COST 75 STSM Report: Main Text Body

BACKGROUND

In Sweden, anomalous propagation (anaprop) of the radar beam causes false precipitation echoes over both land and sea. The characteristics of anaprop over land are well known and can often be treated successfully using intelligent Doppler processing. Sea clutter, however, is much more difficult to deal with as it is usually generated by waves on the sea surface with true velocities. Currently, SMHI uses no method for treating sea clutter. Clutter from land is only treated using Doppler information and this is confined to within a range of 120 km using our radars. This Short Term Scientific Mission serves as a means of learning how the issue of anaprop is being dealt with in Finland, where natural conditions, both seasonal and of the Earth's surface, are very similar to those in Sweden.

Radar manufacturers have made major efforts in recent years to increase the interactivity of operational radars so that the meteorologist and/or engineer can easily tune system parameters and SEE the results directly in radar data. FMI is an excellent example of a weather service which has realized the advantages in investing in a modern network consisting of such radars. FMI's network is presented below.

On the personal level, I am neither a radar meteorologist nor an engineer. I have a background in remote sensing which has given me knowledge in data analysis which can be considered complimentary to that common in radar meteorology. My knowledge of image analysis and the integration of data from multiple sources is relatively new to weather radar and is an advantage in work at a national weather service. I lack knowledge and experience in radar system technology and in the ability to work interactively with system configuration for the purposes of identifying and treating anaprop. This makes an STSM at FMI even more beneficial.

OBJECTIVES

The general objective of this STSM is to study FMI's ability to analyze land and sea clutter using modern operational radars. More specifically, the objectives are to:

• configure and evaluate the use of the IRIS RVP-6 clutter filter,
• evaluate the differences between the use of PP and FFT processing,
• see what effect changes to the PRF have
• analyze the SQI, coherency and the clutter to signal ratio (CSR), using different system parameter configurations,
• compare the differences between the Korpo (FMI) and Arlanda (SMHI) radars, given interesting cases with clutter or combined clutter with precipitation.

In conducting these tasks, FMI's newly installed radar at Korpo was targeted for use. This radar had not yet been added to the Finnish network in operational activities, but it had been installed and accepted for use. This gave us the ability to work with it without disturbing FMI's activities. The Korpo radar is highly interesting for this work since it is located in the Finnish archipelago, where clutter from land and sea is frequent and combined. This radar is also interesting from a Swedish point of view since the overlap with the Arlanda radar (outside Stockholm) is relatively large. Arlanda is an elderly radar with Doppler capabilities out to 120 km range. Korpo is a brand new radar with Doppler out to full range, which in FMI's operations is 250 km. Korpo is beyond Arlanda's Doppler range yet comfortably within non-Doppler. We also targeted the radar at Vantaa, outside Helsinki, for use; this system was installed in 1993.

THE WEATHER

In short, the weather during this STSM was ideal for studying anaprop. The situation was dominated by a warm air mass over a cold sea with a south-easterly air flow.

FMI's RADAR NETWORK

FMI runs a network of 7 operational radars, 6 of which are new C-band Gematronik systems and the seventh is an X-band MRL 5 at Rovaniemi. The C-band radars are all Doppler systems; the X-band radar is not.

Connecting these radar nodes to FMI's central node (Helsinki) is a 64 kbit TCP/IP based internal network. Every 5 minutes, polar volumes (4 elevations) of reflectivity and radial wind velocity are sent from each node to Helsinki. Every 15 minutes, 11-elevation polar volumes of the same parameters are sent to Helsinki. All products are generated centrally or locally and thereafter distributed to regional offices and customers. The national network's capacity enables the generation and transmission of an additional two volumes of data: spectral width (W) and total reflectivity (raw data, dBT). All data collected at Helsinki is archived. The horizontal resolution of the data is 500 m, which is averaged from the 125 m of the raw data collected at each node. The average time delay from volume generation to reception at Helsinki is less than one minute. The host platform for each node is a Silicon Graphics workstation. There are still a few VAXes being used in the radar network but these are successively being replaced by SGI UNIX workstations. Average monthly data availability is greater than 98 % for all data. FMI performs all radar maintenance themselves and they have their own parts reserve.

