1. Introduction and Motivation

The particle size distribution (PSD) has a major impact on radar signals and is also a key quantity affecting several ice microphysical processes, such as aggregation and secondary ice processes. For example, in von Terzi et al., 2022 we found in the framework of the PROM-Imprint project that a wider PSD aloft is correlated to larger aggregates observed in the dendritic growth layer. Further, an accurate characterization of the PSD turned out to be crucial for initializing model simulations, for example with the 1D Lagrangian model McSnow which we are using in PROM-Fragile. Anne-Claire Billault-Roux (EPFL) recently developed an innovative ice particle properties retrieval based on dual-frequency Doppler spectra using a machine learning approach (Billault-Roux et al. 2023).

2. Moving from dual-frequency to triple-frequency

During my stay in Lausanne, we extended the dual-frequency retrieval to consider three frequencies, such that it can be applied to the TRIPEx-pol dataset (von Terzi et al 2022). It is expected that three frequencies can constrain the microphysical parameters better than just two frequencies, since the addition of the third frequency increases the sensitivity to smaller ice-particle sizes. Several in-situ campaigns suggest that the PSD of ice particles is better represented by a gamma distribution compared to an exponential distribution. We have therefore adapted the retrieval to consider a gamma distribution to describe the PSD instead of an exponential distribution used in Billault-Roux et al. 2023. In order to connect observations and retrieval, an accurate forward model is needed. We have therefore changed the forward model used in Billault-Roux et al. 2023 (based on SSRGA, Hogan et al. 2016) to McRadar, a forward model using LUTs of a large database of DDA scattering properties of dendrites, needles and snow aggregates that was recently developed in the framework of PROM-Fragile. To train the Neural Network (NN) which is the basis of the retrieval, Billault-Roux et al. 2023 used PSDs measured by a MASC (Garrett et al. 2012) instrument. The MASC is sensitive to particles larger than 100 𝜇m, possibly underestimating the number concentration of small ice particles. To increase the representation of small ice particles, the training of the NN was conducted using McSnow simulations. This further allowed to train the NN with a larger variety of mass-size, area-size relationships and PSD shapes, giving the potential to apply the retrieval to datasets with a large variety of (micro-) physical properties of ice-particles. All retrieved parameters are listed in Table 1.

3. Case study results

A case study comparing Doppler spectra and Ze from the 22nd January 2022 collected during TRIPEx-pol campaign (von Terzi et al. 2022) reveals a generally good agreement between observations and the reconstruction from the retrieved (micro-) physical properties. A closer look identifies areas of small discrepancies: the Doppler spectra shown Fig. 1 show a slight underestimation of the main peak (slightly too small sZe at ranges larger than approx. 3000m). While the increase of spectral width at approx. 3000m is captured by the retrieval, the second mode observed at heights between 2000m and 3000m is not represented correctly. The underlying representation of the PSD as a gamma distribution is not able to produce a second mode. A similarly good agreement can be seen when comparing several hours (shown in Fig. 2) of Ze observations and reconstructions.

4. Future perspectives and goals

While the agreement between observations and the reconstructions is strikingly similar, the retrieval of particle properties from radar Doppler spectra is ill-posed and the retrieved properties might not be correct even though their forward simulations look promising. Therefore, the retrieval needs to be validated with e.g. in-situ measurements. Such observations are not available for the TRIPEx-pol campaign. We are currently looking into data collected during the BAECC campaign (Petäjä et al. 2016), which provides triple-frequency Doppler spectra alongside in-situ observations of ice particles close to the ground. Further, in a next step, the added value a third frequency provides for retrieving particle properties will be tested. Also, the best combinations of two frequencies (e.g. X-,W-band vs. Ka-,W-band vs. X-, Ka-Band) will be tested to provide a recommendation of dual-frequency setups if only two frequencies are feasible. In the far future, the possibility to move from a fixed description of the PSD (i.e. through assuming a gamma distribution) towards a binned PSD might be explored, to improve the retrieval in cases of multi-modal spectra.