the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 29 Nov 2024
A new aggregation and riming discrimination algorithm based on polarimetric weather radars
Abstract. The distinction between riming and aggregation is of high relevance for model microphysics, data assimilation and warnings of potential aircraft hazards due to the link between riming and updrafts and the presence of supercooled liquid water in the atmosphere. Even though the polarimetric fingerprints for aggregation and riming are similar qualitatively, we hypothesize that it is feasible to implement an area-wide discrimination algorithm based on national polarimetric weather radar networks only. Quasi-vertical profiles (QVPs) of reflectivity (ZH), differential reflectivity (ZDR) and estimated depolarization ratio (DR) are utilized to learn about the information content of each individual polarimetric variable and their combinations for riming detection. High-resolution Doppler spectra from the vertical (birdbath) scans of the C-band radar network of the German Meteorological Service serve as input and ground-truth for algorithm development. Mean isolated spectra profiles (MISPs) of the Doppler velocity are used to infer regions with frozen hydrometeors falling faster than 1.5 ms-1 and accordingly associated with significant riming. Several machine learning methods have been tested to detect riming from the corresponding QVPs of polarimetric variables. The best performing algorithm is a fine-tuned gradient boosting model based on decision trees. The precipitation event on 14 July 2021, which led to a catastrophic flooding in the Ahr valley in western Germany, was selected to validate the performance. Considering balanced accuracy, the algorithm is able to correctly predict 74 % of the observed riming features and thus, the feasibility of reliable riming detection with national radar networks has been successfully demonstrated.
- Preprint
(5938 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 10 Jan 2025)
-
RC1: 'Comment on egusphere-2024-3336', Anonymous Referee #1, 16 Dec 2024
reply
The manuscript presents the development of a novel algorithm for detecting riming in snowfall using polarimetric weather radars. It introduces a gradient boosting machine (GBM)-based classification approach that effectively distinguishes between rimed and aggregated snow particles. The algorithm leverages quasi-vertical profiles (QVPs) of radar variables, including reflectivity (ZH), differential reflectivity (ZDR), and depolarization ratio (DR), with DR highlighted as an underutilized yet highly effective predictor of riming.
The robustness of the algorithm has been demonstrated through its application to multiple weather events, including the catastrophic flooding in Germany in July 2021. A comparative analysis of machine learning methods, such as logistic regression and artificial neural networks, identifies GBM as the best-performing model.
However, the methodology has notable limitations. The algorithm was developed using a limited number of events from a single radar station, which may constrain its generalizability to diverse meteorological settings. Its high sensitivity to input variables, particularly the accurate calibration of ZDR and ρhv, also poses challenges for widespread applicability, especially in operational radar networks outside Germany. These limitations should be explicitly acknowledged in the conclusion, rather than suggesting a direct application to radar networks like NEXRAD in the US without further validation and adaptation.
Citation: https://doi.org/10.5194/egusphere-2024-3336-RC1 -
RC2: 'Comment on egusphere-2024-3336', Anonymous Referee #2, 17 Dec 2024
reply
This manuscript investigates several methods to detect significant ice particle riming using scanning polarimetric radar data only. There are some interesting results indicating a potential utility of the approach even though some robust verifications and testing will be needed in future. Revisions are needed before the formal publication of the manuscript.
Comments.
- Regions of significant riming used for training in this study are those with air density corrected values of the mean Doppler velocities greater than 1.5 m/s above the bright band (as observed using the vertical radar beam measurement geometry). These velocities approximate reflectivity-weighted snow/ice particle fall velocities which depend not only on degree of riming but also on the particle sizes. Larger particles (like those with higher Z values) would fall faster than smaller particles having the same riming degree (or rime fraction) but lower reflectivites (see, for example, eq.10.62 in Ryzhkov and Zrnic 2019). This is a reason for a positive correlation between MDV and reflectivity which is often observed. Decoupling the particle size and riming effects on MDV is challenging and a simple MDV threshold might not be universal. In any case, the authors need to address this issue in their discussions.
- The 15-sec temporal averaging of vertical beam data to minimize vertical air motions influence on Doppler velocity measurements may be insufficient (line 145). More quantitative justification of such averaging is needed.
- The authors suggest that riming leads a decrease in Zdr (line 273). It may not be so for initial stages of riming when supercooled water freezes and is deposited in between the crystal arms, so the bulk density increases but the overall shape remains approximately the same.
- As I understand Fig. 4 show theoretical calculations of DR. The authors also mention that due to QVP averaging DR values were > -28 dB (line 348). What minimal values of DR were obtained for instantaneous measurements with the slanted and vertical beam geometries?
- Did you account for the elevation angle changes of polarimetric variables when constructing QVP from slanted beam measurements at different elevation angles? Are the polarimetric variables used in this study recalculated for the horizontal beam pointing?
- There are some features seen in Fig.6 which need explanations. For example, similar columns of rain reflectivities (~25 dBZ) just prior 4:00 and around 4:40 correspond to very different Zdr values (0.8 and 0.3 dB). High DR values in the region of supposedly drizzle (Z there is less than 0 dBZ) at around 8:00 UTC (Height < 0.5 km) also look strange. Are those artifacts of the QVP approach? By the way, adding the temperature profile to Fig. 6 would be useful.
- There are very high MDVs (~ 2.5 m/s) above the bright band after 8:00 UTC (Fig.3). What kind of riming can be expected there? There could be contributions from vertical air motions. Doppler spectra could provide additional information.
- I wonder why you did not consider any cold cases without rain and melting layer. Without intervening rain and ML, you could use microwave radiometer measurements of supercooled LWP to better identify riming conditions.
- Assuming that particles at cloud tops (~ 7 km) are small enough to be tracers, one can conclude from Fig.3 that vertical air motions of an order of 0.5 m/s could be present. Air motions of such magnitude could also be expected at lower heights. This would affect the identification of riming using the MDV threshold.
- Define D in equation (13). I believe that this equation is written assuming the Rayleigh scattering approximation for spheroidal particles with vertical symmetry axes as viewed with the horizontal radar beam. It also assumes that the particle bulk density is proportional to the reciprocal of particle size. Also, I am not sure about the multiplication factor (D) in the middle of the right-hand side of this equation. Is it a typo? Finally, I believe that this equation is written for CDR in liner units not in logarithmic units of dB as stated in line 280.
Citation: https://doi.org/10.5194/egusphere-2024-3336-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 121 | 45 | 8 | 174 | 1 | 2 |
- HTML: 121
- PDF: 45
- XML: 8
- Total: 174
- BibTeX: 1
- EndNote: 2
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1