SMHI's NETWORK

SMHI runs a network of 11 Ericsson Doppler radars. These were designed during the 1980s. They are completely operational systems with very limited ability to interactively manipulate system parameters and to customize individual measurements. Thus, there is relatively little experience in Sweden in "fine tuning" of operational weather radar systems. Also, the operational data sent to SMHI´s central node (Norrköping) is 500 m pseudo-CAPPI data in cartesian coordinates. All non-Doppler data sent to Norrköping has a horizontal resolution of 2 km. Doppler data has a resolution of 1 km. Data is transferred every 15 minutes. Three of the Swedish nodes are connected through 9 600 baud modems which can only manage a limited traffic load. One node (Gotland) is connected through a 28 800 baud modem. which is considered just fast enough to enable the transport of polar volumes. The remaining nodes are connected through a 64 kbit network. Data availability is highly variable, ranging from 76 % to 99 % depending on the type of 2-D data and node. Poor data availability is not correlated with the method of transport (ie. modem or network) to Norrköping.

Recently, SMHI has started to transfer polar volumes of non-Doppler reflectivity, Doppler reflectivity and radial wind velocity from all but the three slowest nodes to Norrköping every 15 minutes. This is done for the first time and on a purely provisional basis for research purposes; no operational products are generated using this data. Volumes are usually generated in 10 elevations, save those from Norrköping and Gotland which are generated in 20 elevations in non-Doppler and 15 elevations in Doppler. Data availability using this new transfer strategy is currently substandard and subject to review.

ANAPROP TREATMENT AT FMI

SIGMET RVP-6

A new feature in FMI's version of the Sigmet preprocessor (RVP-6) is the ability to apply FFT processing, which is considered to provide higher accuracy results when compared with the more traditional pulse-pair (PP) technique.

Briefly, pulse-pair processing uses the autocorrelation function between pairs of successive pulses to provide an estimate of Vr. The phase shift between two pulses in a pulse pair is used to compute Vr. The average value of the radial velocity is determined by averaging the measured Vr from a series of pulse pairs.

Equally briefly, FFT processing performs a Fast Fourier Transform on a series of 8-256 pulse measurements from one range bin. This produces a Doppler spectrum where the x-axis is the Doppler frequency or phase shift (easily converted to Vr) and the y-axis is the signal power associated with each Doppler frequency (radial velocity). The peak or first moment corresponds with the mean radial velocity of the observed target(s) and the spectral width characterizes their variation in velocity.

The RVP-6 employs a 4th order Infininte Impulse Response (IIR) digital high pass filter to remove low frequency signals due to ground clutter from the linear channel time series. Seven different high pass filters are provided to take into account the clutter width based on differences in antenna rotation rate. This is the clutter filter. These filter stop-band widths vary from 3-14% of the Nyquist interval and stop band attenuation is at least 40 dB. In practise, the operator can adjust the filter from 0 (no filter) to 7 (strongest filter). In operational measurements FMI uses filters 2,3 or 4 depending on the local topography around each radar station.

SIGMET IRIS software

The IRIS software provides a number of tunable parameters, collectively referred to as "data quality thresholding parameters". These are:

The log receiver signal-to-noise ratio (LOG): given in dB, where bin values below the user-defined system noise threshold are removed. We did not modify the default which is 0.8 dB.

The signal quality index (SQI): given between 0 and 1. SQI is a measure of the coherence or Doppler power of the linear channel. SQI=0 indicates white noise (no coherency) and SQI=1 indicates complete coherency. Bin values less than the user-defined SQI are removed.

