Target Separation and Classification using Cloud Radar Doppler-Spectra

The interpretation of cloud radar measurements is sometimes difficult as signals can be contaminated by non-hydrometeor targets as insects or dust (often called as atmospheric plankton). Moreover the meteorological signals are caused by many different types of hydrometeors like cloud droplets, rain, snow or hail. The first step in the interpretation cloud radar measurements therefore is the classification of targets which is usually done by a combination with lidar (ceilometer) measurements and temperature information. But these algorithms can only assign one target type per measuring volume (one range-time point). In cloud radar spectra often different target types occur in the same measuring volume. To be able to separate and classify different target types a new retrieval software has been developed. It is tailored for vertically pointing polarimetric cloud radars as it is based on the difference in the fall velocity and the LDR of the different target types. Only temperature profiles either from numerical models or radiosoundings are needed as additional input.

The new software consists of two programs, 'spcs2dmp' and 'mmclx', which are intended for pipeline like operation.

Spcs2dmp splits each co-polarized spectrum to the velocity ranges containing significant peaks and calculates the first three spectral moments of each peak and the LDR (Linear Depolarization Ratio) using the corresponding velocity range in the cross spectrum. These four values per peak are saved in a very compact data format.

Ideally the spectra should be split so that each piece contains the spectral lines caused by one target type. Different approaches have been tested and a feasible solution to this problem has been found.

Mmclx makes the difficult job of assigning the peaks to different target types. Separate profiles for each target type are saved in NetCdf format. If in one range-time point more than one peak of the same target type was found the moment and LDR values of these peaks are combined to a total moment set.

In a first step mmclx separates plankton and hydrometeors. Most of the plankton below the melting layer can be recognized safely as it has high LDRs. The remaining plankton is filtered by a three dimensional (range, time, velocity) cluster analysis.

In the second step the remaining hydrometeor peaks are classified as follows: Below the melting height they are classified as cloud dropelts (including drizzle) or rain depending on their velocity. Above the melting layer they are classified as ice or supercooled water, depending on their LDR and the temperature (experimental).

For the temperature profiles needed for the peak classification rawinsonde data and alternatively numerical model forecasts are used. In some cases the melting height can be deduced from the radar data by an algorithm which detects the bright band (high LDR, velocity gradient,...). This is very useful to improve the precision of the melting height which is critical for the target classification.

The new retrieval software has been implemented and taken to continuous operation since Nov. 2006 at two polarimetric 36 GHz cloud radars, at the Meteorological Observatory Lindenberg, and at the Max-Planck Institut of Hamburg. Before it had been tested by 5 month of saved spectra. On the conference the new retrieval software will be explained in more detail and experiences based on one year of data will be demonstrated. The performance of the peak classification will be evaluated by comparisons with collocated ceilometer measurements. Examples where different target types have been separated at the same measuring volumes will be shown by Sabrina Melchionna (at this conference).