced aggregation are often accompanied by layers of enhanced differential radar reflectivity (ZDR) and specific differential phase shift (KDP). These observations suggest the growth and increase of concentration of oblate ice crystals at the same height where aggregation is enhanced. Analysis of radar Doppler spectra and mean Doppler velocity (MDV) have further shown a secondary, slow falling peak accompanied by a slow down in the MDV at the same height as the layers of enhanced aggregation and growth of ice particles.  From previous studies it is unclear and often case study dependent where this increase in number concentration of small ice crystals originates and whether it is connected to the enhanced aggregation in the DGZ.

We present a statistical analysis of DGZ observations collected during a three-month-long winter campaign in Jülich, Germany. For our analysis we use observations from a polarimetric W-band Doppler radar and zenith pointing X-, Ka- and W-Band Doppler radars. This unique setup allows us to simultaneously look at the aggregate size, as well as ice crystal shape and concentration. We can therefore look at the described increase of aggregation and ice crystal size and concentration in more detail and see if these signatures can be found in general in mid-latitude clouds.  

Similar to previous studies, our statistical analysis shows a strong increase of aggregation within the DGZ. This increase in aggregation is correlated to a slow down in MDV just below -15°C. The larger the particles in the DGZ, the larger is also the slow down of the MDV. The strong temperature dependence of the slow down and an analysis of the Doppler spectra allowed us to narrow down the origin to an increase in the concentration of small ice crystals in this region as well as enhanced depositional growth leading to a buoyancy effect. Due to the Doppler capabilities of our polarimetric W-band radar, we can derive the maximum of spectral ZDR (sZDRmax), which is not affected by the low ZDR of aggregates. sZDRmax starts to increase at just above -15°C, showing an increase in size of ice crystals at this height. Interestingly, sZDRmax stays constantly elevated until -4°C. KDP shows that the concentration of ice crystals is continuously increasing in the DGZ. This is in contrast to the KDP layers found in previous studies, where KDP was enhanced only around -15°C. We also find KDP to stay constantly elevated until -4°C. Given strong aggregation in the DGZ as a sink for small ice crystals, a source for this continuous increase in ice crystal concentration has to be found. 

How to cite: von Terzi, L., Kneifel, S., and Dias-Neto, J.: Aggregation in the Dendritic Growth Zone: A statistical analysis combining multi-frequency Doppler  and  polarimetric Doppler cloud radar observations, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-11287, https://doi.org/10.5194/egusphere-egu22-11287, 2022.

The dendritic growth zone (DGZ) plays a significant role in the production of precipitation and life cycle of clouds. Previous studies have shown that the DGZ is the region where the ice particle size first starts to increase through aggregation. This increase in ice particle size immensely influences the precipitation on the ground. In-situ cloud observations as well as polarimetric and Doppler radar observations of the DGZ have also shown an increase of ice particle number above the available ice nucleating particles concentration. It is unclear and often case study specific where this large number of new ice particles originates from and if this increase in ice particle concentration influences or even triggers the strong increase in particle size in the DGZ. 

In our work, we combine multi-frequency Doppler and polarimetric Doppler cloud radar observations with Monte-Carlo Lagrangian particle modeling linked by a polarimetric forward operator to test these hypotheses. While polarimetric radar observations are sensitive to small, asymmetric ice particles, the multi-frequency approach can provide information about aggregation and riming. This observational setup allows us to look at the size and shape of ice particles. However, detailed evolution of the particle’s properties and the interaction between ice microphysical processes, such as the aggregation of ice particles and generation of new particles in the DGZ, are difficult to identify using only remote-sensing observations. 

The Lagrangian super-particle model McSnow allows us to describe the microphysical process on the detailed particle level and with that track their individual history.   The newly implemented habit prediction scheme includes ice shape effects that represent various aspects of the particle properties and growth, such as a shape-dependent depositional growth rate, fall velocity, and density evolution, more realistically. Ice habit, fall velocity, and density are core information for radar forward simulations, facilitating the comparison with polarimetric observations.