The clutter-to-signal ratio (CSR): given in dB. CSR compares the ground clutter power to that of the meteorological signal power in the Doppler channel. The meteorological signal is calculated by subtracting clutter power from the total coherent signal power. Bin values less than the user-defined CSR are removed.

The signal level threshold (SIG): given in dB. SIG is the coherent meteorological signal-to-noise ratio in the linear channel and can be used to threshold width and velocity, although SQI is often used solely for this purpose. Bin values less than the user-defined SIG are removed. We did not modify the default which is 5 dB.

The IRIS software also contains a "speckle" filter which removes individual bins where the previous and the following bins are zero or filtered. This was not used. Typical thresholds for each measured quantity are presented in the table below. Note that any combination of the four thresholds is selectable for each measured quantity separately.

MEASUREMENTS

Our activities involved the use of the IRIS software modules Ascope and Antenna; images were displayed using standard IRIS tools. We collected data by "grabbing" these programmes' windows using the xv package and saving them as gif files. All our results are illustrated using these images.

Vantaa

We started on June 10 by visiting the radar at Vantaa, just outside Helsinki, perched atop a water tower. During a 1.5 hour period during mid-day, the network connection to FMI was deliberately down for maintenance which allowed us to freely experiment and conduct our own measurements. Early the same morning, thunderstorm cells had passed through southern Finland. Conditions following these were calm but air mass showers and thunderstorms built up during the afternoon; these were clearly discernable in data from Vantaa.

We limited our activities to making observations. During the course of the afternoon, we observed the following:

• severe Cb cells with reflectivities above 45 dBz,
• real precipitation cells over Estonia, combined with echoes from anaprop near the coast. This example is taken 20 minutes before the one above, the difference being that the clutter filter is switched off here.
• second trip echoes from severe weather which itself was out of range
• a combined sea-breeze and gust front in an E-W direction, having arisin from the outflow from Cb cells to the north and the sea breeze from the south. The front is denoted by insect echoes. This example is compiled as an animation. Notice how the combined front builds up and becomes more easily discernable over time. The visual appearance of the gust front is sensitive to the selected setting of the display palette. If the only interest is precipitation and the lower threshold is set to e.g. 10 dBZ, the front will disappear. On the other hand, if all clear air echoes starting e.g. from -20 dBZ with wide classes are displayed, the image will be contaminated with insect echoes and it is difficult to detect the front. Thus it is important that the software allows flexible selection of both colour class intervals and starting levels.
• a high frequency of insect echoes in the Helsinki vicinity, which can be seen in most of the other examples.

We conducted an experiment where the CSR was modified, given this situation with ground clutter, Cb cells, and insect echoes. We compared the results with the CSR set at 2, 10, and 20 dB. The lowest setting clearly removes real weather echoes and weak insect echoes, along with the clutter. The 10 dB setting is effective in removing clutter and insects but appears to remove some weather as well. The highest setting was not effective at removing the clutter. The default setting of 18 dB appears well suited to this task. A comparison of radial wind velocities can also be made using the CSR set at 10 and 20 dB.

A second experiment dealt with the tuning of the clutter filter. We took a rapid time series of ppi images and treated them with successive filter settings 0, 1, 3, 5, and 7. This example can also be experienced as an animation. One can clearly see how the anaprop is effectively removed using the filter, AND how more of the real precipitation is removed with increasingly strong filtering. The default setting of 4 at Vantaa can be considered appropriate, at least based on this example.

Korpo

As of the morning of June 11, our activities were based at Korpo. Our observation sector was limited to between 80 and 200 azimuth degrees, due to the proximity of a main international telecommunications link. A filter has been installed in July to minimize the possible radar emissions at 5.9 GHz, 300 MHz above the transmitter frequency, enabling scanning the full 360 degrees. Another symptom of the new radar was that an overlay of the Finnish coastline had not yet been installed during the course of this study.