This setup enables us to test observation-based hypotheses such as an increase in number concentration of small, asymmetric ice crystals in the DGZ due to secondary ice or seeder-feeder processes. First results show that the temperature at which the particle is first nucleated is crucial for the particle's habit development. To match the polarimetric radar observations around -15°C, the particles need to be nucleated within the plate-like growth regime at temperatures warmer than -20°C. Particles nucleated at colder temperatures and falling into the plate-like growth regime do not reach the expected habit and the needed aspect ratios to explain the polarimetric radar observations. It is therefore likely that the small particles that cause the distinct polarimetric features at -15°C do not stem from seeder-feeder processes but rather are generated close to the -15°C level. One possible generation process is ice fragmentation which has been found in previous studies to be particularly enhanced at this temperature regime.

How to cite: Welß, J.-N., von Terzi, L., Kneifel, S., Seifert, A., and Siewert, C.: Exploring the origin of increasing ice particle number in the dendritic growth zone combining polarimetric radar observations and novel Lagrangian particle modeling, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-5159, https://doi.org/10.5194/egusphere-egu22-5159, 2022.

Polarimetric observation operators generate virtual observations from the models, which enable a direct comparison of observed and simulated radar signatures of microphysical processes. However, differences in polarimetric fingerprints between observations and models may result both from model deficiencies and faulty assumptions in observation operators. Using the Bonn Polarimetric Radar forward Operator (B-PRO), the evaluation of the German weather forecast model COSMO in radar observation space revealed deficiencies in the ice-snow partitioning and spurious graupel production near the melting layer. Follow-up sensitivity experiments with the model and forward operator (FO) guided the improvement of model parameters, namely the critical diameter of particles for ice-to-snow conversion by aggregation (Dice) and the threshold temperature responsible for graupel production by riming (Tgr), pushing the synthetic radar variables closer to the observations. However, the model still exhibited a low bias (lower magnitude than observation) in simulated polarimetric moments at lower levels above the melting layer (  -3 to   -13 ° C), where snow was found to dominate. Sensitivity experiments with the FO also could not explain this bias indicating shortcoming in the FO or missing cloud microphysical processes in the 2-moment cloud microphysical scheme of the model.

How to cite: Shrestha, P., Mendrok, J., Pejcic, V., Trömel, S., Blahak, U., and Carlin, J.: Evaluation of the COSMO model in polarimetric radar space – impact of uncertainties in model microphysics, retrievals and forward operators, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-12147, https://doi.org/10.5194/egusphere-egu22-12147, 2022.

Numerical weather prediction and climate models provide important information on essential atmospheric variables and extreme events. However, due to model uncertainties arising from initial value and model errors, the simulation results do not match in-situ or remotely sensed measured observations to an arbitrary accuracy. Machine Learning (ML) and/or Deep Learning (DL) methods have shown to be successful tools in closing the gap between models and observations due to high generalization skills and better representation of non-linear and complex relationships. This study focused on using UNet encoder-decoder Convolutional Neural Network (CNN) for extracting spatiotemporal features from model simulations and predicting the actual mismatches between the simulations results and a reference data set. Here, the model simulations serving as input to the CNN were obtained from climate simulations over Europe with the Terrestrial Systems Modeling Platform (TSMP-G2A). The reference data set representing observations was obtained from the COSMO-REA6 reanalysis. The proposed mismatch learning framework was applied to precipitation and surface pressure representing more and less chaotic variables, respectively. The study shows that UNet is able to learn the precipitation and surface pressure mismatches with a daily average correlation coefficient of 0.68 between the actual against predicted mismatches. Seasonal and regional intercomparisons of various precipitation types (e.g., stratiform rainfall, convective rainfall, and snowfall) reveal that the UNet faces challenges in learning the convective-type precipitation mismatches, which may be due to higher random uncertainties in model-based data. After training the UNet network, the reference data is no longer needed for generating the mismatch information. Thus, the UNet weights may be used online during the simulation or as a post-processor to correct predicted variables, which is useful in impact studies. In the first validation experiments, the corrected precipitation data show a strong improvement over the original simulated model data in mean error (47 % on average), correlation coefficient (37 % on average), and root mean square error (22 % on average).

How to cite: Patakchi Yousefi, K. and Kollet, S.: Closing the Gap between Models and Observations: Deep Learning from Mismatches, EGU General Assembly 2022, Vienna, Austria, 23–27 May 2022, EGU22-13008, https://doi.org/10.5194/egusphere-egu22-13008, 2022.