On June 11, we spent most of our time orienting ourselves with the radar and the IRIS software. Since the radar was not yet incorporated into the operational activities, we were able to operate it as though it were a research radar and concentrate on measuring individual phenomena. We measured thus spectra using the Ascope module and we generated PPI images using the Quicklook Menu. We saw no precipitation echoes during the course of the day and instead concentrated on measuring spectra and generating images of anaprop from Estonian islands and the coast.

On June 12, we observed a clear case of the development of thunderstorm cells inland from and parallel to the southern Finnish coast. This radar observation (the image is from Vantaa radar) was confirmed with more-or-less concurrent NOAA AVHRR imagery acquired at SMHI. Since we also observed anaprop from the same locations as during the previous day, we decided to generate images and spectra of both the clutter and the thunderstorm cells.

The following results are emphasized from both days:

• anaprop and sea clutter observed late at night in total reflectivity and radial wind velocity images. The zero-velocity anaprop can be cancelled.
• reflectivity, radial wind velocity and spectrum width images for the line of Cb cells on the 12th. In places, the cells were observed to be quasi-stationary, but the spectrum width image reveals useful information which can help analyze these convective situations.
• effect of high PRF on clutter filtered image quality. Due to the narrow unambiguous velocity range in data using a low PRF, the spectra from targets are so folded that the signal is very to noisy and anaprop treatment becomes more-or-less random. Increasing the PRF increases the unambiguous velocity range and gives a much less noisy signal. This is exemplified using spectrum width images with a PRF of 250 and clutter filters 0, 1, 3, and 5. Raising the PRF to 570 and keeping filter level 5 eliminates the noise yet retains echoes from insects. The same example of this phenomenon, this time using radial wind velocity images, show the same thing with clutter filters 0, 1, 3, 5, and the higher PRF with clutter filter 5. Note that the clutter filter 0 images have shorter ranges than the other images.
• The effective removal of second-trip anaprop echoes using the combined LOG, CSR and SQI filters. In each Ascope panel, the spectrum is obtained at the vertical line of the log(A/D)-range panel. The first panel shows spectra from second trip anaprop and its treatment using LOG and CSR filters. Adding a high SQI threshold (0.4) clearly removes this second trip echo. Based on this example, where second trip anaprop clearly approximates white noise, FMI has plans to introduce a low SQI threshold (around 0) in its operational base reflectivity products. Of course, this risks removing real precipitation with low signal quality.
• the difference between pulse pair and FFT signal processing on anaprop. In the following series of PP images, the effects of increasingly stronger clutter filters to the signal become more and more apparant. Starting with an unfiltered signal trained on anaprop, we can see the effect of the increasingly stronger clutter filter set at 3 and 7. An interesting effect of the software is that it plots a pseudo-peak around the zero velocity area of the spectrum even though one would think that the signal should be completely removed there. When using FFT processing instead, the effects to the spectra are different; no clutter filter provides the recognizable shape, but clutter filter 3 gives a signal which interpolates through the zero velocity region. Clutter filter 7 accentuates this same effect.
• an animation of FFT processed anaprop signals with clutter filters 0, 1, 3, 5, and 7. Note how a ship passes the radar beam during the measurements and how the intensity of anaprop signal varies strongly as a function of time.
• the difference between pulse pair and FFT signal processing on a slowly moving Cb. This example is particularly interesting since it involves a complex target with a broad spectrum (note PRF of 1200 Hz) which goes through the zero velocity area. As in the previous example: PP with clutter filters 0, 3, and 7, and FFT with clutter filters 0, 3, and 7. Clearly, the FFT processing gives signals which should provide higher quality image data, since it allows spectra of echoes without anaprop to retain their shapes. In other words, those parts of weather echoes containing zero velocity information are not removed using this method.

REGRETS

There were a couple of tasks which we did not complete but which would have been desirable. These were the analysis of spectra from sea clutter (which occurred late at night) and the generation of images using FFT processing at Korpo. We regret we did not have time to conduct these tasks.

The objective to compare Korpo and Arlanda radars is a task which we were not able to conduct due to the limitations to the Korpo radar's scan strategy when this mission took place. This is a future task which we look forward to conducting.

SUMMARY AND CONCLUSIONS

This Short Term Scientific Mission deals with FMI's abilities to treat anomalous propagation. The objectives are to experiment with the radar system configurations, try to determine the optimal one for dealing with anaprop, compare the quality of pulse pair and FFT signal processing methods, compare overlapping Finnish and Swedish data if possible, and to conduct a general orientation in the software used for such activities.

Treatment of anaprop

The IRIS software modules are very easy to use and understand and are powerful tools in dealing with anaprop, both in "research" (ie. making custom measurements) and operational modes, and in determining optimal system configurations.

Regarding the IRIS thresholding filters, the two most interesting of which are the CSR and the SQI, the following can be said:

• CSR: the default setting of 18 dB appears well chosen,
• SQI: the combined use of CSR and a low SQI (around 0) should improve the quality of FMI's precipitation products, by removing second trip echoes, without sacrificing significant amounts of real precipitation with low signal quality,

It was also demonstrated that a high PRF led to higher quality radial wind velocity and spectrum width information. This is a result of the wider unambiguous velocity range in data acquired with the higher PRF, which in turn, leads to less noisy data and enhanced abilities to isolate and remove anaprop. Hence it is recommendable that the PRF used in long range (250 km) measurements should be as high as possible (e.g. 570 Hz) instead of the "traditional" PRFs of 200-250 Hz. In the Nordic climate beam overshooting is in most cases so severe that quantitative measurements outside the range of 250 km are not reasonable. A negative side-effect of the higher PRF will be increase in the occurrence of second trip echoes. As has been shown they can be eliminated by SQI thresholding.

The clutter filter's default configuration (4 for Vantaa, 3 for all other radars) seems well chosen.

The comparison of pulse pair and FFT signal processing methods demonstrated that the FFT method gave higher quality spectra which could lead to higher quality radar-based products if/when implemented operationally at all of FMI's nodes.

A number of interesting phenomena were observed and documented.

Due to processing speed limits in the RVP-6, the implementation of FFT processing, to cover the whole measurement range to 250 km in the present sampling (2048 range bins, 25-32 pulses avaraged), would require updating the present RVP-6 dual board versions to 3-board versions.

What SMHI and Sweden can learn from FMI

FMI has generally more modern and competent radar systems than does SMHI. In particular, their systems have Doppler capability out to full operational range which is certainly a major advantage in treating anaprop. They also have a higher performance network for data transfer and experience in running it which has resulted in a high availability of high quality data. The data sent to Helsinki is base data in polar volume form which means that FMI has better abilities to create higher quality radar-based products. This is apparant in the NORDRAD composite images and in FMI's commercial products.

Since there is no cheap, quick and effective method of extending the Ericsson radars' Doppler range to 240 km, the most efficient means of raising the quality of SMHI's radar-based products is to create an infrastructure for guaranteeing the secure transfer of polar volume data operationally and generating products based on this data.

ACKNOWLEDGEMENTS

I would like to thank FMI for having me for the week. In particular I am grateful to Jarmo Koistinen for his efficient and effective logistical planning, which has enabled such a successful mission. The timing of the visit was perfect, both regarding the status of the new Korpo radar and the weather (anaprop) conditions. I would also like to thank Robin King for a quick yet constructive discussion on radar calibration, Elena Saltikoff for valuable discussions, and Kai Sysmelin for accompanying Jarmo and me at Korpo.

REFERENCES

Koistinen, J., 1994: Signal processing and clutter corrections in the Finnish Doppler radar network. COST75 Working Document, 75/WD/65.

Sigmet IRIS and RVP-6 Documentation.